Modelización regional de clima presente y futuro en el entorno de la Península Ibérica mediante conjuntos de modelos regionales

Transcripción

1 UNIVERSIDAD DE CASTILLA-LA MANCHA FACULTAD DE CIENCIAS AMBIENTALES Y BIOQUÍMICA ÁREA DE FÍSICA DE LA TIERRA Modelización regional de clima presente y futuro en el entorno de la Península Ibérica mediante conjuntos de modelos regionales TESIS DOCTORAL RAQUEL ROMERA RUIZ 2015

2

3 UNIVERSIDAD DE CASTILLA-LA MANCHA FACULTAD DE CIENCIAS AMBIENTALES Y BIOQUÍMICA ÁREA DE FÍSICA DE LA TIERRA Modelización regional de clima presente y futuro en el entorno de la Península Ibérica mediante conjuntos de modelos regionales Memoria presentada para optar al grado de Doctor por la Universidad de Castilla-La Mancha Raquel Romera Ruiz Toledo, Noviembre de 2015 Vº Bº de los Directores: Drs. Enrique Sánchez Sánchez y Marta Domínguez Alonso La Doctoranda: Raquel Romera Ruiz

4

5 RESUMEN, MOTIVACIÓN Y OBJETIVOS DE LA TESIS DOCTORAL El objetivo general de la presente tesis doctoral es la utilización de conjuntos de Modelos Regionales de Clima (RCMs) para estudiar el clima en la Península Ibérica (PI) y en las zonas cercanas que más influyen, como el Norte de África y el Mar Mediterráneo. Para la aplicación de los RCMs en el estudio del cambio climático regional futuro, el primer e imprescindible paso es la evaluación de las simulaciones para clima presente, comparando los resultados con observaciones. De estos resultados de la evaluación de clima presente se deducirá la utilidad de las simulaciones de clima futuro para estudiar el cambio climático, de enorme relevancia en la actualidad. En este trabajo se muestran sobre todo los resultados de estos trabajos de evaluación de clima presente, aunque también se han hecho análisis sobre las proyecciones regionales de clima futuro. Las principales variables que caracterizan el clima de una región son la precipitación y la temperatura y por eso son los primeros parámetros que se analizan en una simulación climática. Aunque un primer paso será el análisis de los campos medios, son los eventos extremos tales como olas de calor o de frío, heladas, noches tropicales, sequías, precipitaciones intensas, etc, los que tendrán un impacto igual o superior en la población, influyendo, por ejemplo, en la economía y en la salud. En el trabajo para obtener el Diploma de Estudios Avanzados (Romera, 2010) que antecede a esta tesis doctoral se analizó la capacidad del RCM PROMES para simular la precipitación y la temperatura de la PI, tanto de los campos medios como de los eventos extremos. En este caso se ha realizado un análisis de clima presente sobre la PI de los resultados de un conjunto de RCMs, incluyendo el RCM PROMES, comparando con observaciones los resultados de precipitación y temperatura, tanto de campos medios como de eventos extremos, sobre la misma región (PI). Numerosos estudios avalan la idoneidad de los RCMs para estudiar el clima de una región, pero se ha demostrado que el uso de un conjunto de RCMs da una visión más amplia y acertada del mismo (Gao et al., 2006; Jacob et al., 2007; GomezNavarro et al., 2010). El mismo método se ha usado para evaluar la precipitación de clima presente de un conjunto de RCMs sobre el Norte de África, región con un gran estrés hídrico y muy vulnerable al cambio climático (Niang et al., 2014), analizando tanto los cambios medios como los eventos extremos (sequías y lluvias intensas). También se ha realizado un análisis de los ciclones de carácter tropical que se desarrollan en el Mar I

6 Mediterráneo, llamados habitualmente medicanes (unión de Mediterráneo y la palabra inglesa hurricane ), ya que estos fenómenos afectan a toda la costa Este de la PI. Doswell et al (1998) afirma que los eventos de precipitación intensa que se dan en el oeste del Mar Mediterráneo están asociados frecuentemente con ciclones, ya que esta zona tiene una alta densidad de ciclogénesis a escala mundial (Petterssen (1956). En este caso, además de una evaluación de clima presente se analizan los cambios que sufrirán estos fenómenos en el clima futuro según los resultados de un conjuntos de RCMs, tanto en intensidad como en frecuencia como en distribución espacial o temporal. Otro aspecto a tener en cuenta a la hora de trabajar con conjuntos de RCMs es cómo obtener un resultado común promediando de alguna manera los resultados de cada RCM. En esta línea se ha hecho un esfuerzo por proponer un método para ponderar los resultados obtenidos por cada RCM y poder obtener un resultado de conjunto. II

7 Estructura de la Tesis Doctoral Esta tesis se estructura en los siguientes capítulos. En el capítulo 1 se encuentra la introducción general. En el capítulo 2 se detalla la metodología utilizada durante todo el trabajo de investigación. En los capítulos del 3 al 6 se muestran los resultados de la tesis doctoral, que se corresponden con cuatro artículos. El capítulo 3 (Sánchez et al., 2009) explica un método para ponderar los resultados obtenidos con un conjunto de RCMs para todo el continente europeo, dividiendo el dominio total en subregiones, donde la PI se considera una única región. La mayor parte de los estudios previos sobre la PI consideraban toda la PI como una sola región, sin tener en cuenta la heterogeneidad climática y la complejidad de la orografía de la zona (Boberg et al., 2009, 2010; Fischer y Schär, 2010; Kjellström et al., 2010, Kotlarski et al., 2014). En el capítulo 4 (Domínguez et al, 2013) se divide la PI en subregiones buscando una homogeneidad climática para hacer un estudio de los extremos de precipitación y temperatura usando los datos del proyecto ESCENA. Anteriormente ya se había hecho un análisis de los campos medios (JiménezGuerrero et al., 2013) sobre la PI usando los resultados del mismo proyecto. En el capítulo 5 (Romera et al., 2015a) se describe la evaluación de las simulaciones de un conjunto de RCMs sobre el Norte de África, mientras que en el capítulo 6 (Romera et al., 2015b) se recoge el estudio de los ciclones de características tropicales que tienen lugar en el Mar Mediterráneo. En el capítulo 7 se detallan las principales conclusiones derivadas de los estudios incluidos en la presente tesis doctoral. Finalmente se muestran las referencias utilizadas y un anexo con los trabajos realizados durante todo el período de investigación. III

8 Índice RESUMEN, MOTIVACIÓN Y OBJETIVOS DE LA TESIS DOCTORAL...I 1 INTRODUCCIÓN La Península Ibérica y su entorno Cambio climático antropogénico Cambio climático observado en la PI Herramientas para el estudio del clima Modelos de clima (GCM, RCM) METODOLOGÍA Conjuntos de RCMs El Modelo Regional de Clima PROMES Datos observacionales en malla Cálculos y métricas utilizados Selección de subregiones y bases observacionales Evaluación de campos medios Evaluación de eventos extremos Análisis de ciclones y medicanes: evaluación y clima futuro PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Resumen A weighting proposal for an ensemble of regional climate models over Europe driven by ERA40 based on monthly precipitation probability density functions EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Resumen Present-climate precipitation and temperature extremes over Spain from a set of high resolution RCMs PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Resumen Evaluation of present-climate precipitation in 25 km resolution RCM simulations over North-West Africa MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO Resumen Medicanes description and climate change projections with a multimodel ensemble of RCMs CONCLUSIONES Conclusiones generales Conclusiones específicas por capítulos Propuesta de conjunto ponderado de modelos regionales de clima Extremos de precipitación y temperatura de clima presente sobre España Precipitación de clima presente sobre el Noroeste de África Medicanes: descripción y proyecciones de cambio climático REFERENCIAS IV

9 ANEXO V

10

11 1 INTRODUCCIÓN

12

13 1INTRODUCCIÓN 1.1 La Península Ibérica y su entorno La Península Ibérica (PI de aquí en adelante), situada en el extremo suroeste del continente europeo, en latitudes medias, es la región escogida para realizar el estudio del clima llevado a cabo en esta tesis. La situación y las características de la PI la convierten en todo un reto para el estudio del clima, tanto del clima presente como del clima futuro. En el informe de Impacto del Cambio Climático en España (Castro et al., 2005) así como en el Informe de Evaluación del Cambio Climático Regional de la Red Temática CLIVAR-España (Sánchez y Miguez-Macho, 2010) se recogen las principales características del clima de la PI. Pese al relativamente reducido tamaño de la PI ibérica (por debajo de los km2) la complejidad de su orografía y la variabilidad del clima son sus principales características. La situación de la PI latitudinalmente hablando, entre la zona tropical y la zona templada, rodeada por grandes masas de agua (fuentes constantes de humedad), con muchos kilómetros de costa (unos 6600 km) y su abrupto relieve hacen de ella una región muy interesante de cara al estudio del clima (Font, 1983). El relieve de la PI se caracteriza por amplias mesetas con una altitud media de unos 600 m sobre el nivel del mar y alineaciones montañosas repartidas por todo el territorio, con picos por encima de los 3000 m tanto en el norte (Pirineos) como en el sur (Cordilleras Béticas), mientras que en la zona interior de la PI los picos más destacados llegan a los 2300 m en el Sistema Ibérico y a los 2500 m en el Sistema Central. Otras características de la PI son las depresiones, llanuras y valles que cruzan todo el territorio y que junto a las cordilleras actúan a modo de pasillos naturales (ver Figura 1.1). La variabilidad tanto espacial como temporal (estacional e interanual) de la temperatura y la precipitación (Castro et al., 2005; González-Hidalgo et al., 2011) en la PI son muy importantes. Las diferencias medias de temperatura entre distintos puntos de la PI pueden alcanzar los 18ºC, mientras que la precipitación puede variar entre los 150 mm año-1 y los 2500 mm año -1. La estacionalidad en la PI es muy marcada aunque las características de cada estación varía mucho de una región a otra, mientras que la variabilidad interanual, principalmente en la precipitación da lugar a importantes períodos secos. 3

14 1INTRODUCCIÓN Figura 1.1. Mapa físico de España. (Fuente: En la zona norte de la PI se da el clima oceánico, atlántico o templado húmedo, con lluvias frecuentes a lo largo de todo el año. El resto de la PI recibe la influencia directa de las altas presiones subtropicales, con grandes diferencias entre el verano y el invierno, encontrando clima mediterráneo en sentido estricto en la franja costera del Mediterráneo, clima mediterráneo continental en el centro y sur de la PI y clima mediterráneo oceánico en Portugal y la costa sur. Además, las características de la PI provocan que también se encuentren otros climas a nivel más local, como pueden ser el clima semidesértico (Almería, Murcia y Alicante) o el clima de alta montaña (Pirineos y Sistemas Béticos). En la Figura 1.2 se muestra la clasificación climática de la PI y las Islas Baleares siguiendo la clasificación propuesta por Köppen en 1936, llamada clasificación de Köppen-Geiger (AEMET, 2011). 4

15 1INTRODUCCIÓN Figura 1.2. Clasificación climática de Köppen-Geiger en la Península Ibérica e Islas Baleares. (Atlas Climático Ibérico). ET: Tundra, Dfc: frío sin estación seca y verano fresco, Dfb: frío sin estación seca y verano templado, Dsc: frío con verano seco y fresco, Dsb: frío con verano seco y templado, Cfb: templado sin estación seca con verano templado, Cfa: templado sin estación seca con verano caluroso, Csb: templado con verano seco y templado, Csa: templado con verano seco y caluroso, BSk: estepa fría, BSh: estepa cálida, BWk: desierto frío, BWh: desierto cálido. (AEMET, 2011) 5

16 1INTRODUCCIÓN Según González-Hidalgo et al. (2011) las cadenas montañosas determinan en gran parte la distribución espacial de la precipitación, que es el elemento climático más variable tanto en el espacio como en el tiempo (New et al., 2001; Mitchell and Jones, 2005; Karagianidis et al., 2008). Las cadenas montañosas más importantes se encuentran dispuestas en sentido oeste-este provocando que en esa dirección las masas de aire circulen longitudinalmente a lo ancho de la PI sin ningún impedimento, pero haciendo que las masas de aire que circulan en sentido norte-sur tengan que atravesar todas las cordilleras (Serrano et al., 1999). A las características siempre variables de la precipitación hay que añadirle que los climas del oeste del Mediterráneo se caracterizan por una variabilidad espacial y temporal de la precipitación especialmente relevante (Romero et al., 1998). El máximo de la precipitación se da en el Norte y Oeste de la PI, mientras que el mínimo se encuentra en el E y el S (Nieto et al., 2004). El promedio de la precipitación anual en toda la PI es de unos 600 mm año-1, pero mientras que en Galicia o el País Vasco la precipitación anual puede superar los 2500 mm año -1, en zonas como el Cabo de Gata apenas llega a los 150 mm año -1 (Martín-Vide, 2004). Además, la variabilidad temporal de la precipitación juega un papel muy importante en el clima de la PI ya que en grandes áreas de la PI se da una alta intensidad de precipitación, pero con una distribución temporal muy pobre, según Martín-Vide (1994) se dan altos porcentajes de la precipitación total en pocos días seguidos de períodos de sequía, lo que ocasiona un gran estrés hídrico. Un ejemplo lo tenemos en la región de Valencia, en el este peninsular, donde se da la mayor precipitación diaria de todo el Mediterráneo (llegando a valores de 800 mm, Peñarrocha et al., 2002), con importantes episodios de sequías seguidos de lluvias torrenciales que pueden superar la precipitación media anual. La variabilidad espacial de la temperatura es un aspecto muy importante en el clima de la PI, los valores mínimos de la temperatura media anual varía entre valores inferiores a los 2.5º C en las zonas de alta montaña (Pirineos) y valores superiores a los 18º C que se alcanzan en el sur de la PI. El aspecto más reseñable de la temperatura en la PI es el valor de las temperaturas máximas durante el verano, según el Atlas Climático Ibérico elaborado por la AEMET el número de días al año cuya temperatura mínima supera los 20º C es de 60 en muchas regiones de España mientras que el número de días con temperatura máxima superior a los 25º C es superior a los 150. Una gran influencia que recibe el clima de la PI es la del norte de África. La PI se encuentra separada del Norte de África por 14,4 km que forman el Estrecho de Gibraltar, que a su vez sirve de comunicación entre el Mar Mediterráneo y el Océano 6

17 1INTRODUCCIÓN Atlántico. La cercanía a la PI y la influencia que ejerce en su clima hacen del Norte de África una región de especial interés en este estudio. Se denomina Magreb al Norte de África y comprende los países de Marruecos, Túnez y Argelia, esta zona se encuentra delimitada por el Océano Atlántico, el Mar Mediterráneo y el desierto del Sáhara. La orografía, como ocurre en la PI, es muy variada ya que comprende muchos kilométros de costa y cadenas montañosas muy importantes como el Atlas y el Rif, además de una vasta extensión de desierto. La zona del Magreb también está muy influenciada por las extensiones de agua que la rodean y por la orografía. El clima en esta región tiene una gran variabilidad espacial pasando del clima mediterráneo al clima desértico en pocos kilómetros, sin olvidar el clima de alta montaña. 1.2 Cambio climático antropogénico Según el último informe del Panel Intergubernamental de Expertos sobre el Cambio Climático (IPCC, 2013) el calentamiento global es innegable y los cambios observados desde 1950 no tienen precedente en la Tierra, a pesar de los cambios climáticos que se han dado a lo largo de sus 4500 millones de años de vida. Para poder entender y adaptarse al cambio climático es esencial, por un lado, entender el clima actual y por otro lado, crear las herramientas adecuadas para estudiar el clima futuro y así poder adelantarse e intentar en la medida de lo posible disminuir sus efectos. Las causas de los cambios climáticos antes de la existencia del hombre no está muy clara y se apunta a diversos motivos como variaciones en las concentraciones de los gases de efecto invernadero, mayor o menor actividad volcánica ( volcanic forcing, Gao et al., 2008), impactos de meteoritos, diferencias en la órbita terrestre ( orbital forcing, Laskar et al., 2004), actividad solar ( solar forcing, Wenzler et al.,2005) o movimientos tectónicos con su correspondientes cambios en la orografía (creación de cordilleras, cierres de mares, modificación de corrientes marinas, etc.). Los cambios de clima durante el último milenio se relacionan principalmente con la actividad solar (Magny et al., 2008; Martín-Puertas et al., 2008), aunque también se han observado enfriamientos debidos a las erupciones volcánicas (de Silva y Zielinski, 1998; D'Arrigo et al., 2004; Stone, 2004), pero estos se relacionan con cambios de menor importancia a escala global y de muy corta duración. Sin embargo los estudios sobre el calentamiento sufrido en las últimas décadas indican que la variabilidad interna no puede explicar el calentamiento sufrido desde 1951 y que los cambios en la actividad solar no han contribuido a dicho calentamiento (Bindoff et al., 2013) y apuntan a un origen concreto, por eso se le ha denominado cambio climático antropogénico. Según el IPCC (Bindoff et al., 2013) las actividades 7

18 1INTRODUCCIÓN humanas son las causantes de más de la mitad del incremento observado en la temperatura en superficie media global desde 1951 a 2010 con una fiabilidad entre el 95 y el 100 %. Recientes estudios recogidos en dicho informe también han encontrado evidencias de la influencia de la acción humana en la variación de los eventos extremos. La principal característica del calentamiento antropogénico es su carácter global, la temperatura media de la superficie terrestre se ha incrementado desde el siglo XIX en todo el globo, independientemente que se trate de una zona de tierra como de mar, registrándose en las últimas tres décadas un aumento progresivo década a década, a pesar de la variabilidad interanual o decadal (Stocker et al., 2013). Los valores extremos también han sufrido variaciones a escala global en los últimos 60 años encontrándose un decrecimiento en el número de días y noches fríos y un incremento en el número de días y noches cálidos ( % de probabilidad). También se ha detectado que la longitud y la frecuencia de las olas de calor ha aumentado desde mediados del siglo XX. El IPCC (Stocker et al., 2013) recoge cambios en los eventos de precipitación extrema, cuyo número se ha incrementado en más lugares de los que ha disminuido. En Europa el número de eventos de precipitación intensa ha aumentado en intensidad y frecuencia con algunas variaciones regionales y estacionales. En cuanto a los períodos de sequía no se ha encontrado ninguna tendencia global aunque esto puede ser debido a la falta de estudios y observaciones directas. Sin embargo en el Mediterráneo y el Oeste de África sí se ha encontrado que la frecuencia y la intensidad de las sequías ha aumentado desde 1950 (Hartmann et al., 2013) Cambio climático observado en la PI Según Kovats et al. (2014) la PI es una región amenazada por el cambio climático, un cambio climático que afecta de una manera u otra a todos los climas presentes en dicha región. Esta vulnerabilidad de la PI al cambio climático se ve reflejada en los estudios paleoclimáticos realizados en las últimas décadas y que muestran la gran sensibilidad de la región a la variabilidad climática global a diferentes escalas de tiempo (Cacho et al., 2010). Dada la sensibilidad de la PI se han producido numerosos estudios referentes al cambio climático observado en dicha región, algunos centrados en la PI en su conjunto mientras que otros centran su atención en una determinada subregión. Moberg et al. (2000) indican que se ha producido un calentamiento de hasta un 5% en la PI en el período , pero la mayoría de los estudios se centran en los 8

19 1INTRODUCCIÓN cambios experimentados en el clima en la segunda mitad del siglo XX (Klein Tank et al., 2002). Los diversos estudios coinciden en que el calentamiento global que se ha producido en la PI en el s XX varía según se trate de la temperatura máxima o la mínima y también según la estación del año. Además el calentamiento no ha sido uniforme ni temporal ni espacialmente; durante el siglo XX han habido períodos en los que el calentamiento ha sido mayor y, dependiendo de las regiones, ha sido más o menos notable (Staudt et al., 2005). Estudios sobre toda la PI también han detectado un aumento general de la temperatura durante el siglo XX (Karl et al., 1993; Brunet et al., 2005)) de entre 0.1 y 0.2ºC década-1 (del Río et al., 2011), mientras que del Río et al. (2012) destacan un aumento de 0.3ºC década-1 de la temperatura anual en el 90% de España. Brunet et al. (2006) distinguen entre el aumento experimentado por la temperatura máxima (0.12ºC década-1) y el experimentado por la temperatura mínima (0.10ºC década -1), coincidiendo con del Rio et al. (2012) en que la temperatura máxima ha aumentado más que la mínima. Oñate y Pou (1996) distinguen que el norte y el noroeste de la PI, donde las temperaturas serían ahora menos extremas que a principios del siglo XX y el sudeste y el centro este de la PI donde las temperaturas habrían experimentado un aumento de ambos extremos. Según Sanz-Elorza et al. (2003) la temperatura en las altas montañas de la PI ha aumentado desde 1940, apreciándose el incremento tanto en las temperaturas máximas como en las mínimas mientras que la precipitación mensual se ha redistribuido. Según Castro et al. (2007) a finales del s XXI los modelos climáticos proyectan un aumento de las regiones de la PI y en el norte de África (Gallardo et al., 2013) con un clima seco semiárido. Dada la gran variabilidad climática de la PI y los distintos resultados encontrados según la zona a estudiar se han realizado muchos trabajos centrados en una única subregión de la PI. Algunos ejemplos de estudios centrados en una única región son Serra et al. (2001) en Barcelona, El Kenawy et al. (2012) en el noreste de la PI o del Río et al. (2005) en Castilla y León, por ejemplo. Los estudios de los eventos extremos sobre la PI también indican un aumento de la temperatura durante el s. XX (Klein Tank y Können, 2003). Della-Marta et al. (2007) señalan que la longitud de las olas de calor se ha duplicado durante el último siglo mientras que la frecuencia de los días cálidos (días cuya temperatura máxima es superior al percentil 95) se ha triplicado. 9

20 1INTRODUCCIÓN 1.3 Herramientas para el estudio del clima En las últimas décadas se está realizando un gran esfuerzo en estudiar las causas y las consecuencias del calentamiento global. En esta línea se encuentran los informes del IPCC (Stocker et al., 2013) y a nivel más local el informe del comité CLIVAR-España (Climate VARiability and Predictability, Bladé et al., 2010). Diversos estudios muestran a los modelos numéricos como la mejor herramienta con la que se cuenta actualmente para realizar proyecciones climáticas (Gutiérrez y Pons, 2006; Räisänen, 2007, Flato et al., 2013) Modelos de clima (GCM, RCM) Los modelos de clima son la principal herramienta para simular el clima histórico o paleoclima, para estudiar la variabilidad y el cambio del clima a escala estacional o decadal, para hacer proyecciones de clima futuro y para conseguir esas proyecciones con detalle a escalas regional y local (Flato et al., 2013). Por otro lado, los Modelos Regionales de Clima (RCMs) (Laprise, 2008; Rummukainen, 2010) se han convertido en los últimos años en una herramienta muy importante para mejorar el detalle espacial de las proyecciones de cambio climático obtenidas por los Modelos de Circulación General (GCMs) (Giorgi y Mearns, 1999; Christensen et al., 2007b) y potencialmente también para una mejor descripción de eventos extremos (Durman et al., 2001). Un modelo climático es un programa de ordenador que intenta simular los procesos, tanto físicos como químicos, que ocurren en la atmósfera. Los modelos pueden ser desde muy simples a muy complejos, en una, dos o tres dimensiones y pueden encargarse de un único proceso físico que sea relevante para el clima o intentar simular toda la atmósfera, incluyendo sus interacciones con el suelo y el océano ( Los modelos climáticos son modelos matemáticos que resuelven de manera numérica las ecuaciones matemáticas que rigen los procesos atmosféricos (evolución de un fluido en movimiento, cambio de fase del agua, termodinámica para la conservación de la energía o gases ideales). Para poder plantear estas ecuaciones el espacio se divide en celdillas tridimensionales (x y z), desde la superficie terrestre hasta el tope de la atmósfera y cubriendo todo el dominio que se quiera estudiar; en cada una de estas celdillas se calcularán las variables tales como presión, temperatura, humedad, viento, precipitación... Los GCMs son los modelos elegidos para simulaciones muy largas sobre todo el globo mientras que los RCMs se usan para simular con mayor detalle, tanto espacial como temporal, el clima regional partiendo de los resultados obtenidos con los 10

21 1INTRODUCCIÓN GCMs. Los resultados de los GCMs, debido a su escasa resolución horizontal (entre 1 y 5 grados de latitud), no pueden ser utilizados para analizar el clima regional, pero se usan como datos de partida para las simulaciones de los RCMs (proceso llamado anidamiento), obteniendo así una resolución espacial mucho mayor (entre 50 y 10 km2). Usar una resolución mayor hace posible simular de una mejor manera la línea de costa, la orografía o los tipos de suelo y esto permitirá estudiar los fenómenos locales, este detalle es necesario para hacer estudios de evaluación de impactos (Gao et al., 2006). Al anidar un RCM en un GCM el modelo regional heredará los posibles fallos o limitaciones del GCM por lo que al estudiar los resultados de las simulaciones de un RCM habrá que tener en cuenta este aspecto. Por este motivo en los trabajos de conjuntos de RCMs es conveniente trabajar no sólo con un conjunto de RCMs sino también con un conjunto de GCMs para analizar correctamente la dispersión entre las distintas simulaciones. Una limitación muy importante, y en muchas ocasiones limitante, al realizar las simulaciones de clima por medio de los modelos climáticos es el coste computacional. En los últimos años tanto la velocidad como la capacidad de cálculo de los procesadores ha aumentado considerablemente permitiendo una mayor rapidez en la realización de las simulaciones. Las mejoras informáticas han supuesto un gran avance en el uso de los modelos climáticos ya que permiten hacer simulaciones más largas y con una mayor resolución tanto espacial como temporal al reducir tanto el coste como el tiempo de realizar dichas simulaciones. Los modelos climáticos se encuentran en continua evolución, mejorando la descripción de los procesos físicos, introduciendo nuevos componentes o ampliando la resolución (horizontal, vertical y temporal). Flato et al. (2013) recoge las mejoras experimentadas por los RCMs desde el Cuarto Informe del IPCC como las parametrizaciones que usan dichos modelos para representar los procesos que no pueden ser resueltos explícitamente, por ejemplo los procesos relacionados con la convección atmosférica, con la microfísica de nubes o con los aerosoles (Flato et al., 2013). Estas mejoras han supuesto una mejor descripción por parte de los RCMs de los principales aspectos del clima regional tanto en precipitación como en temperatura, aunque los resultados reflejan una mayor dificultad a la hora de simular la precipitación, dada la complejidad de esta variable. En cuanto a las tendencias registradas en los eventos extremos durante la segunda mitad del s. XX, los RCMs las reproducen bien, pero tienen a sobrestimar el calentamiento de las máximas y a subestimar el de las mínimas. Por otro lado los RCMs han mejorado la simulación de los ciclones extratropicales. Estos resultados avalan la necesidad de evaluar 11

22 1INTRODUCCIÓN exhaustivamente los resultados de los RCMs de la manera más completa y variada posible. Flato et al. (2013) también recoge el valor añadido que suponen los RCMs frente a los GCMs, especialmente en regiones con una orografía variada, como es la zona que nos ocupa, así como para estudiar fenómenos de pequeña escala, como puede ser la precipitación convectiva. Por este motivo todos los trabajos que se recogen en esta tesis se han realizado utilizando conjuntos de RCMs. Para la evaluación de los modelos climáticos es necesaria la disponibilidad de bases de datos observacionales de buena calidad mientras que la obtención de unos resultados fiables de proyecciones de clima futuro está relacionada con el desarrollo de los modelos (Hewitson et al., 2014). Un dato que aporta información muy importante para la evaluación de los modelos será la dispersión entre los diferentes modelos utilizados (Flato et al., 2013) por lo que la utilización de diferentes RCMs en los trabajos de evaluación es fundamental. 12

23 2 METODOLOGÍA

24

25 2METODOLOGÍA 2.1 Conjuntos de RCMs Los métodos de conjuntos de RCMs se usan para estudiar las incertidumbres en las simulaciones de los modelos derivadas de la variabilidad interna, de las condiciones de contorno utilizadas, de los parámetros de los modelos o incluso heredadas de los GCMs en los que se anidan los RCMs (Flato et al., 2013). La dispersión encontrada entre los diferentes resultados aporta una información muy valiosa en la evaluación de los RCMs, especialmente en regiones, variables o períodos en los que las observaciones sean escasas. La discrepancia entre los diferentes resultados y entre estos y las observaciones son un indicador muy importante para conocer las parametrizaciones y características de los modelos más sensibles en la descripción del clima regional. Importantes proyectos europeos han hecho posible la realización de simulaciones conjuntas de RCMs para analizar tanto el clima presente como el clima futuro. Los proyectos más reseñables son PRUDENCE ( Prediction of Regional scenarios and Uncertainties for Defining EuropeaN Climate shcange risks and Effects, Christensen y Christensen, 2007), ENSEMBLES (ENSEMBLES-based Prediction of Climate Changes and their Impacts, Van der Linden y Mitchell, 2009), ESCENA (Generación de escenarios regionalizados de cambio climático en España con modelos de alta resolución, Jiménez et al., 2013) o CORDEX (World Climate Research Program Coordinated Regional Downscaling Experiment, Giorgi et al., 2009), por ejemplo. En todos ellos se han realizado simulaciones tanto de clima presente como de clima futuro intentando abarcar PRUDENCE fue el primer gran proyecto europeo ( ) en el que se consideró un conjunto de RCMs con una resolución de 50 km sobre el continente europeo. Los períodos simulados en este proyecto fueron para las simulaciones de control (Jacob et al., 2007) y para las simulaciones de escenarios futuros de clima (Beniston et al., 2007; Christensen et al., 2007a; Déqué et al., 2007; Boberg et al., 2009; Sánchez et al., 2011). A continuación, en el proyecto europeo ENSEMBLES ( ) participaron más GCMs y más RCMs con el objetivo de obtener una matriz de simulaciones más completa para poder determinar si las incertidumbres son debidas a los modelos globales o a los regionales. Las simulaciones comenzaban en el año 1950, llegando en algunos casos hasta el año 2100 de manera continua y se realizaron con una resolución de 25 km (Christensen et al., 2008; Giorgi y Lionello, 2008; Boberg et al., 2010; Herrera et al., 2010; Kjellstrom et al., 2010; Boberg y Christensen, 2012; López-Franca et al., 2013a, b; Romera et al., 2015b). 15

26 2METODOLOGÍA Otro importante proyecto europeo en esta misma línea es CORDEX. En este proyecto se pretenden realizar simulaciones de cambio climático sobre prácticamente todo el globo, para cubrir zonas poco estudiadas por los modelos. La primera región estudiada en este proyecto fue África (Nikulin et al., 2012). En el caso del continente europeo, que sí ha sido objeto de numerosos estudios previos, se han realizado simulaciones de muy alta resolución (12,5 km) en el marco de este proyecto (Vautard et al., 2013; Jacob et al., 2014; Kotlarski et al., 2014). El proyecto español ESCENA ha centrado su dominio sobre la PI, incluyendo gran parte del Atlántico, así como las Islas Canarias. El dominio europeo usado en los proyectos anteriores (PRUDENCE y ENSEMBLES) dejaba a la PI demasiado cerca de la zona de contorno y en ningún caso cubría el archipiélago Canario. Se han realizado simulaciones a 25 km de resolución usando 5 RCMs, anidados en 4 GCMs y usando distintos escenarios de emisiones SRES. Los períodos simulados son para clima presente y para clima futuro. Para analizar los resultados climáticos procedentes de un conjunto de RCMs se han propuesto diferentes metodologías y aproximaciones. La opción más extendida consiste en hacer un promedio de los resultados de los distintos modelos participantes en el estudio para obtener un único resultado, sin tener en cuenta qué modelos son más cercanos a las observaciones en cada caso. En el proyecto ENSEMBLES se propuso hacer un promedio ponderado de las distintas simulaciones adjudicando un peso a cada unos de los RCMs. Estos pesos o factores dependerían de la región a estudiar y de la variable, ya que un mismo RCM puede dar distintos resultados según la variable o la región. La publicación científica Climate Research dedicó en 2010 un número especial (volumen 44, 2-3) a las propuestas realizadas en este sentido, en ese número especial se encuentra, por ejemplo, el estudio de Christensen et al. (2010) que recoge el primer intento de ponderar los resultados de los modelos y calcula el peso de cada RCM combinando seis métricas distintas para poder tener en cuenta distintos aspectos del RCM. En el capítulo 3 de esta tesis se incluye una propuesta en este línea (Sánchez et al., 2009) El Modelo Regional de Clima PROMES El modelo regional de clima PROMES (Sánchez et al., 2004; Domínguez et al., 2010) ha participado con sus simulaciones en todos los proyectos anteriormente mencionados, así como en los proyectos AMMA (African Monsoon Multidisciplinary Analysis, Redelsperger et al. 2006), CLARIS y CLARIS-LPB (A Europe-South America network for climate change assessment and impact studies La Plata Basin, Boulanger et al., 2010), aportando simulaciones tanto de evaluación, como de clima 16

27 2METODOLOGÍA presente y las proyecciones de clima futuro. En cada proyecto ha sido necesario adaptar PROMES a un dominio nuevo (Europa, España, África y Sudamérica), usando distintos datos de reanálisis para hacer las simulaciones de evaluación (ERA40 o ERA Interim) y se anidando en distintos GCMs según el proyecto, concretamente en el marco del proyecto ESCENA se anidó PROMES en cuatro GCMs distintos. La participación en todos estos proyectos ha hecho que el modelo esté sometido a una continua evolución y mejora, introduciendo nuevas parametrizaciones y adaptando el modelo a los distintos dominios y datos iniciales. Los resultados mostrados en los capítulos del 3 al 6 han sido posibles gracias a la gesión de las simulaciones en los supercomputadores en los que se han llevado a cabo los cálculos, así como el complejo, laborioso y esencial trabajo informático de postproceso necesario para ajustar las salidas del modelo a los estándares netcdf y CF (Climate and Forecast, exigidos y consensuados por la comunidad climática internacional para cada uno de los trabajos de intercomparación en los que se ha participado. Estas tareas son esenciales para que la comunidad científica disponga de bases de datos de simulaciones de modelos robustas y coherentes para su uso, tanto en estudio de clima, como su aplicación como información de entrada para estudios de impacto de clima. Como se muestra en la sección de resultados, en la intercomparación entre los modelos, PROMES ha obtenido resultados comparables al resto de participantes (Jacob et al., 2007 y Jiménez-Guerrero et al., 2013, por ejemplo). 2.2 Datos observacionales en malla La evaluación de los RCMs es un punto crucial en los estudios de clima y la comparación de las simulaciones con observaciones es un requisito previo para poder usar los resultados de una manera fiable (Flato et al., 2013). Aunque se pueden hacer comparaciones con los datos recogidos en estaciones puntuales, la manera más completa de comprobar la habilidad de los RCMs es comparar con datos observacionales dispuestos en una rejilla o malla análoga a la que utilizan los modelos. La calidad de las bases de datos la determinan tanto la densidad de estaciones de medida utilizadas como los algoritmos de interpolación usados para obtener los datos dispuestos en malla. Al trabajar con datos observacionales dispuestos en rejilla hay que tener presente los datos de los que se ha partido para formar dicha base de datos, ya que habrá zonas en las que se disponga de muy pocas estaciones pero al transformar los datos a una rejilla con resolución espacial regular puede resultar engañoso. 17

28 2METODOLOGÍA La disponibilidad de las bases de datos observacionales, su calidad y su resolución tanto espacial como temporal dependerán de la región, el período y la variable que se quiera comparar. La escasez de bases de datos observacionales de calidad es una de las principales dificultades a la hora de evaluar los resultados de los RCMs. Dependiendo de la región, la variable y el período que se quiera evaluar se dispondrá de unas bases de datos u otras. La mayor parte de las bases de datos cuenta con datos mensuales por lo que será especialmente complicado validar ciertos análisis, como la evaluación de los eventos extremos al no disponer de datos diarios. También hay escasez de datos observacionales en los mares y océanos por lo que la comparación de las salidas de los RCMs con observaciones en estas zonas o en ciertas islas será muy difícil de llevar a cabo. Según el IPCC (2014) es importante usar varias bases de datos observacionales, especialmente en zonas con una orografía compleja, debido a la dispersión encontrada entre ellas (Waliser et al., 1999; Kim y Lee, 2003; Nikulin et al., 2012) y de esta manera reducir la incertidumbre en el conocimiento del clima observado. A continuación se detallan los principales aspectos de las bases de datos utilizadas para obtener los resultados que se exponen en esta tesis (ver Tabla 1.1): 18 CRU (Climate Research Unit, Mitchel and Jones, 2005). Esta base de datos tiene una resolución espacial de 0.5º y cuenta con datos mensuales de precipitación y temperaturas media, máxima y mínima, para todo el globo. Ha sido usado en multitud de estudios (Giorgi et al., 2004, Jacot et al., 2007), pero su baja resolución tanto temporal como espacial limita su utilización para estudios más locales o de eventos extremos. ECA (European Climate Assesment & Dataset, Haylock et al., 2008). Esta base de datos tiene una resolución espacial es de 0.25º y cuenta con datos diarios. Por contra no abarca todo el globo ya que se trata de un proyecto europeo y por lo tanto no se ha podido usar en el estudio realizado sobre el norte de África que se incluye en esta tesis (capítulo 5). UDEL (University of Delaware, Willmott y Matsuura, 2000). Las variables incluidas en este caso son precipitación y temperatura media mensuales con una resolución espacial de 0.5º. GPCC (Global Precipitation Climatology Center, Schneider et al., 2011). Esta base de datos sólo tiene datos de precipitación, son datos mensuales con una resolución de 0.5º. CPC (Climate Prediction Center, National Oceanic and Atmospheric

29 2METODOLOGÍA Administration, Chen et al., 2008). En este caso la precipitación es la única variable disponible con una resolución espacial de 0.5º. Al tratarse de datos diarios sobre todo el globo permitirá su utilización en el caso de estudios de precipitación extrema. Spain02 (Herrera et al., 2012). Está base de datos tiene una gran resolución, tanto espacial (0.2º) como temporal (datos diarios) por lo que es de gran utilidad para comparar con las salidas de los RCMs con una resolución similar. Cuenta con datos de precipitación y temperaturas media, máxima y mínima. El dominio de esta base de datos cubre España Peninsular y las Islas Baleares. Resolución Temporal CRU Mensual Resolución Espacial Variables 0.5º Precipitación Temperatura media Temperatura máxima Temperatura mínima ECA Diaria 0.25º Precipitación Temperatura media Temperatura máxima Temperatura mínima UDEL Mensual 0.5º Precipitación Temperatura media GPCC Mensual 0.5º Precipitación CPC Diaria 0.5º Precipitación 0.2 Precipitación Temperatura media Temperatura máxima Temperatura mínima Spain02 Diaria Tabla 1.1: Resumen de las bases de datos observacionales usadas en los estudios recogidos en esta tesis. 19

30 2METODOLOGÍA 2.3 Cálculos y métricas utilizados En esta subsección se recogen los principales aspectos y cálculos que se han realizado para evaluar las simulaciones de clima de los RCMs. Dada la importancia de la trayectoria científica llevada a cabo durante la realización de esta tesis, no sólo se mencionan los cálculos que aparecen en las publicaciones de los capítulos de resultados (capítulos del 3 al 6) sino que se da un punto de vista más amplio haciendo referencia a los cálculos realizados en otras publicaciones científicas en las que la doctoranda ha colaborado y que se pueden consultar en el Anexo, estas publicaciones aparecen en negrita en esta subsección Selección de subregiones y bases observacionales En dominios complejos y para poder hacer un análisis más detallado es conveniente dividir la región a estudiar en subregiones como se ha hecho en Tapiador et al. (2009), Sánchez et al. (2009), Domínguez et al. (2013), Gómez et al. (2015) y en Romera et al. (2015a). Los criterios para elegir las regiones son distintos en cada uno de los estudios. En Tapiador et al. (2009) y en Sánchez et al. (2009) el dominio total comprende todo el continente europeo y las subregiones que se eligieron son las que ya se habían usado en estudios previos (Giorgi et al., 2004; Christensen and Christensen, 2007; Sánchez et al, 2007, por ejemplo), considerando la PI como una única región. En Domínguez et al. (2013) la PI es dividida en seis regiones, partiendo de las once cuencas descritas en Herrera et al. (2010), como ya habían hecho Argüeso et al. (2012) siguiendo la técnica propuesta por Argüeso et al. (2011), mientras que en Gómez et al. (2015) la PI se ha dividido en nueve clusters después de hacer un estudio detallado sobre qué método usar para hacer la división (Jain et al., 1999) y cuál era el número de clusters más apropiado para el dominio y la variable a estudiar (Rokach et al., 2013). Por último, en Romera et al. (2015a) se ha dividido la región del Magreb en ocho subregiones siguiendo el patrón de la precipitación. Otro aspecto muy importante a tener en cuenta a la hora de evaluar las simulaciones de los RCMs es la elección de la base o bases de datos observacionales que se van a utilizar. Es conveniente que las bases de datos ya hayan sido utilizadas en la región de estudio y se haya mostrado su validez en esa zona. También es importante que la resolución espacial de las bases de datos sea lo más parecida posible a la resolución espacial de los datos de salida del RCM, ya que de esa manera la comparación será más efectiva y fiable. Si las resoluciones espaciales de observaciones y RCMs son comparables se podrán interpolar todos los datos a una malla común, como se ha hecho por ejemplo en Jiménez-Guerrero et al. (2013) y 20

31 2METODOLOGÍA en Domínguez et al. (2013), de manera que la comparación entre simulaciones y observaciones es más inmediata, ya que se pueden representar las diferencias en cada punto. Por el contrario, en Romera et al. (2015a) no se ha considerado adecuado hacer una interpolación de los resultados dada la diferencia entre las resoluciones por lo que las diferencias entre simulaciones y observaciones se ha valorado siguiendo otros métodos. En cuanto a la resolución temporal, para el estudio de campos medios es suficiente con tener observaciones mensuales, mientras que para el estudio de campos extremos ha sido necesario contar con datos diarios de observaciones Evaluación de campos medios Para evaluar los resultados de las simulaciones de los RCMs en los distintos estudios realizados se han usado los siguientes métodos: Comparación visual de los resultados representados en mapas: precipitación media durante la estación lluviosa (Romera et al., 2015a) precipitación, temperatura mínima y temperatura máxima estacionales (Jiménez-Guerrero et al., 2013) desviación típica de las temperaturas máxima y mínima y coeficiente de variación de la precipitación (Jiménez-Guerrero et al., 2013) Diferencias entre simulaciones y observaciones del promedio espacial de la precipitación en cada subregión (Romera et al., 2015a) Cálculo del coeficiente de correlación entre simulaciones y observaciones: En Jiménez-Guerrero et al. (2013) se ha calculado el coeficiente de correlación temporal en cada punto así como el promedio espacial tanto para la precipitación como para las temperaturas máxima y mínima tanto anual como estacional. En Tapiador et al. (2009) se ha calculado el coeficiente de correlación espacial para la precipitación y la temperatura media en cada subregión analizada. 21 La función de distribución de probabilidad (PDF) se ha usado en Tapiador et al. (2009) para la precipitación y la temperatura mensuales, en Domínguez et al. (2013) para la precipitación diaria y en Gómez et al. (2015) para los valores diarios de viento a 10 m con objeto de analizar todo el espectro de valores. En el caso de la precipitación se ha representado en escala

32 2METODOLOGÍA logarítmica para poder apreciar con más detalle las colas de la función. En los tres casos se ha realizado una comparación visual de las formas de las PDFs. En Tapiador et al. (2009) se ha comparado la forma completa de las PDFs usando distintos parámetros estadísticos: percentiles 25 y 90 y los cuatro primeros momentos (media, desviación estándar, asimetría y curtosis) En Domínguez et al. (2013) y en Gómez et al. (2015) se ha usado el coeficiente de Perkins (Perkins et al., 2007) para comparar las PDFs. En Jiménez-Guerrero et al. (2013) se han usado diagramas de Taylor (Taylor, 2001) para comparar la desviación estándar al mismo tiempo que la correlación espacial tanto para la precipitación como para las temperaturas máxima y mínima. El ciclo anual de la precipitación en cada subregión se ha calculado en Romera et al. (2015a). Para estudiar la variabilidad interanual en Romera et al. (2015a) se han calculado las anomalías de la precipitación anual media en cada subregión. En Sánchez et al. (2009) se ha buscado un método para tratar los resultados de un conjunto de RCMs, calculando los pesos que habría que aplicar a cada RCM, utilizando las CDFs (función de distribución de probabilidad acumulada) de la precipitación mensual. Al contar con un período simulado de cuarenta años ( ) se han utilizado los primeros treinta años para calcular dichos pesos y los últimos diez años ( ) para validar el método Evaluación de eventos extremos Un aspecto muy importante a la hora de estudiar el clima de una región son los extremos y es necesario saber si los modelos son capaces de reproducir esos eventos extremos tales como períodos secos, eventos de precipitación extrema, olas de calor, olas de frío, etc. Según la climatología de la región a estudiar predominarán unos eventos sobre otros por lo que la elección de los eventos extremos a analizar dependerá de la región elegida. Existen distintos índices y parámetros estadísticos para estudiar este tipo de eventos (Zhang et al, 2011). A continuación se detallan los que se han usado en los estudios de Sánchez et al. (2011), López-Franca et al. (2013b, 2015), Domínguez et al. (2013) y Romera et al. (2015a): 22 Días secos (del inglés, dry days): se considera día seco al día cuya

33 2METODOLOGÍA precipitación no supera 1 mm día -1. A partir de esta definición se puede estudiar la longitud media anual de los períodos de días secos, la longitud máxima de los períodos de días secos o, por ejemplo, el número de períodos de días secos cuya longitud supere cierto número de días. Eventos de precipitación extrema: número de días cuya precipitación supere un cierto valor, por ejemplo 10 mm día -1 (precipitación intensa) o 20 mm día -1 (precipitación muy intensa). Fracción de precipitación debida a los eventos por encima del percentil 95. Percentiles 10 y 90 de las temperaturas máxima y mínima. Ola de calor: en Domínguez et al. (2013) se ha considerado ola de calor al período no inferior a cinco días cuya temperatura máxima supere los 35ºC. Este umbral se ha obtenido a partir del percentil 95 de los datos observados en todo el dominio analizado. A partir de esta definición se han analizado la duración máxima y el número medio por año de las olas de calor. Número de noches tropicales, entendiendo como noche tropical aquella en la que la temperatura mínima es superior a los 20ºC. Combinando las olas de calor con las noches tropicales se obtiene un índice distinto: número de días cuya temperatura máxima es superior a los 35ºC y la temperatura mínima es superior a los 20ºC. La comparación de estos índices entre observaciones y modelos se puede hacer de manera similar a la de los campos medios. Se puede comparar visualmente la distribución espacial mediante la representación en mapas (Sánchez et al., 2011; Domínguez et al., 2013; López-Franca et al., 2013b; Romera et al., 2015a) o por medio de historgramas (Romera et al., 2015a). Una comparación más objetiva se consigue mediante el cálculo, por ejemplo, del coeficiente de correlación de Pearson, como se ha hecho en Domínguez et al. (2013) para todos los estadísticos e índices de extremos analizados en el estudio. En Sánchez et al. (2011), por otro lado, se ha ajustado la distribución de la longitud media de los períodos secos a una distribución Weibull (Lana et al., 2008) y a continuación se ha realizado un test Kolmogorov-Smirnov (Von Storch and Zwiers, 1999) para comprobar si el ajuste era estadísticamente significativo o no. En los siguientes casos se han llevado a cabo otros cálculos para para realizar un análisis de los enventos extremos. En Domínguez et al. (2013) se han analizado la cola de las PDFs por encima del percentil 95 mediante el coeficiente de Perkins 23

34 2METODOLOGÍA (Perkins et al., 2007) para estudiar la precipitación intensa. Por otro lado, LópezFranca et al. (2015) se ha centrado en el estudio de las sequías en la PI analizando la persistencia de los días secos. En este análisis se ha aproximado la persistencia de los días secos a una cadena de Markov de segundo orden. En Drobinski et al. (2015) se ha buscado una relación entre los extremos de temperatura y precipitación en el Mar Mediterráneo usando también un conjunto de RCMs, pero en este caso los datos observacionales proceden de estaciones puntuales Análisis de ciclones y medicanes: evaluación y clima futuro Los resultados que se muestran en el capítulo 6 de la presente tesis doctoral requieren una explicación específica sobre el tipo de cálculos y análisis empleados en el mismo que se detallan en esta subsección. En la zona oeste del Mar Mediterráneo se han relacionado eventos de precipitación intensa y viento fuerte con la alta densidad de ciclones que se da en esa zona (Doswell et al., 1998). Estos procesos, junto con el trabajo anterior desarrollado en Gaertner et al. (2007), dio pie al trabajo de Romera et al. (2015b) donde se realiza un análisis de los ciclones con carácter tropical que se desarrollan en el Mar Mediterráneo (Medicanes) usando los datos de un conjunto de RCMs. La metodología empleada en este estudio se divide en tres pasos: la identificación de los ciclones, el cálculo de las trayectorias seguidas por dichos ciclones y la selección de los ciclones que tengan características tropicales. El primer paso es identificar todos los ciclones, independientemente de sus características, y para ello se ha usado el método descrito en Picornell et al. (2001) basado en la presión a nivel del mar (SLP). Una vez que se han detectado los centros de bajas presiones y se han aplicado los filtros correspondientes para eliminar las bajas demasiado pequeñas o demasiado cercanas, se procede al cálculo de la trayectoria seguida por la baja, utilizando el viento a 700 hpa y el viento máximo diario. En Flaounas et al. (2015) se ha aplicado este mismo método a un conjunto de modelos y se han comparado los resultados con los de otros cinco métodos más. Para clasificar los ciclones detectados y poder hacer un estudio de los medicanes se ha analizado la estructura vertical de cada uno de los ciclones y se han calculado los tres parámetros descritos en Hart (2003), que aportan información sobre la simetría de la baja y la temperatura de su núcleo. Los tres parámetros de Hart son B, que indica la simetría o asimetría del ciclón, -V TL, que aporta información sobre la temperatura del núcleo del ciclón en superficie (entre 900 y 600 hpa) y -V TU, que aporta información sobre la temperatura del núcleo en altura (entre 600 y 300 hpa). 24

35 2METODOLOGÍA Por medio de estos tres parámetros y aplicando los filtros basados en los resultados de Miglietta et al. (2013) se puede determinar si se trata de un ciclón con carácter tropical (medicane) o extratropical. En la Figura 2.1 se muestra un ejemplo de diagrama de Hart, donde se representan los valores de los tres parámetros, para un ciclón real registrado en el Océano Pacífico entre el 21 de agosto y el 9 de septiembre de 2015 (fuente: En un diagrama de Hart se representa toda la trayectoria del ciclón en función de los valores de los parámetros obtenidos y se observa si el ciclón muestra simetría y núcleo cálido (tanto en superficie como en altura) para poder clasificarlo como ciclón con características tropicales. En la Figura 2.1 se indica en color azul la zona de núcleo frío y en rojo la zona de núcleo cálido. Además el tamaño y los colores de los puntos de la trayectoria del ciclón también aportan información sobre la velocidad del viento y la intensidad del ciclón en cada punto. En el estudio de Romera et al. (2015b) se han utilizado este tipo de diagramas para ayudar a la caracterización de los ciclones pero no se muestran explícitamente. En Romera et al. (2015b), una vez seleccionados los medicanes entre todos los ciclones detectados se ha realizado la validación de los resultados obtenidos con los RCMs comparando los datos de los medicanes con los obtenidos por Miglietta et al. (2013). A continuación, para estudiar las variaciones que se pueden esperar en clima futuro se han analizado los siguientes aspectos: Velocidad máxima diaria en el centro de los medicanes (tanto el percentil 95 como la velocidad máxima) Evolución de la frecuencia anual de los medicanes Distribución mensual de los medicanes Distribución espacial de los medicanes dividiendo el Mar Mediterráneo en zonas: según la latitud, en zonas Norte y Sur según la longitud en zonas Oeste, Centro y Este 25 Distribución de los medicanes según su velocidad máxima diaria.

36 2METODOLOGÍA Figura 2.1. Ejemplo de gráfico de Hart para un ciclón real registrado entre el 21 de agosto y el 9 de septiembre de (Fuente: 26

37 3 PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Este capítulo reproduce el texto del siguiente manuscrito: Sánchez E, Romera R, Gaertner MA, Gallardo C, Castro M (2009) A weighting proposal for an ensemble of regional climate models over Europe driven by ERA40 based on monthly precipitacion probability density functions. Atmos Sci Let 10: , doi: /asl.230

38

39 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA 3.1 Resumen Se simula el clima presente sobre Europa usando 12 Modelos Regionales de Clima (RCMs), anidados en el reanálisis de ERA40. Se propone un método para puntuar los modelos según las funciones de densidad acumuladas (CDFs) de la precipitación mensual para el período para cada estación y ocho subregiones, comparando las simulaciones con la base de datos observacional CRU. El conjunto de las curvas CDFs es similar a las observaciones para todas las subregiones y estaciones. Los percentiles más altos (mucha cantidad de precipitación) muestran una dispersión mayor. Se han encontrado diferencias importantes entre las puntuaciones obtenidas por los modelos en cada región y estación. Aplicando esas puntuaciones para obtener un campo de precipitación ponderado para el período se obtienen resultados más cercanos a las observaciones que al considerar el conjunto de los modelos sin ponderar y en algunos casos se observa una amplia mejora. 29

40 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA 30

41 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA 3.2 A weighting proposal for an ensemble of regional climate models over Europe driven by ERA40 based on monthly precipitation probability density functions Abstract Present climate over Europe is simulated by 12 regional climate models (RCMs), forced by ERA40 reanalysis. A method is proposed to score models from the monthly precipitation cumulative density functions (CDFs) for each season and eight chosen subregions, compared with the CRU observational database. Ensemble CDF curves compare well against observations for all the subregions and seasons. Higher percentiles (heavy precipitation amounts) show a larger spread among results. Important differences in scores are obtained among models, regions and seasons. Applying the scores to compute weighted ensemble precipitation, results are slightly closer to observations than the direct (unweighted) ensemble, and some cases show a larger improvement. 1. Introduction Regional climate modelling (RCM) has become in the past years an important tool to improve our understanding of key processes involved in the description of climate mechanisms on regional scales (Giorgi and Mearns, 1999; Christensen et al., 2007). Many analyses from single RCMs focused on present and future regional climate have been made, such as Giorgi et al. (2004); Räisänen et al. (2004); Déqué et al. (2005); Sánchez et al. (2007). Also several efforts with a group of models simultaneously have been made to simulate common period, region and emission scenarios. An ensemble of models allows a more accurate description of the limitations, uncertainties and a probabilistic approach of future climatic projections. The DEMETER project (Palmer et al., 2004) is a well-known example of these studies, with global climate models. When dealing with future climate projections from an ensemble of RCMs, results of several studies are available (Vidale et al., 2003; Tebaldi et al., 2004; Christensen and Christensen, 2007; Jacob et al., 2007; Christensen et al., 2008). The PRUDENCE project (Christensen et al., 2007), focused on European climate, is a recent and successful example of these types of analyses. The result of an ensemble of models (Jacob et al., 2007) is in many cases closer to observations than any individual model (Palmer et al., 2004), perhaps due to compensating errors from different models. One possible approach to compare results further would be giving weights to model results when computing an 31

42 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA ensemble average (a single climate projection obtained from the average of all the models simulating the same period and conditions), which is then expected to obtain a closer result to the observed climate. The error obtained from multiple underdispersive models is sometimes the result of compensating errors, giving rise to a right answer, but for wrong reasons. A weighting or score, based on observations, also tries to improve this potential problem. Depending on the key process or mechanism that is considered, several scores or weightings will be obtained. Another challenging issue of this methodology would be their use to compute also ensemble results for future climatic conditions, supposing that these present-climate weightings will be right for future periods. The procedure proposed here could help in reducing uncertainties, although a more robust method for this purpose under future climate conditions from model simulations would need some convergence criterion, such as in Giorgi and Mearns (2002). Here we show the results obtained from a group of RCMs used in the European ENSEMBLES project (Hewitt, 2005; Christensen et al., 2008), where this type of analysis is one of their main goals. These models have simulated present climate ( ), forced with ERA40 reanalysis (Uppala et al., 2005). The analysis presented here will be focused on precipitation results. Precipitation is a complex and challenging quantity to be studied. It has an intrinsic complex and irregular structure in time and space, and many physical climatic processes are involved in its description. Several numerical parameterizations have been proposed for its representation in climate models (Jacob et al., 2007), and the uncertainties associated with the climatic change projections for future conditions due to increased greenhouse conditions are higher than, for example, temperature (Christensen et al., 2007; Déqué et al., 2007). The use of RCMs to describe precipitation has an additional point of interest when compared with global climate model (GCM) results, owing to the importance of regional scale features (Frei et al., 2003; Fowler et al., 2007). We propose a method to score models from the comparison of precipitation cumulative density functions (CDFs) against the observed values. Due to the complexity of precipitation mechanisms, and the use of the whole probability distribution, we can have confidence that high scores will correspond to a good model in terms of simulating their main climatic features. Several methods have been suggested to score an ensemble of models (generally GCMs), based on monthly or seasonal probability distribution values (Giorgi and Mearns, 2002; Shukla et al., 2006), or from daily ones (Perkins et al., 2007). In particular, the approach shown here is similar to Giorgi and Mearns (2002) reliability ensemble averaging (REA) analysis, but with two modifications here: RCM are used instead of GCM models; and cumulative density 32

43 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA functions are used instead of the statistical tool as the means. The objective of this study is to propose and validate a method to create scores from an ensemble of RCMs in terms of their ability to follow observed CDFs of monthly precipitation. In the ENSEMBLES project both 25 and 50 km horizontal resolution results are available, but we have chosen the 50 km (or 0.5 degrees) ones as the first test of the proposed analysis, considering that this resolution has been the more commonly used one for many of the RCM studies over Europe (Christensen et al., 1998; Gaertner et al., 2001; Vidale et al., 2003; Giorgi et al., 2004; Räisänen et al., 2004; Déqué et al., 2005; Christensen et al., 2007; Jacob et al., 2007). 2. Methodology of the analysis 2.I. Description of RCM models and simulations The results of 12 models (ALADIN (CHMI), ALADIN (CNRM), HIRHAM (DMI), CLM (ETHZ), HadRM3 (HC), RegCM (ICTP), RACMO (KNMI), HIRHAM (METNO), REMO (MPI), CRCM (OURANOS), RCA (SMHI) and PROMES (UCLM)) that take part in the ENSEMBLES project have been used. A detailed description of these models, except OURANOS (Laprise et al., 2003) can be found in Jacob et al. (2007). Precipitation analyses will be made on monthly time scales. Monthly or seasonal scales are of interest for many impact studies (Doblas-Reyes et al., 2006), and can easily be compared against the wellestablished CRU (climate research unit) climatology database (New et al., 1999). Although not shown here, monthly results have also been compared against the European Climate Assessment (ECA) (Haylock et al., 2008) observational database, showing very similar results. CRU results have been then chosen to score the models against the observations, as they have been used as the reference to compare RCM results in many studies over Europe, for basic statistics (Noguer et al., 1998; Gaertner et al., 2001; Rummukainen et al., 2001; Vidale et al., 2003; Hagemann et al., 2004; Giorgi et al., 2004; Räisänen et al., 2004; Jacob et al., 2007; Kjellström and Ruosteenoja, 2007), or the whole probability density functions (Tapiador et al., 2007, 2009). The domain used here covers the whole of Europe and the Mediterranean basin (see Figure 3.1), and uses a horizontal resolution around 50 km for the models with a Lambert conformal projection, or a 0.5 degree cell size for the models with longitude/latitude projection (only two use Lambert). In Jacob et al. (2007) more details of domain characteristics can be obtained. 33

44 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Figure 3.1. Domain and model example topography used for simulations. The eight regions over Europe are shown, where seasonal precipitacion CDFs are computed, defined in Christensen and Christensen (2007): BI (British Isles), IP (Iberian Peninsula), FR (France), ME (Mid Europe), SC (Scandinavia), AL (Alps), MD (Mediterranean) and EA (Eastern Europe) 2.II. Statistical analysis technique For a more detailed regional climatic analysis, the common domain has been divided into eight regions (Figure 3.1), as in Christensen and Christensen (2007). These regions are chosen considering that they show more or less homogeneous climatic characteristics. Cumulative probability density functions are obtained for each of the region and for each season. Typical population amounts are on the order of 10 4 values (3 months per year x 30 years x 500 cells per region at least). Bin size resolution is taken to be 0.5 mm month 1, allowing an accurate description of probability distributions of monthly precipitation, as they typically show values up to 250 or even 300 mm month 1 over many of the selected regions (Tapiador et al., 2007). The description of climate features through CDFs gives an additional point of interest and detail, as it describes not only the usual climatic characteristics (mean or variance), but the whole range of values at the same time (light and heavy precipitation amounts) and also the shape of the distribution, as a complex measure of the precipitation features. The CDF analysis of precipitation (named P in the following expressions) has been studied at other different timescales, such as daily (Perkins et al., 2007), which focused on extreme events, or monthly or even seasonal periods (Dessai et al., 2005; Tapiador et al., 2007; Tapiador and Sánchez, 2008; 34

45 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Tapiador et al., 2009). Here we construct the CDFs for the 8 regions during the first 30 years ( ) of the simulations, leaving the last 10 years ( ) to validate and test the results obtained. The comparison of modelled against the CRU observational CDFs allows us to create a skill score of the models. This is made through the following five factors (i = 1, 5) for each model j (f ij): A RCM A CRU 0.5 f 1 j=1 ( ) 2 A CRU (1) M A MRCM A CRU 0.5 f 2 j=1 ( ) 2 A MCRU (2) j j m f 3 j=1 ( m A RCM A CRU j m 2 A CRU 0.5 (3) ) P RCM PCRU 0.5 f 4 j =1 ( ) 2 PCRU (4) j σ RCM σ CRU 0.5 (5) f 5 j=1 ( ) 2 σcru where ARCMj, ACRU are the areas below the j RCM and CRU cumulative density function precipitation curves, and AM and Am are the fractional areas above (M) and below (m) the 50th percentile. Overbar denotes the spatial and time average and σ the standard deviation of the probability distribution function. Values of f ij factors around 1 indicate that the RCM is close to CRU observations, and values close to 0 mean that they are far from it. Each factor takes into account different aspects of model probability distribution characteristics: the distribution as a whole (through the mean and the total area), the smaller and higher precipitation amounts (50 th percentile limit), and the shape of the distribution (through the variance). Although there is some degree of oversampling among the factors, each one is focused on different aspects of precipitation, and therefore they globally can give a wider view of the ability of models to reproduce precipitation features. A final single weight (W j ) for each model j is obtained as j W j=f 1 j f 2 j f 3 j f 4 j f 5 j (6) This value can be seen as a measure of the capability of each model against observations, but also a method to evaluate the skill of models. 35

46 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Wj can also be used to define a weighted (wg) ensemble precipitation result from N models, which can be compared to the simple (unweighted, unwg) average: N N W j Rj Rwg = j=1 N Rj Rwg = j=1 N ; W j (7) j=1 To compute how close each result (model, season and region) is to observations, the following expression is proposed: np 1 [ M p C p + M p A= k=1 k k k+ 1 C p ( p k+1 p k )] k 1 (8) np 1 [ C p +C p k=1 k ( p k+1 pk )] k+ 1 where Mpk and Cpk are the modelled and CRUobserved precipitation for percentile p k respectively, ranging from 1 to the np percentiles considered. This expression is basically a measure of the area between observed and modelled CDFs, normalized by the total area of the observed curve. Therefore, the smaller the A parameter, the better the modelled distribution when compared with CRU results. 3. Results 3.I cumulative density functions Although the main objective of the study is the computation of skill scores, with the CDFs as the tool for that analysis, it is of interest to make a brief inspection of the comparison of the modelled monthly precipitation CDFs against observations. Figure 3.2 shows some examples (summer (JJA) and winter (DJF) for the Iberian Peninsula (IP), British Isles (BI), Eastern Europe (EA) and Mediterranean (MD) regions) of the modelled and observed CDFs for each model, season and region. To ease the comparison, the mean value at each percentile of all the 12 models, together with a±σ band (σ being the standard deviation among the 12 models) against observations, is shown. Although the shape of the density functions vary greatly between regions and seasons (especially the slope for the higher percentiles), the ensemble of modelled CDFs shows a general good agreement against CRU observations. This is clearly the case, for example, in the winter season over the Iberian Peninsula or the Mediterranean. Nevertheless, there are cases where the comparison is not as good. Sometimes an overestimation is obtained, such as in summer for IP and MD and in winter for EA, which is also larger for higher 36

47 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA percentiles. Sometimes, there is an underestimation for the whole distribution, such as summer at EA, or BI for both JJA and DJF. In contrast with these results, previous studies of current climate simulated over BI usually show an overestimation of mean precipitation for most seasons (Jones and Reid, 2001; Fowler et al., 2005; Jacob et al., 2007). But also some underestimation of mean or heavy daily precipitation values are also obtained (Fowler et al., 2005; Haylock et al., 2006; Jacob et al., 2007) depending on the model used and the season. It must be noticed that these analyses from RCM outputs are usually forced by GCMs, but here ERA40 is the forcing database. Far fewer studies have looked at the whole probability distribution, such as Tapiador et al. (2007, 2009), also with GCM forced RCM results, and there is a slight underestimation of the monthly probability distribution for the whole year when compared with observations. Christensen et al. (2008), using the same RCM simulations as here, although on daily scales, also obtain an underestimation of large precipitation amounts, which is the main bias of the BI CDFs shown here. Some general features can be noted, specifically related to higher percentile results: they show the larger spread among models (probably due to the limitations of the different parameterizations used to describe heavy precipitation processes); and the larger difference against observations. This last point could be partially explained because of the smoothing procedures used in the CRU database (Tapiador et al., 2007), which is likely to be more significant for larger precipitation amounts. 37

48 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Figure 3.2. Cumulative distribution functions for period of RCM ensemble compared with CRU observations. Winter (right column figures) and summer (left column figures) seasons for Iberian Peninsula, British Isles, Eastern Europe and Mediterranean regions are shown. CRU results are in blue and the red curve shows RCM model-averaged value, together with a light red band of ±σ width, where σ is the standard deviation for all the models at each percentile. 38

49 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA 3.II. Weighting values from CDFs Table 3.1 shows seasonal weighting values averaged for the eight regions (mean values for the whole of Europe) for each model, as computed from expression (6). Some general features can be seen: there is no season where models systematically obtain results closer to observations: two models are at their best in DJF, three in spring (MAM) and JJA, and four in autumn (SON). The different precipitation characteristics (local or large scale importance in some seasons compared with others) do not seem to have a significant influence in model performance. Nevertheless, on average (last line of Table 3.1), there is a slightly worse result for winter (0.234) and spring (0.245) than summer (0.280) or autumn (0.270). These differences are statistically significant for 95% bootstrap confidence intervals (Efron and Tibshirani, 1993). Looking at each model performance depending on the season, models exhibit clear differences in their ability to describe precipitation: their best scores are in many cases up to 50% better than their worst one. This large spread in skills could point to significant differences in the performance of numerical precipitation parameterizations used for each model. When comparing models against each other, results indicate that some give a better scores than others. The annual-averages range from of RCM12 to of RCM4. Therefore, the best score is almost twice the worst value. This large spread in the skill scores suggests using these numbers to compute a weighted-ensemble result. 39

50 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA DJF MAM JJA SON ann RCM RCM RCM RCM RCM RCM RCM RCM RCM RCM RCM RCM RCMAVG Table 3.1. Weights for each season (and annual mean on last column) and each model (and model mean on last line) for period, averaged over the eight regions described in Figure III. Validation of weighting for period We test if the weightings obtained from the period are able to give an improved weighted-ensemble result to precipitation values over a different period ( ). As this 10-year period follows the 30-year one used for computing the weightings, it is likely that they will show very similar climatological features, and the period (10 years) makes the analysis a little limited in terms of the inter-annual variability and uncertainty. On the contrary, as this period shows a strong NAO signal (Hurrell and Loon, 1997) that could make the periods more different, the proposed validation analysis would be then more relevant. 40

51 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA BI IP FR ME SC AL MD EA DJF MAM JJA SON wg unwg wg unwg wg unwg wg unwg wg unwg wg unwg wg unwg wg unwg Table 3.2. Weighted (from computed weights shown in Table 3.1) and unweighted ensemble seasonal mean ( ) precipitation. A vlues (formula 8) over each of the eight regions. Equation (8) quantifies how accurate the ensemble results are against observations for any period, and Table 3.2 shows the results obtained. The weighted precipitation values are slightly closer to observations (smaller A values) for most regions and seasons than the unweighted ones (29 results out of the 32 values). There is a small worsening in a few cases (e.g., Alpine region during DJF and MAM), but there are some cases where improvements are larger than the average. This is the case for the neighbouring regions of the Iberian Peninsula and Mediterranean during JJA, and also Mid Europe (ME) and Eastern Europe in DJF. Using a Monte Carlo method based on bootstrapping with replacement (Efron and Tibshirani, 1993; Bhend and von Storch, 2008; Boberg et al., 2009), applied to precipitation populations at each mode, region and season, these differences are statistically significant for 95% bootstrap confidence intervals. The corresponding weighted and unweighted CDFs are shown in Figure 3.3. Summer cases show a better improvement for central percentiles than for the extremes (especially the higher 41

52 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA values). This means that light and up to medium precipitation values are improved. As summer precipitation over these Mediterranean climate regions is closely related to convective (local) mechanisms, results could be related to model convective parameterizations. Both winter cases show smaller improvement than both summer results, no matter the slope they have for higher percentiles (large for ME region or small for EA). These important improvements over some regions and seasons are not inconsistent with the analysis of Table 3.1 weights. First, because an average over all the regions is made there, and so a highly improved region can be masked into a not so good total domain average; and second, due to the different period (close, but not equal) where these numbers are applied. Figure 3.3. Weighted and unweighted ensemble precipitation CDFs for period using weights. Summer (JJA) Iberian Peninsula and Mediterranean, and winter (DJF) Mid and Eastern Europe results are shown, as they are the cases where larger differences are obtained. CRU curve is on dashed black line, unweighted ensemble precipitation in green and weighted results in red. 42

53 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA 4. Conclusions A weighting proposal for an ensemble of 12 regional climate models is analyzed in terms of their monthly precipitation cumulative probability distribution functions for the period , forced with ERA40 reanalysis over Europe with around 50-km cell size resolution. The modelled CDFs used as the tool for the weighting procedure indicate overall a good agreement among models and against observations for most of the seasons and regions. Nevertheless, some discrepancies are also obtained, and a larger spread among models is obtained for higher percentiles. Weights vary between seasons and regions, but there are clear differences among model performances, as the best models have almost double the skill score compared with the worst. Therefore, it is likely that some changes can be obtained by using a weighted against an unweighted ensemble average from all the models. The application of weights of each model to obtain a weighted ensemble precipitation result gives a slight improvement for most of the regions and seasons, with some cases where the result is much better. Several lines of work follow the analysis presented here. One is a deeper analysis of the CDFs used to compute the weights, the differences shown among models and against observations for each of the seasons, and the physical mechanisms that could be behind those differences. Another is to extend this analysis to 25-km horizontal resolution model results, to test if the results obtained here are consistent when resolution is increased. The weighting procedure will then be applied, in the frame of the ENSEMBLES project, to future climate conditions, and compared with other weighting proposals. The use of CDFs to compute a weighting is expected also to be extended to other quantities, such as temperature. Acknowledgements This work was supported by the EU-funded ENSEMBLES project ( , GOCE-CT ). We thank the anonymous reviewers for their interesting and useful comments that have helped us to improve the quality and contents of the manuscript. We thank ECMWF for making the computing resources, used for the simulations with PROMES, available under Special Project SPESMG06. 43

54 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA 5. References Bhend J, von Storch H Consistency of observed winter precipitation trends in northern Europe with regional climate change projections. Climate Dynamics 31: Boberg F, Berg P, Thejll P, Gutowski WJ, Christensen JH Improved confidence in climate change projections of precipitation evaluated using daily statistics from the PRUDENCE ensemble. Climate Dynamics 32: Christensen JH, Boberg F, Christensen OB, Lucas-Picher P On the need for bias correction of regional climate change projections of temperature and precipitation. Geophysical Research Letters 35(L20709): doi: /2008gl Christensen JH, Carter TR, Rummukainen M, Amanatidis G Evaluating the performance and utility of regional climate models: the PRUDENCE project. Climatic Change 81(S1): 1 6. Christensen JH, Christensen OB A summary of the PRUDENCE model projections of changes in european climate during this century. Climatic Change 81(S1): Christensen O, Christensen J, Machenhauer B, Botzet M A very-high resolution regional climate simulations over scandinaviapresent climate. Journal of Climate 11: Christensen JH, Hewitson B, Busuioc A, Chen A, Gao X, Held I, Jones R, Kolli RK, Kwon W-T, Laprise R, Rueda VM, Mearns L, Menéndez C, Räisänen J, Rinke A, Sarr A, Whetton P Regional Climate Projections. Climate change 2007: The physical science basis. Contribution of working group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Technical report. Déqué M, Jones RG, Wild M, Giorgi F, Christensen JH, Hassell DC, Vidale PL, Rockel B, Jacob D, Kjellstrom E, Castro M, Kucharski F, den Hurk BV Global high resolution versus Limited Area Model climate change projections over Europe: quantifying confidence level from prudence results. Climate Dynamics 25: Déqué M, Rowell D, L uthi D, Giorgi F, Christensen J, Rockel B, Jacob D, Kjellström E, Castro M, Van den Hurk B An intercomparison of regional climate simulations for europe: assessing uncertainties in model projections. Climatic Change 81(S1):

55 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Dessai S, Lu X, Hulme M Limited sensitivity analysis of regional climate change probabilities for the 21st century. Journal of Geophysical Research 110(D19108): doi: /2005jd Doblas-Reyes FJ, Hagedorn R, Palmer TN Developments in dynamical seasonal forecasting relevant to agricultural management. Climate Research 33: Efron B, Tibshirani R An Introduction to the Bootstrap. Chapman & Hall. Fowler HJ, Ekstr om M, Blenkinsop S, Smith AP Estimating change in extreme european precipitation using a multimodel ensemble. Journal of Geophysical Research 112(D18104): doi: /2007jd Fowler HJ, Ekström M, Kilsby CG, Jones PD New estimates of future changes in extreme rainfall across the UK using regional climate model integrations. 1. Assessment of control climate. Journal of Hydrology 300: Frei C, Christensen JH, Déqué M, Jacob D, Jones RG, Vidale PL Daily precipitation statistics in regional climate models: evaluation and intercomparison of the European Alps. Journal of Geophysical Research 108: doi: /2002jd Gaertner M, Christensen O, Prego J, Polcher J, Gallardo C, Castro M The impact of deforestation on the hydrological cycle in the western Mediterranean: an ensemble study with two regional climate models. Climate Dynamics 17: Giorgi F, Bi X, Pal J Mean, interannual variability and trends in a regional climate change experiment over Europe: I. present-day climate ( ). Climate Dynamics 22: Giorgi F, Mearns LO Regional climate modeling revisited. An introduction to the special issue. Journal of Geophysical Research 104: Giorgi F, Mearns LO Calculation of average, uncertainty range, and reliability of regional climate changes from AOGCM simulations via the reliability ensemble averaging (REA) method. Journal of Climate 15: Hagemann S, Machenhauer B, Jones R, Christensen OB, Déqué M, Jacob D, Vidale PL Evaluation of water and energy budgets in regional climate models applied over Europe. Climate Dynamics 23: Haylock MR, Cawley GC, Harpham C, Wilby RL, Goodess CM Downscaling heavy precipitation over the United Kingdom: a comparison of dynamical and 45

56 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA statistical methods and their future scenarios. International Journal of Climatology 26: Haylock M, Hofstra N, Tank AMGK, Klok EJ, Jones P, New M A european daily high-resolution gridded dataset of surface temperature and precipitation for Journal of Geophysical Research 113(D20119): doi: /2008jd Hewitt CD The ENSEMBLES project: Providing ensemblebased predictions of climate changes and their impacts. EGGS Newsletter 13: Hurrell J, Loon HV Decadal variations in climate associated with the north atlantic oscillation. Climatic Change 36: Jacob D, Barring L, Christensen O, Christensen J, de Castro M, Déqué M, Giorgi F, Hagemann S, Hirschi M, Jones R, Kjellstr om E, Lenderink G, Rockel B, Sánchez E, Schär C, Seneviratne S, Somot S, Ulden AV, den Hurk BV An inter-comparison of regional climate models for europe: model performance in presentday climate. Climatic Change 81(S1): Jones PD, Reid PA Assessing future changes in extreme precipitation over britain using regional climate model integrations. International Journal of Climatology 21: Kjellström E, Ruosteenoja K Present-day and future precipitation in the baltic sea region as simulated in a suite of regional climate models. Climatic Change 81(S1): Laprise R, Caya D, Frigon A, Paquin D Current and perturbed climate as simulated by the second-generation Canadian regional climate model (CRCM-II) over northwestern North America. Climate Dynamics 21: New M, Hulme M, Jones P Representing twentieth-century space-time climate variability. Part I: development of a mean monthly terrestrial climatology. Journal of Climate 12: Noguer M, Jones R, Murphy J Sources of systematic errors in the climatology of a regional climate model over Europe. Climate Dynamics 14: Palmer T, Alessandri A, Andersen U, Cantelaube P, Davey M, Delecluse P, Déqué M, Diez E, Doblas-Reyes FJ, Feddersen H, Graham R, Gualdi S, Gueremy J-F, Hagedorn R, Hoshen M, Keenlyside N, Latif M, Lazar A, Maisonnave E, Marletto V, Morse AP, Orfila B, Rogel P, Terres J-M, Thomson MC Development of a european multi-model ensemble system for seasonal to inter-annual prediction 46

57 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA (DEMETER). Bulletin of the American Meteorological Society 85: Perkins S, Pitman AJ, Holbrook NJ, McAneney J Evaluation of the AR4 climate models simulated daily maximum temperature, minimum temperature, and precipitation over australia using probability density functions. Journal of Climate 20: Räisänen J, Hansson U, Ullerstig A, Döscher R, Graham L, Jones C, Meier H, Samuelson P, Willen U European climate in the late twenty-first century: regional simulations with two driving global models and two forcing scenarios. Climate Dynamics 22: Rummukainen M, Räisänen J, Bringfelt B, Ullerstig A, Omstedt A, Willén U, Hansson U, Jones C A regional climate model for northern Europe: model description and results from the downscaling of two GCM control simulations. Climate Dynamics 17: Sánchez E, Gaertner MA, Gallardo C, Padorno E, Arribas A, Castro M Impacts of a change in vegetation description on simulated european summer present-day and future climates. Climate Dynamics 29: Shukla J, DelSole T, Fennessy M, Kinter J, Paolino D Climate model fidelity and projections of climate change. Geophysical Research Letters 33(L07702): doi: /2005gl Tapiador FJ, Sánchez E Changes in the european precipitation climatologies as derived by an ensemble of regional models. Journal of Climate 21: Tapiador FJ, Sánchez E, Gaertner MA Regional changes in precipitation in europe under an increased greenhouse emissions scenario. Geophysical Research Letters 34(L06701): doi: /2006gl Tapiador FJ, Sánchez E, Romera R Exploiting an ensemble of regional climate models to provide robust estimates of projected changes in monthly temperature and precipitation probabilistic distribution functions. Tellus A 61A: Tebaldi C, Mearns LO, Nychka D, Smith RL Regional probabilities of precipitation change: A bayesian analysis of multimodel simulations. Geophysical Research Letters 31(L24213): doi: /2004gl Uppala S, Kallberg P, Simmons A, Andrae U, da Costa Bechtold V, Fiorino M, Gibson J, Haseler J, Hernandez A, Kelly G, Li X, Onogi K, Saarinen S, Sokka N, Allan R, Andersson E, Arpe K, Balmaseda M, Beljaars A, Van de Berg L, Bidlot J, Bormann N, 47

58 3PROPUESTA DE CONJUNTO PONDERADO DE MODELOS REGIONALES DE CLIMA Caires S, Chevallier F, Dethof A, Dragosavac M, Fisher M, Fuentes M, Hagemann S, Hlm E, Hoskins B, Isaksen L, Janssen P, Jenne R, McNally A, Mahfouf J-F, Morcrette J-J, Rayner N, Saunders R, Simon P, Sterl A, Trenberth K, Untch A, Vasiljevic D, Viterbo P, Woollen J The ERA-40 re-analysis. Quarterly Journal of the Royal Meteorological Society 131: , doi: /qj Vidale PL, Luthi D, Frei C, Seneviratne S, Schar C Predictability and uncertainty in a regional climate model. Journal of Geophysical Research 108: doi: /2002jd

59 4 EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Este capítulo reproduce el texto del siguiente manuscrito: Domínguez M, Romera R, Sánchez E, Fita L, Fernández J, Jiménez-Guerrero P, Montávez JP, Cabos WD, Liguori G, Gaertner MA (2013) Present-climate precipitation and temperature extremes over Spain from a set of high resolucion RCMs. Clim Res 58: , doi: /cr01186

60

61 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 4.1 Resumen En el marco del proyecto español ESCENA, se han analizado los extremos de precipitación y temperatura sobre España obtenidos a partir de la simulaciones de 5 modelos regionales de clima. La simulación de los extremos de precipitación y temperatura supone un reto para los modelos dada la complejidad del clima en el dominio analizado. Hay diferencias importantes respecto de estudios similares previos (como los proyectos Europeos PRUDENCE y ENSEMBLES): el dominio de la simulación de clima presente ( ) está centrado sobre la PI y las simulaciones están anidadas en el reanálisis ERA-Interim. Dos de los modelos (WRF y MM5) no formaron parte del proyecto ENSEMBLES. Para validar los resultados se ha usado una nueva base de datos (Spain02) de temperatura y precipitación diarias de alta resolución (0.2ºx0.2º) sobre España. El comportamiento de los modelos depende de las características climáticas de las subregiones. Los resultados son mejores en la costa norte de España, mientras que en otras cuencas atlánticas hay algunas diferencias que no pueden relacionarse con las limitaciones en la representación de los episodios convectivos. Algunos modelos subestiman la cantidad de días con precipitación intensa. Este aspecto en general no mejora comparándolo con los resultados de ENSEMBLES, a pesar de las diferencias de configuración. En otros aspectos, todos los modelos están en un nivel similar al del mejor subgrupo de modelos de ENSEMBLES. Las simulaciones de ESCENA (que cubre más escenarios de emisión y más modelos globales de clima que el proyecto (ENSEMBLES) pueden ser consideradas un complemento valioso a los resultados del proyecto europeo para estudios de impacto en España. 51

62 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 52

63 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 4.2 Present-climate precipitation and temperature extremes over Spain from a set of high resolution RCMs Abstract In the frame of the Spanish project ESCENA, 5 regional climate model simulations were analyzed in terms of their representation of precipitation and temperature extremes over Spain. The climate complexity of the target domain poses a challenge for the models in terms of their ability to simulate precipitation and temperature extremes. There are important differences in comparison to previous similar studies (such as PRUDENCE and ENSEMBLES European projects): the domain of the present-climate simulations (1990 to 2007) is centered over the Iberian Peninsula (IP), and the simulations are nested in ERA-Interim reanalysis. Two of the models (WRF and MM5) were not part of the ENSEMBLES project. A new high-resolution ( ) database (Spain02) of daily temperature and precipitation observations over Spain was used to validate the results. The performance of the models depends on the climatic characteristics of the subregions. Results are better for the northern coastal area of Spain, but for other Atlantic basins there are some biases that cannot be linked to limitations in the representation of very heavy convective events. Several models underestimate the amount of rain for heavy-precipitation days. This aspect has not improved in general in comparison to ENSEMBLES results, despite the differences in setup. In other respects, all models are at a similar level as the best subgroup of ENSEMBLES models. The ESCENA simulations (covering more emissions scenarios and GCMs than ENSEMBLES project) can thus be considered a valuable complement of that European project for impact studies over Spain. Keywords: Climate extremes Extreme precipitation Extreme temperatures Regional climate models Iberian Peninsula Heavy precipitation Droughts Heat waves 53

64 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 1. Introduction The analysis of extreme events under the present climate is important in order to assess the reliability of climate change projections, because it is likely that the frequency and intensity of extreme events and the size of affected areas will increase in the future (Sánchez et al. 2011, Field et al. 2012). Beniston et al. (2007) defines an extreme event as rare, intense and severe. Thus, extreme events are usually analyzed in terms of frequency distribution tails of climate variables, including high percentiles and record-breaking events, known to have caused severe impacts. A global tendency towards an increased frequency of some extreme events during the 20th century has been found in several studies, but the results show large spatial variability (Alexander et al. 2006, Vincent & Mekis 2006, López-Moreno & Beniston 2009, Hirschi et al. 2010). The socio-economic impact of extreme temperature is not only related to hot or cold individual days, but also to persistent events like cold and heat waves (Fischer & Schär 2010). More frequent and intense summer heat waves have been observed at global and European scales (Alexander et al. 2006, DellaMarta et al. 2007, Barriopedro et al. 2011). To understand the mechanisms behind extreme events in present and future climate, global circulation models (GCMs) and regional climate models (RCMs) have been used. The increased frequency of observed extreme events has been studied at different spatial scales, from global (Frich et al. 2002), to continental (Beniston & Stephenson 2004, Schär et al. 2004) and down to regional scale (Huth et al. 2000). On this respect, the Intergovernmental Panel on Climate Change (IPCC 2001, 2007) recommended more adequate analyses and regional detail for model projections. At the same time, end-users, such as impact modellers, demand climate data at higher spatial resolution than given by GCMs (Castro et al. 2007). There is, thus, a need for high resolution climate projections. Additionally, model uncertainties are explored by means of multi-model ensembles of simulations. The generation of high-resolution regional climate information through an ensemble of RCM simulations has been the basis for several international projects (e.g. PRUDENCE, ENSEMBLES in Europe, NARCCAP in North America). Currently, such an approach is being extended to most continental regions of the world in the frame of CORDEX project (COordinated Regional climate Downscaling EXperiment; Giorgi et al. 2009). Among other advantages, the estimation of the spatial and temporal distribution of extreme precipitation episodes has been improved using climate models with high resolution (Beniston 2006, Frei et al. 2006, Beniston et al. 2007) compared to those using lower resolutions. 54

65 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA In the Mediterranean region, extreme events have strong impacts on ecosystems, socio-economic activities and agriculture (Vicente-Serrano et al. 2004, Lana et al. 2006, Vicente-Serrano 2006a). Results from RCMs including the Iberian Peninsula (IP) in their domains (Sánchez et al. 2004, 2011, López-Moreno & Beniston 2009, Herrera et al. 2010) highlight the importance of high spatio-temporal resolution for the study of extreme events over this area. The irregularity of precipitation and the strong seasonal variations in the IP make this region parti cularly interesting for the study of precipitation and temperature extremes, which caused considerable damage over the last century (García-Herrera et al. 2005, Vicente-Serrano 2006b). Instead of considering the IP as a single region as in previous studies performed in the frame of PRUDENCE and ENSEMBLES European Projects (e.g. Boberg et al. 2009, 2010, Fischer & Schär 2010, Kjellström et al. 2010), in the present study the analyzed domain has been divided in 6 subregions to take into account the high spatial variability of the IP climate. A correct simulation of present climate is needed to detect changes in key variables that could modify the availability of water resources (MartinVide 2004, López-Moreno & Beniston 2009). In previous modelling studies over Europe, RCM simulations were compared with the gridded observational climate database E-OBS (Haylock et al. 2008). E-OBS has been very useful in many analyses of temperature extremes (Fischer et al. 2007, Hirschi et al. 2010). However, due to the high spatial variability of the climate in the IP, the amount of stations underlying E-OBS is insufficient for this region. High resolution climate data (Brunetti et al. 2001, González-Hidalgo et al. 2003) and gridded data based on much denser station networks than E-OBS are needed for RCM evaluation studies (Kysely & Plavcova 2010). Here we use the gridded daily precipitation and temperature dataset Spain02 (see section 2.III), developed for Peninsular Spain (PS) and the Balearic Islands (BA) (Herrera et al. 2012). Spain02 was tested (Herrera et al. 2012) against point observations and compared to E-OBS database. This new database (Spain02) proved to represent better the intensity and spatial variability of several standard extreme climate indices defined by Sillmann & Roeckner (2008). The present study analyzes the results for present climate from 5 RCM simulations, focusing on their ability to represent extreme events. The RCM simulations have been performed within the Spanish project ESCENA, and complement and extend ENSEMBLES results (using the same horizontal resolution of 25 km) through the use of improved and/or additional RCMs, new GCM/RCM matrix combinations, and a larger set of emission scenarios (A1B, A2 and B1). Results of ESCENA project are publicly available for impact or regional climate studies ( 55

66 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA proyectoescena.uclm.es). An important difference with respect to PRUDENCE and ENSEMBLES is the simulation domain. It is centered over the IP and covers larger parts of the Atlantic Ocean, including the Canary Islands, which were not included in these European projects. Additionally, evaluation model runs are forced by the ERAInterim reanalysis instead of the ERA-40 reanalysis used in ENSEMBLES. We focused this study on PS and the BA, where the high-resolution Spain02 database is available. 2. Methods and data 2.I. Regional climate models The RCM ensemble includes 5 different models (PROMES, 2 versions of WRF model, MM5 and REMO). These models have been developed at different institutions. A summary of the physics parameterizations for each of them is shown in Table 4.1. These models are described in more detail in Jiménez-Guerrero et al. (2013). The RCM PROMES (Castro et al. 1993, Sánchez et al. 2004) was developed at the Complutense University of Madrid and the University of Castilla-La Mancha (Spain). Several changes have been introduced in PROMES with respect to previous simulations. The most significant changes are the coupling to ORCHIDEE land surface model (Krinner et al. 2005) and the change of radiation parameterization, as explained in Domínguez et al. (2010). The Weather Research and Forecasting (WRF) model is a state-of-theart limited area model developed in collaboration between the National Center for Atmospheric Research (NCAR) and a number of research institutions in the US. In this study we used the Advanced Research WRF core (version 3.1.1; Skamarock et al. 2008), which is the research version of the model. WRF model simulations were run by the Santander Meteorology Group of the University of Cantabria, Spain. The model includes modifications introduced by this group to get averaged and extreme values of surface variables (Fita et al. 2010) and was run through the WRF4G execution workflow (Fernández-Quiruelas et al. 2010). Two different configurations were used in the simulations (labeled as WRF-A and WRF-B in Table 4.1). They only differ in the Planetary Boundary Layer (PBL) scheme. WRF-A uses a local scheme, whereas WRF-B uses a non-local scheme. MM5-UM is the climate version of the fifthgeneration Pennsylvania State University NCAR mesoscale model (Dudhia 1993, Grell et al. 1994, Gómez-Navarro et al. 2010). The physical configuration (Table 4.1) has been chosen in order to minimise the computational cost, since none of the tested configurations provides the best performance for all kinds of synoptic events and regions of the IP (Fernández et al. 56

67 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 2007). REMO is a hydrostatic, 3-dimensional regional climate atmospheric model, developed at the Max-Planck-Institute for Meteorology in Hamburg. It is based on the Europa Model, a former numerical weather prediction model of the German Weather Service and it is described in Jacob (2001) and Jacob et al. (2001). REMO uses the physical package of the global circulation model ECHAM4 (Roeckner et al. 1996). All models are configured with a relaxation boundary of 8 to 10 grid points. 2.II. Simulation set up The RCM simulations cover part of Europe, with a domain centered on the IP (Figure 4.1 showing the land mask and orography for each model) and km of horizontal resolution in Lambert Conformal projection (PROMES, both versions of WRF and MM5) or in a rotated latitude longitude coordinate system (REMO). The topography has been interpolated from GTOPO30 data (Verdin & Greenlee 1996). The horizontal coordinates are arranged in Arakawa-B (MM5) or Arakawa-C (PROMES, REMO and both versions of WRF) grids. PROMES model used 37 sigma levels in the vertical, WRF 33, MM5 30 and REMO 31 levels (up to 10 hpa for all models). The simulations cover the period from 1990 to 2007, with 1 yr of spin-up. The ERA Interim reanalysis derived from the latest version of the operational ECMWF system has several differences to ERA-40 (Uppala et al. 2005), such as improvements in model physics, new humidity analysis, variational bias correction of satellite radiance data, among others (Dee & Uppala 2009). In this study all models used boundary conditions from ERA-Interim with a spatial resolution of , updated every 6 h. 57

68 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Model Micrphysics PROMES Cumulus Radiation PBL Land surface Includes ice processes (Hong Kain & Kritsch (1993) et al., 2004) ECMWF (2004) with fractional cloud cover (Chaboreau & Bechtold, 2002, 2005) Cuxart et al. (2000) ORCHIDEE LSM (Krinner et al., 2005) WRF-A Single- moment 5-class microphysics (Hong & Lim, 2006) Grell & Devenyi (2002) CAM 3.0 (Collins et al., 2006) Local Mellor-Yamada- NOAH LSM (Chen & Janjić (Janjić, 1990, Dudhia, 2001a, b) 1994) WRF-B Single- moment 5-class microphysics (Hong & Lim, 2006) Grell & Devenyi (2002) CAM 3.0 (Collins et al., 2006) Non-local Yonsei University (Hong et al., 2006) MM5 Simple ice (Dudhia, 1989) Grell (1993) Rapid radiative transfer Medium-range NOAH LSM (Chen & model (RRTM) (Mlawer forecast (MFR) (Hong Dudhia, 2001a, b) et al., 1997) & Pan, 1996) REMO Stratiform clouds (Sundqvist, Mass flux convection Morcrette et al. (1986) Higher-order closure 1978), (Roeckner et al., 1996) scheme after Tiedke with modifications after scheme (Brinkop & (1989) with modifications Giorgetta & Wild (1996) Roeckner, 1995) after Nordeng (1994) NOAH LSM (Chen & Dudhia, 2001a, b) Types of vegetation (FAO). Bucket scheme for hydrology. Five layers for thermal processes Table 4.1. Physics parameterizations used by the models in the simulations analyzed in the study. PBL: planetary boundary layer, LSM: Land Surface Model, CAM: Community Atmosphere Model. 58

69 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Figure 4.1. Simulation domain of the 5 models used in the study: MM5, PROMES, REMO, WRF-A and -B. 2.III. Spain02 observation database The Spain02 precipitation and temperature database is a daily gridded dataset developed by Herrera et al. (2012) using surface station data from a set of 2756 quality-controlled stations over PS and the BA. There are currently insufficient surface observations to include the Canary Islands in the database. Here we present the most important features of this observational data. This data set spans from 1950 to 2008, with daily frequency and resolution. The fields were interpolated in a 2-step process. First, the monthly means were interpolated using thin-plate splines. Second, daily departures from the monthly means were interpolated using a kriging methodology. In the case of precipitation, occurrence and amounts are interpolated through indicator and ordinary kriging, respectively. The interpolation procedure for precipitation is similar to that of the E-OBS European database (Haylock et al. 2008), except for the absence of the elevation-dependent splines applied to E-OBS database, which was not used in Spain02 (the topography is well represented by the large amount of stations). 59

70 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA For temperature, the Spain02 database applied a more stringent filter to the station data in order to provide a product better suited for trend analyses. Unlike the precipitation product, where the maximum number of available stations at a given day were used, the temperature data set was built from a reduced set of stations with the longest records. In particular, only those stations with a record longer than 40 yr and <1% of missing data were included in the interpolation. This led to a set of 186 stations. These stations were gridded using the same methodology as E-OBS (including the use of elevation as covariable, given the smaller amount of stations). The differences to E-OBS are the better quality and larger amount of stations. The large number of stations used in this product com pared with E-OBS database, provides a better representation of the precipitation (Herrera et al. 2012) and temperature variability over Spain. Details of the interpolation procedure can be found in Herrera (2011) and Herrera et al. (2012). 2.IV. Statistical analysis The analyzed variables present a high spatial variability in a relatively small area, especially precipitation (Herrera et al. 2010). In the study of the climate features of a given region, information provided by the frequency distribution is complementary to simpler statistical analyses from which an integrated representation of climatic conditions over a region is provided (Giorgi et al. 2004a,b, Christensen & Christensen 2007, Déqué et al. 2007, Jacob et al. 2007). Additionally, the use of subregions provides more useful results for some impact studies. Thus, our precipitation analysis was performed using the upper tail of the precipitation distribution for the subregions, and by examining the values of some standard extreme climate indices. 60

71 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Figure 4.2. Spain02 database orography and subregions: North (NO, 178 grid points), Center-South (CESU, 756), Levant (LE, 183), Ebro (EB, 234), Cataluña (CA, 57) and Balearic Island (BA, 37), The probability density function (PDF) for precipitation was approximated by a histogram of the relative frequencies of daily precipitation amounts (days with precipitation 1 mm) binned into 1 mm d 1 bins. A clearer visualization of the highest precipitation values is provided using a logarithmic scale. The PDFs are normalized by dividing by the total amount of precipitation data (1 value per day and grid point) in order to take into account the different size of the subregions. Comparison of the simulated results with observed values was performed using the PDFs calculated directly at the original grids for each subregion. Due to the similar resolution of both sources ( for Spain02 and km for RCMs), the differences in subregion limits and extension are rather small among the different grids. This avoids the smoothing of extreme values which may arise from regridding. In order to take into account the different precipitation regimes, the PDFs and the related analyses were carried out for different subregions obtained by grouping the 11 basins described in Herrera et al. (2010) according to the similarity of their annual cycles. In Figure 4.2 these subregions are shown for the Spain02 grid; for the RCMs, equivalent subregions in the original grid of each model have been used. The 6 subregions are North (NO), Center-South (CE-SU), Levant (LE), Ebro (EB), Cataluña 61

72 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA (CA) and the BA. For the purpose of quantifying the overlap of the upper tail of the precipitation PDF between every RCM and Spain02, we applied a skill score (Perkins et al. 2007), previously used by Boberg et al. (2009), Boberg et al. (2010) and Kjellström et al. (2010). We introduced a small variation in Perkins score, computing only the upper tail of the PDF, defined as the truncated PDF obtained using events larger than the 95th percentile of the observed precipitation. This metric has been calculated on all grid points and results are given as spatial averages for each subregion. It takes values between 0 to 1, where 1 indicates maximum overlap. Extreme temperatures are represented selecting different percentiles of daily maximum and minimum temperature depending on the season: the 10th percentile in winter (December to February) and the 90th percentile in summer (June to August). Moreover, in order to analyse extreme events in a complementary way, we used several indices (Klein Tank et al. 2009) applied over the IP by authors like de la Cruz Gallego et al. (2004), Sánchez et al. (2004), Gallego et al. (2005) and Herrera et al. (2010), among others. The selected indices are indicated in Table 4.2. For precipitation, we selected medium-high intensity indices representing droughts or high intensity precipitation episodes. As frequency indices, we took the consecutive dry days (<1 mm) index (CDD) and the very heavy precipitation (>20 mm) days index (R20mm). As intensity index, we chose the fraction of precipitation due to events above the 95th percentile (R95PTOT). We calculated some other indices, but their spatial distribution was similar to the indices shown. All indices (see Table 4.2) are presented as yearly averages except R95PTOT. 62

73 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Index Description Units CDD Consecutive dry days index (yearly mean) d yr-1 R20MM Very heavy precipitation days index (yearly mean) d yr-1 R95PTOT Precipitation fraction dur to 95 percentile % HWd Maximum hear wave duration per year (yearly mean) d yr-1 HWn Heat wave number index (yearly mean) Heat waves yr-1 TR Tropical nights index (Tn>20ºC) d yr-1 CHT Combined HW and TR (Tn>20ºC and Tx>35ºC) d yr-1 Precipitation Temperature Table 4.2. Precipitation and temperature extreme indices. Tx(Tn): maximum (minimum) temperature. For temperature, we used the yearly number of heat waves (HWn), the yearly maximum heat wave duration (HWd) and a set of temperature indices. A heat wave is defined as an event with maximum temperatures >35 C during at least 5 consecutive days. This threshold has been calculated by averaging over the whole domain the 95th percentile of Spain02 climatology (1961 to 1990). These indices are more sensitive to model biases, since they are defined by absolute values rather than relative ones. The complementary temperature indices include the tropical nights index (TR) (number of days with minimum temperature >20 C) and the combined tropical nights and summer days index (CHT, number of days with minimum temperature >20 C and maximum temperature >35 C; Fischer & Schär 2010). The spatial representation of extreme indices is analysed with the Pearson correlation for the whole PS and BA. Regarding ensemble values, the ensemble mean of the models indices was calculated for each statistical index. In contrast, the PDF ensemble was created using all daily precipitation data from all the models without averaging. 63

74 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 3. Results 3.I. Precipitation 3.I.a. PDF tail analysis Figure 4.3 shows the simulated precipitation PDF for the different RCMs, the multimodel ensemble and the observed Spain02 data for the 6 subregions. The large spatial variability of precipitation over the IP is clearly seen from the observed precipitation PDFs. Differences are relatively high: CE-SU subregion shows comparatively low frequency of the most extreme values (>100 mm d 1), whereas LE, CA and BA show a relatively high frequency of such values. This reflects the division between Atlantic and Mediterranean precipitation regimes. Figure 4.3. Precipitation PDF of Spain02 data (black dots) and the 5 RCMs: PROMES (red), WRF-A (dark blue), WRF-B (light blue), MM5 (pink) and REMO (green), and the ensemble mean (ENS, yellow) over 6 Spanish subregions (North: NO, Center-Soutch: CE-SU, Ebro: EB, Levant: LE, Cataluña: CA, Balearic Islands: BA) from January 1990 to December Gray lines: 95th percentile (p95) for each region. The spread of the simulated PDFs is highly dependent on the considered subregion. The lowest spread of the simulations is seen in the NO subregion. At the same time, differences between simulated and observed values are the lowest there, basically due to the small temporal variability in the area. Atlantic frontal systems are very frequent here. The precipitation associated with these systems is in general well represented by RCMs (Herrera et al. 2010). By contrast, Mediterranean subregions have a marked seasonal precipitation regime with high spatio-temporal variability. Extreme precipitation is quite often related to severe convective events which are more difficult to simulate by RCMs (e.g. Herrera et al. 2010). This result in a high spread of simulated precipitation for the CE-SU subregion. This area includes At 64

75 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA lantic basins, where precipitation is typically linked to Atlantic depressions; however, the seasonality of precipitation is much higher than for the NO subregion. Most models tend to underestimate precipitation for the CE-SU subregion, as found in previous studies (e.g. Herrera et al. 2010). Figure 4.3 also shows the 95th percentile of Spain02 data, used as the threshold for the modified Perkins score. Subregions like BA or CA have the highest 95th percentile values (~30 mm d 1), while for the rest of the domain it most often does not exceed 25 mm d 1. Table 4.3 shows the calculated skill scores values. The values at the NO and CE-SU subregions are rather high ( 0.90). The spread is smaller there than for the eastern part of PS. Maximum differences of about 25% are found at the BA. Subregions with higher spreads have lower scores. The RCMs present in general a good agreement with the observed PDF, with score values between 0.75 and 0.95, except for the BA that have the lowest score (<0.74 for all models). These lower values could be related to the small number of grid points representing the islands, and the consequent worse representation of orography in this subregion. The ensemble scores are consistently higher than the scores obtained for the individual models. NO CE-SU LE EB CA BA PROMES WRF-A WRF-B MM REMO ENS Table 4.3. Results for the modified Perkins' skills score (see section 2.IV) of extreme precipitation (higher than p95 of Spain02) for the 6 Spanish subregions (North: NO, Center-South: CE-SU, Levant: LE, Ebro: EB, Cataluña: CA, Balearic Island: BA), the 5 RCMs (PROMES, WRF-A, WRF-B, MM5 and REMO) and the ensemble (ENS) mean. 65

76 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 3.I.b. Extreme precipitation indices Regarding dry spells, the CDD index (Figure 4.4, left column) has a spatial distribution with a strong north south gradient. Center and South PS show the highest observed values, with maximum values >100 d yr 1. This reflects the summer dry period. Differences among RCMs are not high. Results show generally an underestimation of observed values over most of the domain, except some models (WRFA, WRF-B and REMO) in parts of the Mediterranean subregions (LE, CA and BA). RCMs and observations differ considerably in some areas, particularly near the Mediterranean southern coast, where for example PROMES or MM5 simulate values of ~40% of the observed CDD. However, spatial correlations of ~0.90 for all models indicate a good concordance between observed and simulated spatial distributions of this index (see Figure 4.7). The ensemble mean only gives a better value of the index in comparison with some models in certain subregions. 66

77 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Figure 4.4. Spatial distribution of precipitation extreme indices (CDD: consecutive dry days index, R20MM: very heave precipitation days index, and R95PTOT: precipitation fraction due to 95 th percentile) of Spain02 data (1st row) and the RCM biases for PROMES, WRF-A, WRF-B, MM5, REMO and the ensemble mean (rows 2-7, respectively) from January 1990 to December

78 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA The observed spatial distribution for R20MM index and the corresponding model biases are shown in the middle column of Figure 4.4. The highest values are found in the northern part of PS, in coherence with its precipitation regime. The next highest values are observed in mountainous areas of the CE-SU subregion, followed by CA subregion. The spatial distribution of this index shows no clear relationship to the distribution of CDD index, as high numbers of days >20 mm occur both for low and high CDD areas. This illustrates well the large differences of rain regimes in this small domain. The smallest differences with respect to observations are obtained by PROMES, whereas the other RCMs underestimate this index, especially both versions of WRF model (spatial correlation values vary from 0.72 [WRF-A] to 0.79 [WRF-B]). The R95PTOT index (Figure 4.4, right column) is re lated to the heaviest precipitation events. Observed values obtained from Spain02 data show a clear maximum for the Mediterranean coastal areas. The RCMs are able to reproduce the observed climatic contrast between areas with high and low values of the heaviest rain events. However, they slightly overestimate this index over most of the do main. PROMES and MM5 show the largest positive bias extending over the widest areas. WRF-A values are the closest of all models to the observed data over most of the domain. The R95PTOT spatial correlations (see Figure 4.7) are the lowest among all precipitation in di ces. These low R95PTOT correlation values could indicate the difficulty of simulating the precise spatial location of heavy convective events using RCMs. 68

79 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Figure 4.5. Spatial distribution for temperature seasonal percentiles: Tx10 (Tn10) = 10th percentile for maximum (minimum) temperature for winter; Tx90 (Tn90) = 90 th percentile for maximum (minimum) temperature for summer. Top row: Spain02 data. Rows 2-7: RCM biases for PROMES, WRF-A, WRF-B, MM5, REMO and the ensemble mean. 69

80 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 3.II. Maximum and minimum temperature 3.II.a. Seasonal percentiles Figure 4.5 presents the seasonal percentiles for maximum (Tx) and minimum (Tn) temperature. The first and second columns show the 10th percentile respectively of maximum (Tx10) and minimum (Tn10) temperatures in winter. Spain02 shows the lowest values of Tx10 in the northern part of PS (reaching 3 C in the Pyrenees) while the lowest values (down to 10 C) for Tn10 are found in the central-eastern highlands around the Iberian Mountain Range. This reflects the continental characteristics of the interior climate (Castro et al. 2007). The 90th percentile for maximum (Tx90) and minimum (Tn90) temperatures in summer are represented in the third and fourth columns of Figure 4.5. The highest values of Tx90 (~40 C) appear in the CE-SU subregion of the IP, related to the typical summer thermal lows over this area (Hoinka & de Castro 2003). The thermal inertia of the oceans leads to smaller differences between Tx90 and Tn90 in coastal areas (with values between 4 and 14 C in NO, LE and CA subregions, not shown) and to higher differences in continental areas (with values between 14 and 24 C in CE-SU and EB subregions). In general, a similar spatial distributionwas obtained for all models, with alow spread among them, except for REMO Tn percentiles. The spatial distribution of the Spain02 percentiles was generally well reproduced by the models. The highest correlation was found for Tx (see Figure 4.7). Correlation values ob tained for percentiles of Tx are0.83 to 0.95,and 0.79 to 0.86 for Tn percentiles. REMO has the lowest values for cold nights (0.79). Most of the RCMs underestimate the Tx10 values, with moderate differences in some areas. REMO is the model with values closer to Spain02. Both versions of WRF and MM5 also underestimate the Tx90, while PROMES matches Spain02 percentiles in the western part of PS and overestimates them, particularly on the Mediterranean coast and the BA. Regarding minimum temperature percentiles, the values obtained by REMO are higher than the simulated by the rest of models, and also higher than the observations in most of the domain, as discussed below. These biases of REMO reach up to 10 C in areas where these percentiles reach their minimum values. 70

81 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Figure 4.6. Spatial distribution for yearly heat wave indices: HWn (heat wave frequency), HWd (heat wave duration), tropical nights (TR), combined tropical and summet days (CHT). Top row: Spain02, rows 2-7: RCM biases for PROMES, WRF-A, WRF-B, MM5, REMO and the ensemble mean. 71

82 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 3.II.b. Extreme temperature indices Figure 4.6 presents the indices that have been used to describe temperature extreme events like heat waves (defined in Table 4.2). Spain02 shows that the highest heat wave frequency (HWn; first column) is located in the CE-SU subregion (up to 5 heat waves yr 1), with a maximum duration (HWd; second column) that varies between 6 and 16 d in PS. It should be noted that these indices are sensitive to model biases as they are related to a fixed observational threshold. The spread of HWn values is large among the RCMs. There is some tendency of the models to underestimate this index compared with observations, with the exception of PROMES and REMO, which overestimate the number of heat waves particularly in southern and eastern coastal areas. The models which underestimate the number of heat waves also tend to underestimate their duration. In contrast, PROMES and REMO simulate longer than observed heat waves, especially over the southwestern part of PS. The spatial distribution of these indices presents correlation values between 0.43 (HWd for WRF-A) and 0.88 (HWn for MM5) (Figure 4.7); HWd is the index with the lowest spatial correlation values for all models. Figure 4.6 also shows other indices related to factors that have a social and economical impact on the studied area (AEMET 2007, Fischer & Schär 2010): TR (third column) and CHT (last column), defined in Table 4.2. Spain02 present the highest TR value in the eastern coasts and the BA, with >100 d yr 1. Values around 20 or 30 d yr 1 are observed in most of the southern half of PS and the Ebro valley, while in the remaining areas TR has a value of <2 d yr 1. The thresholds used to compute CHT are the same as those used for TR and HW, but the spatial distribution of the CHT index is more similar to HWn than to the TR index, with highest values in CE-SU subregion (up to 40 d yr 1). This fact highlights the need for studying the TR index separately in extreme temperature analysis. For instance, in the case of the BA, the value of TR is >100 d yr 1, while CHT is <15 d yr 1. 72

83 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Figure 4.7. Spatial correlations between simulated and observed precipitation and temperature indices of the 5 RCMs (PROMES, WRF-A, WRF-B, MM5 and REMO) and the ensemble mean, and the mean correlation of all indices for each model (last column). Index definitions in Table 4.2 and Figure 4.5. Differences of TR values among RCMs are related to the differences obtained for Tn90 index (see last column in Figure 4.5), with higher values for the BA and the southern part of PS. In general, all the models underestimate TR in the central area of PS except REMO, which overestimates this index in most of the domain. CHT is also clearly overestimated by REMO in many parts of PS, reaching values up to 3 times higher than the observed ones. If we consider the biases of maximum and minimum temperature percentiles (Figure 4.5), this latter bias is probably more related to the minimum temperature biases. Higher than observed values are also simulated by PROMES in the southern and eastern coasts. This positive bias for PROMES is also noticeable in the BA, which is in agreement with Tx90 percentile (Figure 4.5) and HWn and HWd indices (Figure 4.6). All models except REMO underestimate CHT for the area with the highest observed values of this index. Spatial correlations for both indices vary between 0.60 and The ensemble mean of the CHT index does not present an important improvement compared to the results of the individual models. 73

84 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 4. Discussion and conclusions We have analyzed the ability of 4 RCMs (WRF, MM5, PROMES and REMO) to reproduce climate extremes in an evaluation simulation over PS and the BA, for the period This work is part of a Spanish project (ESCENA) for generating a set of regional climate change scenarios over this region. In comparison to previous climate validation studies over this domain, there are important differences: the domain is centered over the IP, unlike the domain used in ENSEMBLES project in which the lateral boundaries were very near to this region, and the models are nested in the ERA-Interim reanalysis. The multi-model ensemble also includes WRF (WRF-A and -B, with different PBL schemes) and MM5, which were not included in ENSEMBLES. There is a very large range of precipitation regimes in this small analyzed region. Three areas stand out for at least one of the extreme indices: the northern coast, the southwestern area and the Mediterranean coast. The northern coast is an area with low CDD and high R20MM values. Precipitation occurs throughout the year, and the interannual variability (shown in the companion paper, Jiménez-Guerrero et al. 2013) is rather low. For this regime, the spread of the model results is the lowest, as indicated by the Perkins skill score (Table 4.3). CDD and R95PTOT are rather well reproduced by all models, but they clearly differ in the R20MM index, for which several models show an important negative bias. By contrast, the spread of the results for the southwestern area is higher, and there is a clear tendency of most models to underestimate precipitation. This occurs despite the fact that the PDF shows comparatively low frequency for very high precipitation values, and therefore model errors cannot be linked mainly to problems in simulating very heavy convective precipitation events. This is an area where model biases in a previous study with ENSEMBLES project results (Herrera et al. 2010) were also large. A possible reason for those biases in ENSEMBLES simulations could have been the proximity of lateral boundaries, or limitations of moisture values of ERA40 re analysis. ERA-Interim, among other improvements, has introduced a new humidity analysis. But the present setup, with western lateral boundaries far away from this region and nesting in ERA-Interim, does not improve previous results. We can obtain more information about biases in this area from the companion paper (JiménezGuerrero et al. 2013). As shown there, the interannual variability is high and is not well captured by most models. But at the same time, results shown in this companion paper indicate that all models show high temporal correlations. This suggests that the timing of precipitation events is well captured (a result that could be linked to the fact 74

85 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA that major precipitation events are frequently due to large scale Atlantic depressions), but the precipitation amounts are not. The fact that there is one model (PROMES) showing better results over this area indicates that the negative bias could be associated with some aspect of the model formulation. For the Mediterranean coast, the number of days with >20 mm of precipitation is smaller than for the southwestern part, but R95PTOT is much larger, and the PDF tails show a large frequency of very high precipitation. A similar negative bias in R20MM can be seen for all models except PROMES, but R95PTOT is rather well captured by all models. In contrast to the southwestern area, the temporal correlation (Jiménez-Guerrero et al. 2013) is much lower for all the models here. This indicates that the errors for the Mediterranean coast are also linked to deficiencies in the simulation of the temporal distribution of precipitation. An interesting result, specific to the Mediterranean subregions, is that the ensemble mean is worse than the individual models in representing the frequency of very high precipitation events. The ensemble mean underestimates this frequency. A possible reason is that strong convective events are spatially localized, and the specific locations are simulated differently by the models. Thus, the average would tend to smooth out these very high precipitation amounts. Modelled values of maximum temperature percentiles are generally lower than observed, with the exception of REMO. By contrast, this model shows a considerable overestimation of minimum temperature percentiles and of TR and CHT indices, indicating a problem with minimum temperatures. The REMO version used here already includes the improvements discussed in Kjellström et al. (2010) for winter minimum temperatures. The present biases in REMO results must have different causes, which are still unknown. The spatial distribution of temperature percentiles is particularly well simulated by all models, as illustrated by the high spatial correlation values. ESCENA also explored a very limited part of the uncertainty accounted for by the physical parameterizations using the WRF model in 2 configurations. The differences between the local and non-local PBL closure schemes used are most noticeable in the minimum temperature indices, since the local closure configuration (WRF-A) shows a weaker low level mixing during nighttime (García-Díez et al. 2013), which leads to cooler minimum temperatures (and associated percentiles). WRF-B used a non-local closure scheme, very similar to that used by MM5, leading to very similar biases in the winter cold extremes (Figure 4.5, Tn10). 75

86 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA The present multi-model ensemble can be compared to the one of ENSEMBLES project (Herrera et al. 2010). The spatial correlations of the precipitation indices of models involved in both projects (PROMES and REMO) are similar in both cases. In the case of ENSEMBLES models, for some indices there was a clear division between models with high and low correlation values. In particular, results for R20MM index can be compared directly. The comparison indicates that all models of the present ensemble are at a similar level to the best subgroup of ENSEMBLES models, and clearly above the worse models of that project. Therefore, the ESCENA simulations (covering more emissions scenarios and GCMs) can be considered a valuable complement to the ENSEMBLES results for impact studies over Spain. Acknowledgements The Spanish Ministerio de Medio Ambiente y Medio Rural y Marino funded this research through project ESCENA (Ref: ). REMO results (from UAH) were obtained through the Spanish project CGL C Project CGL , financed by the Spanish Ministerio de Economía y Competitividad, also contributed to this study. 76

87 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 5. References AEMET (Agencia Estatal de Meteorología) (2007) Resumen de extremos climatológicos en España (2007). AEMET (in Spanish) Alexander L, Zhang X, Peterson T, Caesar J and others (2006) Global observed changes in daily climate extremes of temperature and precipitation. J Geophys Res 111: D05109, doi: /2005JD Barriopedro D, Fischer E, Luterbacher J, Trigo R, García-Herrera R (2011) The hot summer of 2010: redrawing the temperature record map of Europe. Science 332: Beniston M (2006) August 2005 intense rainfall event in Switzerland: not necessarily an analog for strong convective events in a greenhouse climate. Geophys Res Lett 33: L05701, doi: /2005GL Beniston M, Stephenson DB (2004) Extreme climatic events and their evolution under changing climatic conditions. Global Planet Change 44: 1 9 Beniston M, Stephenson DB, Christensen OB, Ferro CAT and others (2007) Future extreme events in European climate: an exploration of regional climate model projections. Clim Change 81: Boberg F, Berg P, Thejll P, Gutowski WJ, Christensen JH (2009) Improved confidence in climate change projections of precipitation evaluated using daily statistics from the PRUDENCE ensemble. Clim Dyn 32: Boberg F, Berg P, Thejll P, Gutowski WJ, Christensen JH (2010) Improved confidence in climate change projections of precipitation further evaluated using daily statistics from ENSEMBLES models. Clim Dyn 35: Brinkop S, Roeckner E (1995) Sensitivity of a general circulation model to parameterizations of cloud turbulence interactions in the atmospheric boundary layer. Tellus A 47: Brunetti M, Colacino M, Maugeri M, Nanni T (2001) Trends in the daily intensity of precipitation in Italy from 1951 to Int J Climatol 21: Castro M, Fernández C, Gaertner MA (1993) Description of a mesoscale atmospheric numerical model. In: Diaz JI, Lions Jl (eds) Mathematics, climate and environment. Mason, Paris, p Castro M, Gallardo C, Jylhä K, Tuomenvirta H (2007) The use of a climate-type 77

88 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA classification for assessing climate change effects in Europe from an ensemble of nine regional climate models. Clim Change 81(S1): Chaboureau JP, Bechtold P (2002) A simple cloud parameterization derived from cloud resolving model data: diagnostic and prognostic applications. J Atmos Sci 59: Chaboureau JP, Bechtold P (2005) Statistical representation of clouds in a regional model and the impact on the diurnal cycle of convection during Tropical Convection, Cirrus and Nitrogen Oxides (TROCCINOX). J Geophys Res 110: D17103, doi: /2004JD Chen F, Dudhia J (2001a) Coupling an advanced land surface-hydrology model with the Penn State-NCAR MM5 modeling system. I. Model implementation and sensitivity. Mon Weather Rev 129: Chen F, Dudhia J (2001b) Coupling an advanced land surface-hydrology model with the Penn State-NCAR MM5 modeling system. II. Preliminary model validation. Mon Weather Rev 129: Christensen JH, Christensen OB (2007) A summary of the PRUDENCE model projection of changes in European climate by the end of this century. Clim Change 81: 1 30 Collins WD, Rasch PJ, Boville BA, Hack JJ, McCaa JR, Williamson DL, Briegleb BP (2006) The formulation and atmospheric simulation of the Community Atmosphere Model version 3 (CAM3). J Clim 19: Cuxart J, Bougeault P, Redelsperger JL (2000) A turbulence scheme allowing for mesoscale and large-eddy simulations. QJR Meteorol Soc 126: 1 30 de la Cruz Gallego M, García JAG, Vaquero JM (2004) Distribución espacial de índices de frecuencia de precipitación diaria en la Peninsula Ibérica. Física de la Tierra 16: Dee DP, Uppala S (2009) Variational bias correction of satellite radiance data in the ERA-Interim reanalysis. QJR Meteorol Soc 135: Della-Marta P, Haylock M, Luterbacher J, Wanner H (2007) Doubled length of western European summer heat waves since J Geophys Res 112: D15103, doi: / 2007JD Déqué M, Rowell DP, Lthi D, Giorgi F and others (2007) An intercomparison of regional climate simulations for Europe: assessing uncertainties in model projections. 78

89 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Clim Change 81(S1): Domínguez M, Gaertner M, De Rosnay P, Losada T (2010) A regional climate model simulation over West Africa: parameterization tests and analysis of land-surface fields. Clim Dyn 35: Dudhia J (1989) Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model. J Atmos Sci 46: Dudhia J (1993) A nonhydrostatic version of the Penn State-NCAR mesoscale model: validation tests and simulation of an Atlantic cyclone and cold front. Mon Weather Rev 121: ECMWF (European Centre for Medium-Range Weather Forecasts) (2004) IFS documentation CY28r1. ECMWF, Reading, p 7 32, CY28r1/pdf_files/Physics.pdf Fernández J, Montávez JP, Sáenz J, González-Rouco JF, Zorita E (2007) Sensitivity of the MM5 mesoscale model to physical parameterizations for regional climate studies: annual cycle. J Geophys Res 112(D4): D04101, doi: /2005JD Fernández-Quiruelas V, Fita L, Fernández J no AC (2010) WRF workflow on the Grid with WRF4G. Proc 11th WRF Users Workshop, June, Boulder, CO. www. meteo.unican.es/files/pdfs/wrfusers_wrf4g.pdf Field C, Barros V, Stocker T, Qin D and others (eds) (2012) Managing the risks of extreme events and disasters to advance climate change adaptation. A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge Fischer EM, Schär C (2010) Consistent geographical patterns of changes in highimpact European heatwaves. Nat Geosci 3: Fischer EM, Seneviratne SI, Lüthi D, Schär C (2007) Contribution of land atmosphere coupling to European summer heat waves. Geophys Res Lett 34: L06707, doi: / 2006GL Fita L, Fernández J, García-Díez M (2010) CLWRF: WRF modifications for regional climate simulation under future scenarios. Proc 11th WRF Users Workshop, June, Boulder, CO Frei C, Schöll R, Fukutome S, Schmidli J, Vidale PL (2006) Future change of precipitation extremes in Europe: intercomparison of scenarios from regional climate models. J Geophys Res 111: D06105, doi: /2005JD

90 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Frich P, Alexander LV, Della-Marta P, Gleason B, Haylock M, Klein Tank AM, Peterson T (2002) Observed coherent changes in climatic extremes during the second half of the twentieth century. Clim Res 19: Gallego MC, García JA, Vaquero JM(2005) The NAO signal in daily rainfall series over the Iberian Peninsula. Clim Res 29: 103 García-Díez M, Fernández J, Fita L, Yagüe C (2013) Seasonal dependence of WRF model biases and sensitivity to PBL schemes over Europe. QJR Meteorol Soc 139: García-Herrera R, Díaz J, Trigo R, Hernández E and others (2005) Extreme summer temperatures in Iberia: health impacts and associated synoptic conditions. Ann Geophys 23: Giorgetta M, Wild M (1996) The water vapour continuum and its representation in ECHAM4. MPI Rep 162 Giorgi F, Bi X, Pal JS (2004a) Mean, interannual variability and trends in a regional climate change experiment over Europe. I. Present-day climate ( ). Clim Dyn 22: Giorgi F, Bi X, Pal JS (2004b) Mean, interannual variability and trends in a regional climate change experiment over Europe. II. Climate change scenario ( ). Clim Dyn 23: Giorgi F, Jones C, Asrar G (2009) Addressing climate information needs at the regional level: the CORDEX framework. WMO Bulletin 58: 3 Gómez-Navarro JJ, Montávez JP, Jiménez-Guerrero P, Jerez S, García-Valero JA, González-Rouco JF (2010) Warming patterns in a regional climate projections over the Iberian Peninsula. Met Zeit 19: González-Hidalgo JC, De Luis M, Raventós J, Sánchez JR (2003) Daily rainfall trend in the Valencia region of Spain. Theor Appl Climatol 75: , martin.deluis/gonzalez-hidalgo%20et% 20al.,% %20 (TAC).PDF Grell GA (1993) Prognostic evaluation of assumptions used by cumulus parameterizations. Mon Weather Rev 121: Grell GA, Devenyi D (2002) A generalized approach to parameterizing convection combining ensemble and data assimilation techniques. Geophys Res Lett 29: L1693, doi: / 2002GL

91 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Grell GA, Dudhia J, Stauffer DR (1994) A description of the fifth-generation Penn State/NCAR mesoscale model (MM5). Tech Note TN-398+ IA. NCAR: Haylock MR, Hofstra N, Klein-Tank AMG, Klok EJ, Jones PD, New M (2008) A European daily high-resolution gridded data set of surface temperature and precipitation for J Geophys Res 113: D20119, doi: /2008JD Herrera S (2011) Desarrollo, validación y aplicaciones de Spain02: una rejilla de alta resolución de observaciones interpoladas para precipitación y temperatura en España. PhD thesis, Universidad de Cantabria, available at Herrera S, Fita L, Fernández J, Gutiérrez JM (2010) Evaluation of the mean and extreme precipitation regimes from the ENSEMBLES regional climate multimodel simulations over Spain. J Geophys Res 115: D21117, doi: / 2010JD Herrera S, Gutiérrez JM, Ancell R, Pons MR, Frías MD, Fernández J (2012) Development and analysis of a 50-year high-resolution daily gridded precipitation dataset over Spain (Spain02). Int J Climatol 32: Hirschi M, Seneviratne SI, Alexandrov V, Boberg F and others (2010) Observational evidence for soil-moisture impact on hot extremes in southeastern Europe. Nat Geosci 4: Hoinka KP, de Castro M (2003) The Iberian peninsula thermal low. QJR Meteorol Soc 129: Hong SY, Lim JOJ (2006) The WRF Single-Moment 6-Class Microphysics Scheme (WSM6). J Korean Meteorol Soc 42: Hong SY, Pan HL (1996) Nonlocal boundary layer vertical diffusion in a mediumrange forecast model. Mon Weather Rev 124: Hong S, Dudhia J, Chen S (2004) A revised approach to ice microphysical processes for the bulk parameterization of cloud and precipitation. Mon Weather Rev 132: Hong SY, Noh Y, Dudhia J (2006) A new vertical diffusion package with an explicit treatment of entrainment processes. Mon Weather Rev 134: Huth R, Kyselý J, Pokorná L (2000) A GCM simulation of heat waves, dry spells, and their relationships to circulation. Clim Change 46:

92 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA IPCC (2001) Climate Change 2001: the scientific basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Houghton JT, Ding Y, Griggs DJ, Noguer M and others (eds) Cambridge University Press, Cambridge IPCC (2007) Climate Change 2007: the physical science basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Solomon S, Quin D, Manning M, Chen Z and others (eds) Cambridge University Press, Cambridge Jacob D (2001) A note to the simulation of the annual and inter-annual variability of the water budget over the Baltic Sea drainage basin.meteorol Atmos Phys 77: Jacob D, Van den Hurk B, Andrae U, Elgered G and others (2001) A comprehensive model inter-comparison study investigating the water budget during the BALTEXPIDCAP period. Meteorol Atmos Phys 77: Jacob D, Barring L, Christensen OB, Christensen JH and others (2007) An intercomparison of regional climate models for Europe: model performance in presentday climate. Clim Change 81(S1): Janjic ZI (1990) The Step-Mountain Coordinate: physical package. Mon Weather Rev 118: Janjic ZI (1994) The Step-Mountain Eta coordinate model: further developments of the convection. Mon Weather Rev 122: Jiménez-Guerrero P, Montávez JP, Domínguez M, Romera R and others (2013) Mean fields and interannual variability in RCM simulations over Spain: ESCENA project. Clim Res 57: Kain JS, Fritsch JM (1993) Convective parameterization for mesoscale models: the Kain Fritsch scheme. The representation of cumulus convection in numerical models. Am Meteorol Soc 46: Kjellström E, Boberg F, Castro M, Christensen JH, Nikulin G, Sánchez E (2010) Daily and monthly temperature and precipitation statistics as performance indicators for regional climate models. Clim Res 44: Klein Tank AMG, Zwiers FW, Zhang X (2009) Guidelines on analysis of extremes in a changing climate in support of informed decisions for adaptation. Climate Data and Monitoring, WCDMP 72: WMO TD No. 1500, WMO, Geneva, wcdmp_ series/documents/ WCDMP_72_TD_ 82

93 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA 1500_ en_ 1. pdf Krinner G, Viovy N, de Noblet-Ducoudré N, Ogée J and others (2005) A dynamic global vegetation model for studies of the coupled atmosphere biosphere system. Global Biogeochem Cycles 19: GB1015, doi: / 2003 GB Kysely J, Plavcova E (2010) A critical remark on the applicability of E-OBS European gridded temperature dataset for validating control climate simulations. J Geophys Res 115:D23118, doi: /2010jd Lana X, Martínez MD, Burgueño A, Serra C, Martín-Vide J, Gómez L (2006) Distributions of long dry spells in the Iberian Peninsula, years Int J Climatol 26: López-Moreno JI, Beniston M (2009) Daily precipitation intensity projected for the 21st century: seasonal changes over the Pyrenees. Theor Appl Climatol 95: Martin-Vide J (2004) Spatial distribution of a daily precipitation concentration index in peninsular Spain. Int J Climatol 24: Mlawer EJ, Taubman SJ, Brown PD, Iacono MJ, Clough SA (1997) Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J Geophys Res 102(D14): Morcrette JJ, Smith L, Fourquart Y (1986) Pressure and temperature dependance of the absorption in longwave radiation parameterizations. Beitr Phys Atmos 59: Nordeng TE (1994) Extended versions of the convective parametrization scheme at ECMWF and their impact on the mean and transient activity of the model in the tropics. Tech Rep ECMWF, Tech Memo No. 206, Reading Perkins SE, Pitman AJ, Holbrook NJ, McAneney J (2007) Evaluation of the AR4 climate models simulated daily maximum temperature, minimum temperature, and precipitation over Australia using probability density functions. J Clim 20: Roeckner E, Arpe K, Bengtsson L, Christoph M and others (1996) The atmospheric general circulation model ECHAM-4: model description and simulation of presentday climate. MPI Meteorol Rep 218, Hamburg Sánchez E, Gallardo C, Gaertner MÁ, Arribas A, de Castro M (2004) Future climate extreme events in the Mediterranean simulated by a regional climate model: a first approach. Global Planet Change 44:

94 4EXTREMOS DE PRECIPITACIÓN Y TEMPERATURA DE CLIMA PRESENTE SOBRE ESPAÑA Sánchez E, Domínguez M, Romera R, López-Franca N, Gaertner MA, Gallardo C, Castro M (2011) Regional modeling of dry spells over the Iberian Peninsula for present climate and climate change conditions. Clim Change 107: Schär C, Vidale PL, Lüthi D, Frei C, Häberli C, Liniger MA, Appenzeller C (2004) The role of increasing temperature variability in European summer heatwaves. Nature 427: Sillmann J, Roeckner E (2008) Indices for extreme events in projections of anthropogenic climate change. Clim Change 86: Skamarock WC, Klemp JB, Dudhia J, Gill DO and others (2008) A description of the advanced research WRF version 3. NCAR TN 475, wrf/ users/docs/ arw_v3.pdf Sundqvist H (1978) A parameterization scheme for non-convective condensation including prediction of cloud water content. QJR Meteorol Soc 104: Tiedtke M (1989) A comprehensive mass flux scheme for cumulus parameterization in large scale models. Mon Weather Rev 117: Uppala SM, Kallberg PW, Simmons AJ, Andrae U and others (2005) The ERA-40 reanalysis. QJR Meteorol Soc 131: Verdin KL, Greenlee SK (1996) Development of continental scale digital elevation models and extraction of hydrographic features. Proc 3rd Int Conf/Workshop on Integrating GIS and Environmental Modeling, Santa Fe, NM, January 21 26, National Center for Geographic Information and Analysis, Santa Barbara, CA Vicente-Serrano SM (2006a) Differences in spatial patterns of drought on different time scales: an analysis of the Iberian Peninsula. Water Resour Manage 20: Vicente-Serrano SM (2006b) Spatial and temporal analysis of droughts in the Iberian Peninsula ( ). Hydrol Sci J 51: Vicente-Serrano SM, Lasanta T, Romo A (2004) Analysis of spatial and temporal evolution of vegetation cover in the Spanish central Pyrenees: role of human management. Environ Manag 34: Vincent LA, Mekis E (2006) Changes in daily and extreme temperature and precipitation indices for Canada over the Twentieth Century. Atmos-Ocean 44:

95 5 PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Este capítulo reproduce el siguiente manuscrito: Romera R, Sánchez E, Domínguez M, Gaertner MA, Gallardo C (2015) Evaluation of present-climate precipitation in 25 km resolution RCM simulations over North-West Africa, Clim Res. In press. DOI: /cr01330

96

97 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 5.1 Resumen Los riesgos relacionados con los futuros cambios en las características de la precipitación sobre el Norte de Africa (reducción de la precipitación media, inundaciones y sequías) ha sido señalada por varios autores, pero la escasez de estudios en este área de África ha sido apuntada por el IPCC. Se ha analizado las características de la precipitación sobre el Norte de África en clima presente ( ) por un conjunto de cinco modelos regionales de clima (RCMs) de alta resolución (25 km) del proyecto español ESCENA. La evaluación de las simulaciones de clima presente de los RCMs se realiza comparando los resultados con cuatro bases de datos observacionales diferentes. El análisis de la distribución espacial de la precipitación media y el ciclo anual revela la compleja distribución espacial de la precipitación y los diferentes patrones dependiendo del área, por lo que se realiza un análisis más específico subdividiendo el dominio en ocho regiones. La heterogeneidad tanto espacial como temporal del patrón de precipitación en el dominio estudiado se muestra también en los eventos extremos (los días de precipitación fuerte oscila entre 4 y 16 días al año y los días secos consecutivos entre 240 y 330 días) y en el análisis de anomalías anuales (valores entre -50% y 100 %). El estudio revela características de precipitación muy robustas, pero también diferencias significativas entre los modelos, con una dispersión mayor del 70% en algunas subregiones, apuntando a la necesidad de usar un conjunto de RCMs y la importancia de la alta resolución en un dominio tan complejo. Como los modelos regionales son capaces de describir las características básicas de la precipitación sobre la región, pueden usarse para evaluar los cambios proyectados por las simulaciones de cambio climático. 87

98 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 88

99 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 5.2 Evaluation of present-climate precipitation in 25 km resolution RCM simulations over North-West Africa Abstract Risks related to precipitation future changes characteristics over Northern Africa (decreasing mean precipitation, floods and droughts) have been highlighted by several authors, but the scarcity of studies in this area of Africa has been noted by IPCC. Northern Africa present-climate ( ) precipitation features from a set of five high resolution (25 km) RCMs from ESCENA Spanish project are analysed. The evaluation of present-climate RCMs simulations is done comparing the results with four different observational databases. The analysis of the spatial distribution of mean precipitation and the annual cycle reveals the complex spatial distribution of the rainfall and the different patterns depending on the area, so a more specific analysis subdividing the domain in 8 regions is done. The spatial and temporal heterogeneity of the precipitation pattern in the studied domain is also shown in the extreme events (heavy precipitation days oscillate between 4 and 16 days year -1 and consecutive dry days between 240 and 330 days) and annual anomalies analyses (values between -50% and +100%). The study reveals many robust precipitation features over the region, but also significant differences between the models, with a spread larger than 70% in some subregions, pointing to the need of a set of RCMs and the importance of high resolution in a very complex domain. As regional models are able to describe the basic precipitation features over the region, they can be used to assess the changes projected by climate change simulations. 1. Introduction According to IPCC (2014) Africa will be very affected by the climate change due, in part, to its limited adaptive capacity. The report also highlights the scarcity of studies and observed data in such continent. Climate model studies over Africa have been localized in Western Africa (Wang & Eltahir 2000 and Druyan 2011), East Africa (Sun et al. 1999) and South Africa (Hudson 1997 and Joubert & Hewitson 1997), but rarely reach the north coast of the continent, especially the coast of Morocco, Algeria and Tunisia where the populated areas are very important. Other studies have been centered in the north of Africa (Driouech et al in Morocco or Bargaoui et al 2014 in Tunisia), but most of them do not use Regional Climate Models (RCMs) (Knippertz et al. 2003). The North-West Maghreb (Morocco, Algeria and Tunisia) is critically dependent on climate because the main occupancy and mean of subsistence is the agriculture so 89

100 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA the increase of rainfall variability could seriously affect whole countries (Mougou et al. 2011). In these cases the knowledge of the present climate and the changes projected is crucial to adapt the agriculture (Rosenberg 1992). The Maghreb is an area of dense settlement bounded by desert to the south and by large water bodies in other directions. The complexity of the orography of these countries is a great challenge to the RCMs, as it includes the coast, agriculture areas, very high mountains (over 4000 m) and even desert areas, so the resolution of the RCMs is very important (Knippertz et al. 2003, Huebener & Kerschgens 2007 and Born et al. 2008b). Therefore, this region is very interesting in terms of precipitation as it includes areas with precipitation as large as 2000 mm year -1, mountainous areas (orographic precipitation), synoptic forcings with differences on the Atlantic and Mediterranean sides, and a large desert such as the Sahara with almost no rain all along the year. During the 20th century, the Maghreb has been the area that has suffered the mayor warming over whole Africa (Hulme et al. 2001). According to Driouech et al. (2010) a decrease in Morocco mean precipitation and a change in extreme events in this country is expected in the middle of the century. Agriculture areas are very vulnerable in medium latitudes to projected changes, like intense precipitations occasioning floods and decreasing water availability (Born et al. 2008b). The precipitation in Morocco has been decreasing from 1960 (Knippertz et al. 2003), increasing the frequency and persistence of the droughts (Benassi 2008). The Maghreb area has not been included in the domain of the most important Mediterranean projects with ensembles of RCMs, as PRUDENCE (Christensen & Christensen 2007) or ENSEMBLES (Van der Linden & Mitchell 2009) European projects, which could explain the scarcity of studies over this region. However there have been some efforts of analysing the climate over Northern Africa using data from ENSEMBLES, as the study of extreme precipitation in Morocco of Tramblay et al., (2012) or the study of seasonal precipitation variability in the north of Tunisia of Bargaoui et al. (2014). There have been other efforts at studying African climate through RCM simulations, as AMMA (Redelsperger et al. 2006) and IMPETUS (Born et al. 2008a) projects centered over West Africa, and other studies as Patricola & Cook (2010). Future climate studies using statistical downscaling have also included the North of Africa as Hertig and Jacobeit (2008), Hertig et al. (2012, 2013) or Jacobeit et al. (2014). Due to the small amount of RCMs simulations over this continent, CORDEX (COordinated Regional climate Downscaling Experiment) project has centered its first runs and results over such domain (Giorgi et al. 2009), due to the necessity of 90

101 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA analysing its future climate. In this frame, 50 km RCM runs have been performed over the whole African continent. For this reason, in the last years many evaluation studies with RCMs over Africa have been published. Some of this studies have been centered in a specific region of the continent, as Endris et al. (2013) in Eastern Africa, Gbobaniyi et al. (2014) in Western Africa or Kalognomou et al. (2013) in Southern Africa, but none of them has focused its efforts in the Northern part of the continent. Other studies (Laprise et al. 2013, J. Kim et al or Hernández-Díaz et al. 2013) analyse the complete domain and propose specific analysis over smaller subregions, being Maghreb one of them, but they do not finally show any explicit result over this region. The principal weakness of CORDEX simulations to study Maghreb climate is the resolution because CORDEX resolution is not high enough to reproduce the Maghreb complex orography (Knippertz et al. 2003). Nikulin et al. (2012) analyses the complete African domain, concluding that RCMs simulations reproduce the details of African climatology but, at the same time, present important biases depending on the region and the season. In the framework of CORDEX new RCM simulations at high spatial resolution (12 km) are becoming available (Euro-CORDEX (Kotlarski et al. 2014, Vautard et al. 2013) and Med-CORDEX) over the North of Africa, providing a very useful tool to analyse the climate over this region (Tramblay et al. 2013). Panitz et al. (2014) also analyses the complete African domain as well as studying different regions, Maghreb among them; in this case 50 and 25 km simulation resolution are used, but they use only one RCM. The study concludes that a higher resolution does not improve the results over the continent. In such study Maghreb is considered as one region despite its climatic diversity, turning out that it is a region with very scarce precipitation however only two seasons (JFM and JAS) are studied so the complete rainy season of the region is not represented, a more detailed analyses would show that North Africa present an important rainy season. Liebmann et al. (2012) determine the average start and end dates of wet season from October to March as it was already used in Driouech et al. (2009), where it was called extended winter, while other authors include April in the rainy season (Schilling et al. 2012). Hertig and Jacobeit (2008) determine the rainy season from October to May, but in this case the study include all the Mediterranean coast, so they have to consider different climates with different rainy periods. Canary Islands, an archipelago of volcanic origin sited in the Atlantic Ocean, in front of Morocco Coast, are also object of this study. Such archipelago has also a very complex orography, with a total area of less of 7500 km 2 and seven principal islands. 91

102 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Its highest summit reaches more than 3700 m (Teide). Despite its limited extension, the climate presents a noticeable heterogeneity, with a marked influence by the ocean but also being affected by the dry dusty wind from the Sahara. At such latitude, the amount of annual precipitation could be much scarce but the ocean currents, trade winds and the abrupt orography makes some of the islands much wetter than expected (Herrera et al. 2001). The detected strong decreasing trend in the Canary Island precipitation for the second half of 20 th century (García-Herrera et al. 2003) makes interesting the study of future precipitation changes. As Tramblay et al. (2013) remark, the study of precipitation extremes (dry days and heavy daily precipitation) is very important in an area with a strong variability of precipitation where the agriculture is the first economic activity and its production depends on the precipitation. Several publications have proved the need of RCMs for this kind of studies, as they are a better way to reproduce extreme precipitation (Gao et al. 2006, Domínguez et al and Sánchez et al. 2004) compared with GCMs (General Circulation Models) which are not able to reproduce frequency and magnitude of extreme precipitation (Frei et al. 2006, Fowler et al and Giorgi & Lionello 2008). Section 2 describes the observed databases and RCMs simulations used in this study. Also the methodology is explained. In section 3.I mean annual precipitation values are analysed: observed and simulated accumulated precipitation are compared in the studied domain, the mean biases for the rainy season among models and observations are analysed and the annual cycle for each subregion is studied. Precipitation extremes are analysed in section 3.II by mean of different precipitation indices. In section 3.III annual precipitation anomalies are considered. Finally, in section 4, the most important conclusions of the study are summarised. 2. Data and Methodology 2.I. Regional Climate Simulations The Maghreb climate (Moustadraf et al. 2008) ranges from arid to semi-arid Saharian climate in the southern part, oceanic in the western part and Mediterranean in the northern part; these different zones are delimited by important mountain ranges (Atlas and Rif) with mountainous climate. This climatic heterogeneity is very similar to the nearby European region (Spain and Portugal), especially in the winter season (Driouech et al. 2009). Therefore, RCMs from ESCENA project, already evaluated in Iberian Peninsula (Jiménez-Guerrero et al and Domínguez et al. 2013), have been used in this study. 92

103 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Five RCMs simulations from the Spanish project ESCENA have been analysed: PROMES (Domínguez et al. 2010), two different versions of WRF (Weather Research and Forecasting) (Skamaroch et al. 2008, Fita et al. 2010), MM5 (Grell et al. 1995, Gómez-Navarro et al. 2010) and REMO (Jacob 2001). The high resolution (25 km) runs are nested in ERA-Interim reanalysis. For further information about the RCMs, the simulations and the ESCENA project see Jiménez-Guerrero et al. (2013) and Domínguez et al. (2013). Figure 5.1. RCMs study domain including orography (left panel). Poposed subregions (right panel): Northern Rif (NR), Southern Rif (SR), Western Atlas (WA), Eastern Atlas (EA), Mediterranean Coast (MC), Inner Argelia (IA), Tunisian (TU) and Canary Island (CI). 2.II. Observational datasets In order to evaluate the RCMs simulations four gridded 0.5º resolution observational databases have been used in this analysis: three monthly precipitation datasets (Climate Research Unit Data, CRU, from the University of East Anglia (Mitchell & Jones 2005), Terrestrial Precipitation from the University of Delaware, UDEL (Willmott & Matsuura 2000) and Global Precipitation Climatology Center, GPCC (Schneider et al. 2011)) and one daily precipitation dataset (NOAA, National Oceanic and Atmospheric Administration, Climate Prediction Center, CPC (Chen et al. 2008)), needed to analyse precipitation extremes. The use of several different observations databases is important due to the spread found among them (Nikulin et al. 2012; Kim & Lee 2003; Waliser et al. 1999), especially in areas with complex orography with a scarce number of stations (IPCC, 2014). 93

104 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 2.III. Methodology The complete domain sumulated by all the RCMs is shown in Figure 5.1 (left panel). Due to the heterogeneity of the studied domain, both climatic and orographically, the division into smaller subregions is necessary (Panitz et al. 2014). Figure 5.1 (right panel) shows the 8 subregions proposed in this study: Northern Rif (NR), Southern Rif (SR), Western Atlas (WA), Eastern Atlas (EA), Mediterranean Coast (MC), Inner Algeria (IA), Tunisian (TU) and Canary Islands (CI). As the southern part of the domain correspond to the Sahara desert, with basically zero precipitation, no subregions are selected there, because a more specific analysis of the precipitation in such area will not provide any interesting result. Although any distribution of subregions is always somewhat arbitrary, the proposal here is based on the search of more homogeneous precipitation values over smaller areas compared with the whole domain. In section 3.I annual precipitation is calculated as the mean over the 20 years of the accumulated annual precipitation. The mean ( ) precipitation biases (relative difference, Kotlarsky et al. 2014) between CPC database and every observational dataset and every RCM have been calculated over each subregion for the rainy season (October to March). Also the annual cycle for each subregion (spatial average) is studied (Panitz et al. 2014). Precipitation extremes are analysed in section 3.II by means of two different precipitation core ETCCDI (Expert Team on Climate Change Detection and Indices) indices in each different region (R10MM and CDD). The heavy precipitation days index (R10MM) is calculated as the 20 years-mean number of days with precipitation over 10 mm. Both the spatial distribution and the subregions-averaged histogram are shown. The consecutive dry days index (CDD) is the subregions-averaged maximum period of consecutive dry days in 20 years (histogram shown), where dry days are defined as days with precipitation less than 1 mm. Observational CPC database is used to validate the simulated extreme indices as it is the only daily observed dataset. In section 3.III spatial averaged annual precipitation anomalies bring into focus the variability of the climate in the different subregions. The anomalies have been calculated, in each region, as the difference between the annual accumulated precipitation and the 20 years-mean precipitation; the anomaly has been normalized dividing by the 20 years-mean precipitation (Panitz et al. 2014) and it has been expressed in percentage. 94

105 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 3. Results 3.I. Mean precipitation A first overview of the annual precipitation fields obtained from the four observational datasets (Figure 5.2) indicate that this region is very complex in terms of the rainfall distribution. There are two clear maximum areas, around the Gibraltar Strait and also around the limit between Algeria and Tunisia Mediterranean coast. It is also evident a relative minimum in precipitation on the Mediterranean coast between both maximum areas. In Algeria and Tunisia, rainfall amounts decrease from the coast to the inner regions, to almost zero values when reaching the Sahara desert. In Morocco, orographic maxima can be seen over the Atlas and Rif mountain chains. Over the Atlantic coast of Morocco, a clear reduction in precipitation from the maximum values at Gibraltar Strait is also observed. Figure 5.2. Mean accumulated annual precipitation (mm year-1) from observations: CPC (upper left), CRU (upper right), GPCC (lower left) and UDEL (lower right). 95

106 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Despite these common precipitation features, also some relevant differences can be seen when these four precipitation datasets are compared. CRU dataset underestimate maximum values when compared with the other datasets, giving for example values around 600 mm year -1 for the Algeria-Tunisia maximum, meanwhile the other give more than 1000 mm year -1 there. Orographic features over the Atlas or Rif can be seen for GPCC or UDEL datasets, but CPC or CRU exhibit a much smoother pattern. It must be considered that surface rain-gauge stations are likely to be just a few over the domain of study, so the interpolation procedures and their representativity of the regional surrounding climate can be highly questionable. This uncertainty on the comparison among observational results is very relevant when analysing the results obtained from the RCM simulations. Regional climate modeling results from the five RCM simulations (Figure 5.3) are able to reproduce the same global patterns described by the observational datasets, that is, the location of the maximum precipitation values, the gradient of decrease over the inner part of the continent, and also the reduction over the Atlantic coast. RCMs tend to give a more detailed description over the mountainous areas, which is likely to be the result of the higher resolution of the RCMs against the observational datasets (50km vs 25km), together with the additional reduction of observational information over mountain regions (Briggs & Graham Cogley 1996), and perhaps the tendency of regional models to overestimate orographical precipitation. As it is shown in Nikulin et al. (2012), 50 km simulations are not able to reproduce orographical precipitation over Atlas. It is interesting to notice the relative minimum of precipitation obtained by all the RCMs not only over the Mediterranean coast between both maximums, but over all the Alboran Sea and the Almeria region (in Spain). Also the Canary Islands, despite the limited amount of land points used by the RCMs there exhibit an interesting pattern, all the RCMs show a relative minimum of precipitation on its southern part, as it can be seen when looking at detailed observational information (Herrera et al. 2001). It is a well known feature, due to their steep orography, combined with the typical trade-wind synoptic conditions of this region. 96

107 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Figure 5.3. Mean accumulated annual precipitation (mm year -1) from RCMs: REMO (upper left), WRA (upper right), WRB (center left), PROMES (center right) and MM5 (lower left). 97

108 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Figure 5.4. Mean rainy seasonal (Oct-Mar) relative precipitation bias (%) respect to CPC for observational data-sets: CRU (square), UDEL (circle) and GPCC (triangle) and for models: REMO (sky blue), WRFA (violet), WRFB (blue), MM5 (green) and PROMES (red), over 8 subregions indicated in Figure 5.1. Some discrepancies among the RCMs and when compared with observations are also obtained, as already pointed by Jiménez-Guerrero et al. (2013), while analysing mean climate conditions simulated over the Iberian Peninsula, such as the overestimation of precipitation values obtained by PROMES or the more smoothed results by MM5. Comparing high resolution observations to datasets based on a limited amount of stations Herrera et al. (2012) concluded that observational datasets underestimate the maximum values over the nearby Iberian Peninsula. Figure 5.4 shows the mean rainy seasonal (Oct-Mar) relative precipitation bias for each observational data-set and each RCMs respect to CPC. The spread among observational data-sets is smaller than the spread among RCM in every region except CI, where the number of grid points is very scarce. UDEL and GPCC biases are very similar over every region (Figure 5.2 also shows a very similar pattern between both observational data-set) except MC and CI. The largest bias (more than 100%) is presented by UDEL over CI, where the grid points are very scarce, as it can be seen in Figure

109 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA In most of the regions, except NR, IA and CI, the bias among the observations is smaller than 25 %, being EA the region with the smallest bias. The bias presented by the RCMs is negative in all the regions. Only two RCMs change this pattern: REMO in SR respect to CRU and in WA respect to CPC and PROMES that only present a negative bias over NR. This results agree with Jiménez-Guerrero et al. (2013) that uses the same RCMs over the neighbour Iberian Peninsula, where only REMO in two seasons and PROMES in every one have positive biases, and with Bargaoui et al. (2014) that conclude a 20% underestimation of the precipitation by RCMs in Tunisia. The complexity of the precipitation patterns shown on Figure 5.2 points to the interest of separating the domain of study on several subregions, for a more detailed analysis, as proposed on Figure 5.1. This subregional description allows us, for example, to describe the structure of the annual cycle of precipitation over each of the subregions, as can be seen on Figure

110 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Figure 5.5. Annual cycles of mean monthly precipitation (mm month -1) ( ) for four observational datasets (dashed lines: CPC green, UDEL red, GPCC blue, CRU black) and five RCMs (solid lines: MM5 green, PROMES red, WRFB blue, WRFA violet, REMO sky blue) over 8 sub-regions indicated in Figure

111 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA On Figure 5.5, the observational datasets (dashed lines) allow for a detailed description of the monthly precipitation evolution along the year for all the subregions. For most of them, the summer minimum and the winter maximum are very apparent, highlighting two seasons: the rainy season (October-March) and a very dry summer (JJA). Nevertheless, some regions exhibit a much larger winter precipitation maximum, as it is the case of NR, SR or MC, with values up to 70 mm month-1 for several months. On the other side, EA or IA range between 20 and 30 mm month-1 all over the year, with even a more homogeneous behaviour all along the year. EA annual cycle is clearly different from its surrounding NR or MC. IA is the region where lower precipitation values are obtained from all the regions, as it can be expected when looking at the spatial maps, due to its closer location to the Sahara desert. Other neighbouring regions with very different precipitation results are TU and MC. It is interesting to see that NR and SR show a comparable annual cycle, although one is related to Mediterranean forcings, and the other to more Atlantic influence. The WA region is quite different to the SR one, due to their more southern position, being both influenced by the Atlantic Ocean. The spread among observational datasets is large for NR during autumn, winter and spring, with values ranging from 50 to 80 mm month-1 over some months. This is also the case for CI region, with an even larger spread in November (50 mm month -1), but there the main reason is probably that just few and probably not totally consistent land points are considered on each observational database to compute this island chain. These annual cycles contrast with the one published by Panitz et al. (2014), where the North-West of Africa is presented as a very dry area along the year, with precipitation lower than 30 mm month -1. These results point out the interest of the present analyses over the studied domain. Regional model description of annual cycles shows the capability of the simulations to describe the climatological evolution of precipitation along the year for each of the eight selected subregions. Most of the regions exhibit a relatively good agreement among models and observations, but others present a larger spread. Thus, both RCMs and observational datasets present an important spread over NR during the rainy season, however all the RCMS underestimate the observed values, as it has been observed also in Figure 5.4. This is partly the case also for SR region, where some RCMs clearly underestimate observed values, but PROMES even slightly overestimate observed timeseries. PROMES, as already seen on the spatial maps and pointed out by the calculated biases, is the RCM that gives larger precipitation values, resulting in some clear overestimation compared with observations over other regions than SR, as it is the case for IA, MC or TU. This overestimation is largest in 101

112 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA spring and in the transition from summer to autumn, when convective precipitation contributes much to total precipitation. The other models, and particularly WRF and MM5, underestimate precipitation in all subregions, for most of the rainy months, with a negative bias respect to CPC larger than 25% over most of the regions. This underestimation is specially noteworthy for winter precipitation over the Atlantic subregion, where WRF presents a negative bias of 50%, which is mostly linked to relatively large scale Atlantic depressions. In contrast, winter precipitation is well captured by PROMES, which points to very different origins of biases among the models. A special case is the CI islands chain, where the complex orography results in a very irregular precipitation distribution that can hardly be described with neither the gridded observations or the RCMs at the resolution used here. 3.II. Extreme Events In order to analyse the extreme events over the region, the R10MM and the CDD indices have been chosen. The projected change of dry days over the studied domain could reach high values in middle-term climate ( ) (Bouagila & Sushama 2013), so a good evaluation of RCMs ERA-Interim performance is crucial. Regarding heavy precipitation episodes (R10MM), the CPC observations have a spatial distribution with north-south gradient, showing a clear dipole of maximums (Figure 5.6, upper-left panel and Figure 5.7, left panel): one peak over strait of Gibraltar and surrounding areas (NR and SR regions) and another covering most of MC region. This pattern is very similar to the pattern observed in Figure 5.2, where the mean accumulated annual precipitation in represented, and it is coherent not only with the precipitation regime of CPC but also with all observational databases used in this paper. The maximum values reach up to 37 days year

113 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Figure 5.6. Spatial distribution of R10MM precipitation extreme index (days year-1) from CPC observed data (upper left) and from 5 ERA-Interim regional climate simulations: REMO (upper right), both versions of WRF (middle), PROMES (lowef left) and MM5 (lower right). 103

114 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA In general, all the models simulate a similar spatial distribution of observed R10MM (Figure 5.6), having remarkable differences among them, specially when averaged spatially (Figure 5.7, left panel). The RCMs underestimate the observed values, except PROMES, that simulates larger areas that exceeding the 50 days per year. Both versions of WRF model show a similar distribution, while the model with the values more similar to the observations is REMO. MM5 does not simulate properly the R10MM amplitude, with values below 12 days per year over all the subregions (Figure 5.7).Based on CPC results, the CDD index seems to have an EasternWestern gradient, being the regions with Atlantic influence (CI, WA and SR) which present a greatest value of CDD. Canary Islands is the region with larger CDD period, possibly due to the scarce number of land points and to the underestimation of the relief as it is the main factor that affect the local rainfall (Herrera et al. 2001). A similar pattern has been found in the annual mean CDD. IA is one of the regions with the smallest values of both indices (R10MM and CDD), following to CI (R10MM) and TU (CDD). Most of the models underestimate CDD index. REMO (sky blue bar) is the RCM which present values closer to the observations, overestimating the index in some regions. Differences among models are smaller for the annual mean CDD, remaining REMO close to CPC while the spread among RCMs is reduced. While PROMES (red bar) and REMO (sky blue bar) overestimate one of the two extreme indices calculated, the rest of the models (MM5 and two versions of WRF) present a smoother distribution among subregions, underestimating both extreme precipitation events and dry periods. These results agree with the ones obtained by Dominguez et al. (2013) over the neighbour Iberian Peninsula. Figure 5.7. Precipitation indices histograms: R10Mm (days year-1, left) and CDD (consecutive days (20 years)-1, right) from CPC observed data (yellow column) and from 5 RCMs: REMO (sky blue bar), WRF-A (pink bar), WRF-B (dark blue bar), PROMES (red bar) and MM5 (green bar). 104

115 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 3.III. Annual anomalies In Figure 5.8 simulated and observed annual precipitation anomalies with respect to the climatology ( ) in each subregion are represented. In light of Figure 5.8 it is possible to divide the regions in two big groups: NR, SR and WA (Morocco, Atlantic Sea) have important precipitation anomalies, especially positive ones (1996 and 2008) while the rest of the regions present a smoother evolution, balancing out positive and negative anomalies is the year of the studied period with larger precipitation in Morocco, as also reflected in Benassi (2008), this positive anomaly is very important in NR, SR and WA, even exceeding 100% while in the rest of the regions this anomaly is lower than 40%, although always positive. The negative anomaly, from 1998 to 2002 is presented in all the regions, with an unique little break in 1999 in MC, pointing out a very persistent drought over all the studied domain. All the observational databases present very similar behaviour, coinciding both in the positive anomalies (1996 in NR; SR, WA and CI, 2005 in TU and CI and 1989 in CI) as in the negative ones (1994 and 1998 in NR, SR, WA and EA and 2000 in MC, IA, TU and CI), with small differences in specific years and regions, as CPC in WA in 2003 or CRU in TU in In TU and CI regions the spread among databases is more notable due possibly to the small size of the regions and the greater number of kilometers of coast line, so the masks of the databases can present more differences among them. The spread among the RCMs anomalies depends on the region. In SR and WA all RCMs have a very similar behaviour, while in NR, EA, MC or TU there are some noticeable differences as in NR in , in EA in or MM5 with respect to the rest of the models in MC and TU in In IA, TU and CI the differences among RCMs anomalies are not so infrequent as there are several years in which different models reflect anomalies of different sign. The spread among observational datasets in these subregions is also larger than in the others. 105

116 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Figure 5.8. Normalized annual precipitation anomalies (%) in each selected region (see Figure 5.1). Observed data in dashed lines: CRU black, GPCC yellow, CPC blue, UDEL green. Simulated data in solid lines: REMO sky blue, WRFA violet, WRFB blue, MM5 green, PROMES red. 106

117 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 4. Conclusions The ability of five high resolution (25 km) RCM simulations from ESCENA Spanish project to reproduce the present climate precipitation in Northern Africa is analysed in this study. The scarcity of RCMs studies over such region gives an added value to the present one; for this reason a wide range of precipitation characteristics has been analysed: spatial and temporal distribution, mean values, extreme events (high precipitation episodes and dry periods) and seasonality. The analysis reveals both orographic and climatic heterogeneity, posing an important challenge to the RCMs involved in the study. The studied region presents also a great difficulty regarding the observational databases as demonstrated by the spread obtained among them in all parameters analysed. CPC and CRU observational databases show a spatially smoother precipitation while GPCC and UDEL are able to show clearer orographic precipitation maxima. The basic spatial distribution of the annual accumulated precipitation is reproduced by the models: the maxima and the relative minimum along the Mediterranean Coast, as well as the N-S gradient from the Mediterranean Sea to the Sahara Desert and the orographic precipitation (Atlas and Rif Chains), presenting a positive mean rainy seasonal relative bias in the case of PROMES and a negative one for the rest of models in almost all the regions. The spread among RCMs is larger than observations' over all the regions. The annual cycle and the annual anomalies of precipitation in each of the 8 subregions proposed show a large spatial heterogeneity. The neighbouring NR and SR regions, despite its different forcings (Mediterranean and Atlantic, respectively), reveal a very similar annual cycle with a long and important wet season, in contrast to other regions, as IA or WA, with less precipitation along such season. IA and TU regions are the only ones with a significant precipitation during summer. RCMs reproduce both spatial distribution and seasonality of precipitation over the studied domain, being the precipitation overestimation by PROMES and the smoothed precipitation by MM5 the more remarkable aspects. In a region where floods and droughts largely affect the population, the simulation of precipitation extremes is important. Two precipitation extreme indices have been chosen: R10MM to study heavy precipitation events and CDD to study dry periods. The coastal regions NR, SR and MC present the higher values of R10MM with 15 or 16 days year-1 while the greatest number of CDD over 20 years vary from 100 days in TU to 180 days in WA, where the dry season is longer, as was reflected in the annual cycle. Worth mentioning is the case of CI with the greatest values of CDD and the lowest of R10MM; the complex orography of the islands over a very limited extension 107

118 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA of terrain and the ocean influence present a true challenge to the RCMs. PROMES and REMO represent better some aspects of these extreme indices, while MM5 and two versions of WRF present smoother results, underestimating both indices. The analysis of the annual anomalies show the ability of the RCMs to reproduce the temporal series in each region. All the models reflect the relatively flat behaviour of the annual anomalies in MC, the rainy 1996 in NR, SR, WA and IA and the dry years in WA and EA. Biases between models and against observations can be related to the convection scheme employed, when interacting with the complex orography of the region, as pointed also in Nikulin et al. (2012) for the whole African continent and previously seen in Jiménez-Guerrero et al. (2013) and Domínguez et al. (2013), with the same models, focused over the Iberian Peninsula. Similar overestimations were obtained in Paxian et al. (2015). This paper presents a global vision of the ability of five RCMs to reproduce present climate over a domain presenting a high climate heterogeneity and a very complex orography, from desert to very high mountains showing that the precipitation over North-West of Africa should not be analysed as one only region. This study presents a starting point to analyse future climate with the future climate simulations available alse at ESCENA project with the same ensemble or RCMs. Acknowledgements This work has been funded by the grant CGL (Spanish Ministry of Science and Innovation) and grant CGL R (Spanish Ministry of Economy and Competitivity), and by ESCENA project (Ref: , Spanish Ministry of Environment and Rural and Marine Affairs). These projects have been cofunded by the European Regional Development Fund. 108

119 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA 5. References Bargaoui Z, Tramblay Y, Lawin E, Servat E (2014) Seasonal precipitation variability in regional climate simulations over Northern basins of Tunisia. Int J Climatol 34: Benassi M (2008) Drought and climate change in Morocco. Analysis of precipitation field and water supply. In : López-Francos A. (ed.). Drought management: scientific and technological innovations. Zaragoza : CIHEAM, p (Options Méditerranéennes : Série A. Séminaires Méditerranéens; n. 80) Born K, Christoph M, Fink AH, Knippertz P, Paeth H, Speth P (2008a) Moroccan Climate in the Present and Future: Combined View from Observational Data and Regional Climate Scenarios. In Climate Changes Ande Water Resources in the Middel East and North Africa, Berlin: Springer Verlag, p Born K, Fink AH, Paeth H (2008b) Dry and Wet Periods in the Northwestern Maghreb for Present Day and Future Climate Conditions. Meteorol Z 17 (5): doi: / /2008/0313. Bouagila B, Sushama L (2013) On the Current and Future Dry Spell Characteristics over Africa. Atmosphere 4(3): Briggs PR, Cogley JG (1996) Topographic Bias in Mesoscale Precipitation Networks. J Climate 9: doi: / (1996)009<0205:tbimpn>2.0.co;2. Chen M, Shi W, Xie P, Silva VBS, Kousky VE, Higgins RW, Janowiak JE (2008) Assessing Objective Techniques for Gauge-Based Analyses of Global Daily Precipitation. J Geophys Res-Atmos 113. doi: /2007jd Christensen JH, Christensen OB (2007) A Summary of the PRUDENCE Model Projections of Changes in European Climate by the End of This Century. Climatic Change. doi: /s Domínguez M, Gaertner MA, De Rosnay P, Losada T (2010) A Regional Climate Model Simulation over West Africa: Parameterization Tests and Analysis of LandSurface Fields. Clim Dynam 35: doi: /s Domínguez M, Romera R, Sánchez E, Fita L,Fernández J, Jiménez-Guerrero P, Montávez JP, Cabos WD, Liguori G,Gaertner MA (2013) Present-Climate Precipitation and Temperature Extremes over Spain from a Set of High Resolution RCMs. Clim Res 58 (2): Driouech F, Déqué M, Mokssit A (2009) Numerical Simulation of the Probability 109

120 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Distribution Function of Precipitation over Morocco. Clim Dynam 32: doi: /s Driouech F, Déqué M, Sánchez-Gómez E (2010) Weather regimes-moroccan precipitation link in a regional climate change simulation. Global Planet Change 72: 1-10 Druyan LM (2011). Studies of 21st-Century Precipitation Trends over West Africa. Int J Climatol. doi: /joc Endris HS, Omondi P, Jain S, Lennard C,Hewitson B, Chang a L, Awange JL et al. (2013) Assessment of the Performance of CORDEX Regional Climate Models in Simulating Eastern Africa Rainfall. J Climate, 26: doi: /jclid Fita L, Fernández J, García-Díez M (2010) CLWRF: WRF Modifications for Regional Climate Simulation under Future Scenarios. In Proc 11th WRF Users Workshop. Boulder, CO. Font Tullot I (1983) Climatología de España y Portugal. Instituto Nacional de Meteorología. I.S.B.N Fowler HJ, Ekström M, Blenkinsop S, Smith AP (2007). Estimating Change in Extreme European Precipitation Using a Multimodel Ensemble. J Geophys ResAtmos 112. doi: /2007jd Frei C, Schöll R, Fukutome S, Schmidli J, Vidale PL (2006) Future Change of Precipitation Extremes in Europe: Intercomparison of Scenarios from Regional Climate Models. J Geophys Res-Atmos 111. doi: /2005jd Gao X, Pal JS, Giorgi P (2006) Projected Changes in Mean and Extreme Precipitation over the Mediterranean Region from a High Resolution Double Nested RCM Simulation. Geophys Res Lett 33. doi: /2005gl García Herrera R, Gallego D, Hernández E, Gimeno L, Ribera, P, Calvo N (2003). Precipitation trends in the Canary Islands. Int J Climatol, 23(2): Gbobaniyi E, Sarr A, Sylla MB, Diallo I, Lennard C, Dosio A, Dhiédiou A, et al. (2014) Climatology, Annual Cycle and Interannual Variability of Precipitation and Temperature in CORDEX Simulations over West Africa. Int J Climatol 34: doi: /joc Giorgi F, Jones C, Asrar GR (2009) Addressing Climate Information Needs at the Regional Level: The CORDEX Framework. Bulletin - World Meteorological Organization 58:

121 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Giorgi F, Lionello P (2008) Climate Change Projections for the Mediterranean Region. Global Planet Change 63: doi: /j.gloplacha Gómez-Navarro JJ, Montávez JP, Jiménez-Guerrero P, Jerez S, García-Valero JA, González-Rouco JF (2010) Warming Patterns in a Regional Climate Projections over the Iberian Peninsula. Meteorol Z 19: Grell GA, Dudhia J, Stauffer DF (1995) A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5). NCAR Technical Note, 1995, 122pp Hernández-Díaz L, Laprise R, Sushama L, Martynov A, Winger K, Dugas B (2013) Climate Simulation over CORDEX Africa Domain Using the Fifth-Generation Canadian Regional Climate Model (CRCM5). Clim Dynam 40 (5-6): doi: /s z. Herrera RG, Puyol DG, MartÍn EH, Presa LG, Rodríguez PR (2001). Influence of the North Atlantic oscillation on the Canary Islands precipitation. J Climate, 14(19): Herrera S, Gutiérrez JM, Ancell R, Pons MR, Frías MD, Fernández J (2012) Development and analysis of a 50-year high-resolution daily gridded precipitation dataset over Spain (Spain02) Int J Climatol 32: doi: /joc Hertig E, Jacobeit J (2008) Assessments of Mediterranean precipitation changes for the 21st century using statistical downscaling techniques. Int J Climatol, 28: Hertig E, Seubert S, Paxian A, Vogt G, Paeth H, Jacobeit J (2012) Changes of total versus extreme precipitation and dry periods until the end of the 21 st century: statiscial assessments for the Mediterranean area. Theor Appl Climatol, 111: 120 Hertig E, Seubert S, Paxian A, Vogt G, Paeth H, Jacobeit J (2013) Statistical modeling of extreme precipitation for the Mediterranean area under future climate change. Int J Climatol, doi: /joc.3751 Hudson DA (1997) Southern African Climate Change Simulated by the GENESIS GCM. S Afri J Sci 93: Huebener H, Kerschgens M (2007) Downscaling of Current and Future Rainfall Climatologies for Southern Morocco. Part I: Downscaling Method and Current Climatology. Int J Climatol 27: Hulme M, Doherty R, Ngara T, New M, Lister D (2001) African Climate Change: 111

122 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Clim Res 17: doi: /cr IPCC (2014) Climate Change 2014: Impacts, Adaptation and Vulnerability. Part B: Regional Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Barros VR, Field CB, Dokken DJ, Mastrandrea MD, Mach KJ, Bilir TE, Chatterjee M, Ebi KL, Estrada YO, Genova RC, Girma B, Kissel ES, Levy An, MacCracken S, Mastrandrea PR, White LL(eds)] Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 688 pp Jacob D (2001) A Note to the Simulation of the Annual and Inter-Annual Variability of the Water Budget over the Baltic Sea Drainage Basin. Meteorol Atmos Phys 77: Jacobeit J, Hertig E, Seubert S, Lutz K (2014) Statistical downscaling for climate change projections in the Mediterranean region: methods and results. Reg Environ Change 14, doi: /s Jiménez-Guerrero P, Montávez JP, Domínguez M, Romera R, Fita L, Fernández J, Cabos WD, Liguori G, Gaertner MA (2013) Mean Fields and Interannual Variability in RCM Simulations over Spain: The ESCENA Project. Clim Res 57 (3): Joubert AM, Hewitson BC (1997) Simulating Present and Future Climates of Southern Africa Using General Circulation Models. Prog Phys Geog 21: doi: / Kalognomou EA, Lennard C, Shongwe M, Pinto I, Favre A, Kent M, Hewitson B, et al. (2013) A Diagnostic Evaluation of Precipitation in CORDEX Models over Southern Africa. J Climate 26: doi: /jcli-d Kim J, Waliser DE, Mattmann CA, Goodale CE, Hart AF, Zimdars PA, Crichton DJ, et al. (2014) Evaluation of the CORDEX-Africa Multi-RCM Hindcast: Systematic Model Errors. Clim Dynam 42: doi: /s Kim J, Lee J (2003) A Multiyear Regional Climate Hindcast for the Western United States Using the Mesoscale Atmospheric Simulation Model. J Hydrometeorol. doi: / (2003)004<0878:amrchf>2.0.co;2. Knippertz P, Christoph M, Speth P (2003) Long-Term Precipitation Variability in Morocco and the Link to the Large-Scale Circulation in Recent and Future Climates. Meteorol Atmos Phys 83: doi: /s y. Kotlarski S, Keuler K, Christensen OB, Colette A, Déqué M, Gobiet A, Goergen K et 112

123 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA al. (2014) Regional Climate Modeling on European Scales: A Joint Standard Evaluation of the EURO-CORDEX RCM Ensemble. Geosci Model Dev 7, 4: doi: /gmd Laprise R, Hernández-Díaz L, Tete K, Sushama L, Šeparović L, Martynov A, Winger K, Valin M (2013) Climate Projections over CORDEX Africa Domain Using the Fifth-Generation Canadian Regional Climate Model (CRCM5). Clim Dynam 41: doi: /s Liebmann B, Bladé I, Kiladis GN, Carvalho LMV, Senay GB, Allured D, Leroux S, Funk C (2012) Seasonality of African Precipitation from 1996 to J Climate 25: doi: /jcli-d Mitchell TD, Jones PD (2005) An Improved Method of Constructing a Database of Monthly Climate Observations and Associated High-Resolution Grids. Int J Climatol 25: doi: /joc Mougou R, Mansour M, Iglesias A, Chebbi RZ, Battaglini A (2011) Climate Change and Agricultural Vulnerability: A Case Study of Rain-Fed Wheat in Kairouan, Central Tunisia. Reg Environ Change 11: doi: /s Moustadraf J, Razack M, SinanM (2008) Evaluation of the Impacts of Climate Changes on the Coastal Chaouia Aquifer, Morocco, Using Numerical Modeling. Hydrogeol J 16: doi: /s Nieto S, Frías MD, Rodríguez-Puebla C (2004) Assessing two different climatic models and the NCEP-NCAR reanalysis data for the description of winter precipitation in the Iberian Peninsula. Int J Climatol 24: , doi: /joc.999 Nikulin G, Jones C, Giorgi F, Asrar G, Büchner M, Cerezo-Mota R, Christensen OB, et al. (2012) Precipitation Climatology in an Ensemble of CORDEX-Africa Regional Climate Simulations. J Climate 25: doi: /jcli-d Panitz HJ, Dosio A, Büchner M, Lüthi D, Keuler K (2014) COSMO-CLM (CCLM) Climate Simulations over CORDEX-Africa Domain: Analysis of the ERA-Interim Driven Simulations at 0.44 and 0.22 Resolution. Clim Dynam 42: doi: /s Patricola CM, Cook KH (2010) Northern African climate at the end of the twenty-first century: an integrated application of regional and global climate models. Clim Dynam 35:

124 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Paxian A, Hertig E, Seubert S, Vogt G, Jacobeit J, Paeth H (2015) Present-day and future mediterranean precipitation extremes assessed by different statistical approaches. Clim Dynam 44: doi: /x Redelsperger J-L, Thorncroft CD,Diedhiou A, Lebel T, Parker DJ, Polcher J (2006) African Monsoon Multidisciplinary Analysis: An International Research Project and Field Campaign. B Am Meteorol Soc. doi: /bams Rosenberg NJ (1992) Adaptation of Agriculture to Climate Change. Climatic Change, doi: /bf Sánchez E, Gallardo C, Gaertner MA, Arribas A, de Castro M (2004) Future Climate Extreme Events in the Mediterranean Simulated by a Regional Climate Model: A First Approach. Global Planet Change 44: doi: /j.gloplacha Schilling J, Freier KP, Hertig E, Scheffran J (2012) Climate change, vulnerability and adaptation in North Africa with focus on Morocco. Agriculture, Ecosyst Environ 156: Schneider U, Becker A, Finger P, Meyer-Christoffer B, Rudolf A, Ziese M (2011) GPCC Full Data Reanalysis Version 6.0 at 0.5 : Monthly Land-Surface Precipitation from Rain-Gauges Built on GTS-Based and Historic Data. Mimeo. doi: /dwd_gpcc/fd_m_v6_050. Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Duda, MG, Wang W, Powers JG (2008) A description of the Advanced Research WRF Version 3, NCAR Sun L, Semazzi FHM, Giorgi F, Ogallo L (1999) Application of the NCAR Regional Climate Model to Eastern Africa: 2. Simulation of Interannual Variability of Short Rains. J Geophys Res. doi: /1998jd Tramblay Y, Badi W, Driouech F, El Adlouni S, Neppel L, Servat E (2012) Climate change impacts on extreme precipitation in Morocco. Global Planet Change 8283: Tramblay Y, Ruelland D, Somot S, Bouaicha R, Servat E (2013) High-resolution MedCORDEX regional climate model simulations for hydrological impact studies: a first evaluation of the ALADIN-Climate model in Morocco. Hydrol Earth Syst Sc 17: Van der Linden P, Mitchell JFB (2009) ENSEMBLES: Climate Change and Its Impacts: Summary of Research and Results from the ENSEMBLES Project. 114

125 5PRECIPITACIÓN DE CLIMA PRESENTE SOBRE EL NORTE DE ÁFRICA Mediterranean. Vol ley.com/doi/ / ch9/summary. Vautard R, Gobiet A, Jacob D, Belda M, Colette A, Déqué M, Fernández J, GarcíaDíez M, Goergen K, Güttler I, Halenka T, Karacostas T, Katragkou E, Keuler K, Kotlarski S, Mayer S, van Meijgaard E, Nikulin G, Patarcić M, Scinocca J, Sobolowski S, Suklitsch M, Teichmann C, Warrach-Sagi K, Wulfmeyer V, Yiou P (2013) The simulation of European heat waves from an ensemble of regional climate models within the EURO-CORDEX project. Clim Dynam doi: /s z Waliser DE, Shi Z, Lanzante JR, Oort AH (1999) The Hadley Circulation: Assessing NCEP/NCAR Reanalysis and Sparse in-situ Estimates. Clim Dynam 15: doi: /s Wang G, Eltahir EAB (2000) Ecosystem Dynamics and the Sahel Drought. Geophys Res Lett 27: doi: /1999gl Willmott CJ, Matsuura K (2000) Terrestrial Air Temperature and Precipitation: Monthly and Annual Climatologies. Centre for Climate Research, Department of Geography, University of Delaware. 2_clim.html\nhttp://scholar.google.com/scholar? hl=en&btng=search&q=intitle:terrestrial+air+temperature+and+precipitation: +Monthly+Climatologies#4. 115

126

127 6 MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO Este capítulo reproduce el texto del siguiente manuscrito: Romera R, Sánchez E, Domínguez M, Gaertner MA, Miglietta MM (2015) Medicanes description and climage change projections with a multimodel ensemble of RCMs. Global Planet Change. Under Review

128

129 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO 6.1 Resumen Este artículo presenta la capacidad de un conjunto de Modelos Regionales de Clima (RCMs) para describir los ciclones tropicales que se desarrollan sobre el Mar Mediterráneo (medicanes) y cómo se espera que cambien en condiciones de clima futuro. Debido al pequeño tamaño de este fenómeno no puede ser estudia con Modelos de Circulación General (GCMs) por lo que se usan RCMs. En primer lugar se estudian las simulaciones de diez RCMs con una resolución aproximada de 25 km, forzados con reanálisis de ERA40 ( ) y se comparan con la limitada información observacional disponible sobre medicanes. Los medicanes simulados se obtienen usando un método de detección de ciclones basado en la presión de superficie y en la velocidad del viento en niveles altos, y a continuación se analiza la estructura vertical en términos del viento térmico y de la asimetría de la estructura. Los resultados indican que casi todos los RCMs son capaces de reproducir la frecuencia de los medicanes (en torno a 1 ó 2 por año) observados, aunque en un período de tiempo más limitado comparado con las simulaciones de clima regional. Se obtiene también una importante variabilidad en el número de medicanes, su localización y el mes de desarrollo. Una vez se ha realizado el estudio de validación, se analizan las simulaciones de los RCMs, forzados con varios GCMs y cubriendo el período de 1951 a 2050 en todos los casos y hasta 2100 en algunos de ellos. Se obtiene una ligera reducción (significante en más de la mitad de los casos) en la mayor parte de las combinaciones GCM/RCM disponibles, especialmente hasta Al mismo tiempo, parece que la fuerza de los medicanes podría incrementarse hacia el final del siglo XXI. Se han visto importantes diferencias dependiendo de la subregión analizada (oeste, central o este) y también entre el sur y el norte del Mar Mediterráneo. El número de eventos es mayor en el norte, sobre los mares de Liguria, Tirreno, Adriático y Egeo, pero en el futuro estas zonas sufrirán un mayor descenso en el número de medicanes que en el sur por lo que los eventos se repartirán de manera más uniforme a lo largo del Mar Mediterráneo. Los eventos más intensos tendrán lugar básicamente sobre el Sur del Mar Mediterráneo. 119

130 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO 120

131 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO 6.2 Medicanes description and climate change projections with a multimodel ensemble of RCMs Abstract The capability of an ensemble of Regional Climate Models (RCMs) to describe the tropical-like cyclones developed over the Mediterranean Sea (medicanes) and how they are projected to be changed for future climate conditions is here presented. Due to the small size of this phenomena, it cannot be studied with General Circulation Models (GCMs) so RCMs are used. First, ten RCMs simulations with roughly 25 km resolution, forced with ERA40 reanalysis ( ) are studied and compared with the limited available observational information about medicanes. Simulated medicanes are obtained using a cyclone detection method based on surface pressure and wind speed at higher levels, and then the vertical structure is analysed in terms of the thermal wind and the asymmetry of the structure. Results indicate that almost all the RCMs are able to reproduce the frequency of medicanes (around 1-2 per year) observed, although in a more limited time period compared with the regional climate simulations. An important variability in the medicanes number, location and month of development is also obtained. Once this validation study is performed, those RCMs, forced by several GCMs, and covering up to 2050 for all of them and to 2100 in some combinations, are also analysed. A slight reduction (significant in more than the half of the cases) is obtained for most of the available GCM/RCM combination of simulations, specially up to At the same time, it seems that the strongest medicanes could be increased by the end of the XXIst century. Important differences are seen depending on the subregion (western, central or eastern), and also between the South and the North parts of the Mediterranean Sea. The number of events is larger in the North, over Ligurian, Tyrrhenian, Adriatic and Aegean Seas but in the future these zones will suffer a greater decrease of medicanes than in the South so the events will be sparced in a more uniform mode along the Mediterranean Sea. The more intense events will take place basically over the South of Mediterranean Sea. Keywords: Medicanes, climate change, Mediterranean cyclones, Regional Climate Models 121

132 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO 1. Introduction Mediterranean Sea is a region with a high density of cyclones (more than 1500 cyclones year-1 according to Campins et al., 2011), being related to heavy precipitations and strong winds events (Jansà et al., 2001; Bocheva et al., 2007; Trigo et al., 2000; Nissen et al., 2010). Mediterranean cyclones are usually responsible of great damages in Mediterranean islands and coastal zones, due to strong winds and floods. A better understanding and description of these extreme meteorological phenomena for present conditions, and also how is projected to change under climate change conditions can be crucial to understand and predict the response of the Mediterranean climate to global climate change during the current XXIst century. But not all the Mediterranean cyclones have the same characteristics, recent studies have focuses their efforts in detecting the presence of cyclones with tropical characteristics (axial symmetry and warm core) over the Mediterranean Sea, the so called Medicanes (Mediterranean Hurricanes). The singular situation of the Mediterranean Sea favours the development of tropical-like cyclones. Gibraltar Straight is the only communication with the Atlantic Ocean so the water exchange is very limited and the Mediterranean water temperature reaches very high values in September. Other special feature of the Mediterranean Sea is the surrounding orography: important high mountains as the Alps Chain or the Iberian Peninsula and the Sahara desert are located near the Mediterranean coast. Ulbrich et al. (2009) pointed out to this zone as one of the most favourable places over the North Hemisphere to develop tropical-like cyclones. A more detailed analysis showed that medicanes do not happen in any place of the Mediterranean Sea with the same probability. Thus, Miglietta et al. (2013) analysed fourteen medicanes, finding out two preferred regions of occurrence (Ionian Sea and Balearic Islands). From a modelling perspective, it would seem quite interesting the usage of General Circulation Models (GCMs) to analyse mediterranean features, and so medicanes. But this analysis is not possible due to their typical size related to the resolution of GCMs, as these medicanes have typically a diameter of less than 300 km (Walsh et al., 2014). Therefore, it seems to be necessary the use of higher resolution models, such as Regional Climate Models (RCMs). Chapter 14 of the last IPCC report (Christensen et al., 2013) pointed out the lack of extratropical cyclones studies and the important spread among RCMs, which is similar to the future changes expected or to the natural interannual variability. It is quite important also to notice that the use of a model ensemble would allow to having into account the high spread among climate models, avoiding a biases analysis, unlike other studies which uses only one 122

133 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO RCM, as Walsh et al. (2014). Gaertner et al. (2007) showed the risk of tropical cyclone development over the Mediterranean in September using an ensemble of nine RCMs from PRUDENCE European Project (Christensen and Christensen, 2007), with an horizontal resolution of 50 km2. That work analysed future climate comparing A2 scenario runs ( ) with control simulation ones ( ). In that study, high resolution allowed for the first time to detect the risk of tropical cyclone development over the Mediterranean Sea, and the need of higher resolution and a more complete ensemble of RCMs (larger number of RCMs, the use of more GCMs or different emission scenarios) was remarked. It is expected that the number of detected cyclones can be increased with higher resolution models, as the smaller cyclones could be missed if the used resolution is too small (Pinto et al., 2006). Giorgi and Lionello (2008) already showed the capability of PRUDENCE ensemble of RCMs to describe the main climate change features over the Mediterranean basin. ENSEMBLES European Project (Van der Linden and Mitchell, 2009) completes the analyses from PRUDENCE increasing the number of RCMs, the number of GCMs, and the resolution (25 km2). Also the simulated period was increased compared to PRUDENCE being or depending on the model. Some studies based on this ensemble of RCMs have already been focused on the main features over the Mediterranean Sea (Sanchez-Gomez et al., 2009, 2011; Boberg and Christensen, 2012). Some other ensemble of regional climate models have also been performed in the past years and are currently being developed, with special focus over the Mediterranean region (Gualdi et al., 2013; Jacob et al., 2014). Several works have focused their efforts to study medicanes in a limited area over the Mediterranean Sea (Lionello, 2005; Lionello et al., 2003), or in the study of some observed medicanes (Pytharoulis et al., 2000; Fita et al., 2007), and other studies use only one RCM (Lionello et al., 2008a; Walsh et al., 2014, Lionello et al., 2002). The present study aims at advance from the previous work of Gaertner et al. (2007), using here an ensemble of 25 km 2 horizontal resolution RCM simulations, nested in different GCMs, from ENSEMBLES project. The focus of the study is the analysis of medicanes over the whole Mediterranean Sea from that ensemble of RCMs. The cyclone detection method described by Picornell et al. (2001) has been applied to detect all the cyclones in the studied domain, following by the calculation of the Hart (2003) parameters to select the medicanes among all the detected cyclones, as already made by Gaertner et al. (2007). First, the ERA40 forced simulations for period is evaluated against the available observational information about medicanes to inspect the capability of these RCMs to describe the observed 123

134 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO medicanes. Then, both present ( ) and future climate periods (up to 2050 for all the simulations and to 2100 for some of them) are analysed. This study then pretends to provide a complete vision about this infrequent phenomenon, analysing frequency, intensity and regional and monthly distribution and the changes under future climate change conditions. Figure 6.1: Topography of the studied domain and subregions chosen to the study. 124

135 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO 2. Methodology Ten RCMs from ENSEMBLES European Project (see Table 6.1), with an horizontal resolution of 25 km2, nested in ERA40 (evaluation period) and in six different GCMs (scenarios) have been used. ERA40 simulations ( ) are compared with the list of medicanes studied in Miglietta et al. (2013) as the methodology to detect medicanes is very similar to the one applied in the present study. The ten RCMs have been forced with different GCMs (Table 6.1 show the different combinations) to simulate present climate (control run) from 1951 to 2000 and future climate (scenario run) from 2001 to 2050 (thirteen simulations) and from 2001 to 2100 (nine simulations). The months selected for the study have been from August to January, as most of the medicanes occur during last summer and autumn (Cavicchia and von Storch, 2012) with a peak activity in January (Cavicchia et al., 2014). The Mediterranean Sea has been divided in three zones (see Figure 6.1) in order to study medicanes regional distribution: West (longitude lower than 10º), Central (longitude larger than 10º and lower than 24º) and East (longitude larger than 24º). Those subregions are comparable to the ones described in Giorgi and Lionello (2008). 125

136 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO RCM (Institution) ERA40 RCA3 ALADIN CLM HadRM3Q16 HadRM3Q3 RACMO HIRHAM REMO RCA PROMES (C4I) (CNRM) (ETHZ) (HC) (HC) (KNMI) (METNO) (MPI) (SMHI) (UCLM) A2 A1B A1B A1B ECHAM A1B ARPEGE A1B A1B BCM GCM A1B A1B A1B HCQ0 A1B HCQ A1B A1B HCQ3 Table 6.1: Summary of models, institutions, periods and scenario used in the present study. 126

137 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO 2.I. Cyclone Detection Method The cyclone detection method described by Picornell et al. (2001) based on sea level pressure (SLP) has been used. August to January daily SLP is analysed to identify the minima values. A Cressman filter with a radius of 200 km is used (Sinclair, 1997) to smooth neighbouring SLP minima. A radius of 400 km around every SLP minimum is applied to filter weaker cyclones through a SLP gradient threshold, so the final analysed domain is smaller then the original available one. The cyclone tracks have been calculated using the horizontal wind at 700 hpa. A daily maximum surface wind filter has been applied, dismissing all the cyclones which daily maximum wind speed is less than 17,5 m s -1 in all the days of its life, as it is the threshold to distinguish between a tropical depression and a tropical storm (or the threshold between near gale force and gale force in Beaufort Scale). 2.II. Vertical Structure of the Cyclones (Medicanes classification) The geopotential fields between 900 and 300 hpa have been used to apply the cyclone phase space method (Hart, 2003) in order to select the medicanes among all the detected cyclones. Hart method permit to calculate three parameters: B, -V TU and -VTL. B parameter is a measurement of the symmetry or asymmetry of the cyclone; a value of B near zero indicates that the cyclone is symmetric, being 10 m a suitable threshold to select symmetric cyclones. -VTU (thermal wind upper) and -VTL (thermal wind lower) parameters determine the existence of a warm or cold core between 900 and 600 hpa (-VTL) and between 600 and 300 hpa (-V TU). To select tropical characteristics cyclones, warm cores are searched so -V TL and -VTU must both be positive (Hart, 2003). In this study all the cyclones with -V TL positive and -VTU greater than -10 m have been selected. B parameter has been calculated using a radius of 150 km but it has not been used to filter the cyclones. These thresholds are based in the results of Migletta et al. (2013). 127

138 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO 3. Results 3.I. RCM evaluation ( ) against available observations From a pure observational point of view, it is difficult to detect medicanes, as many of them do not touch the coastline and ships try to avoid them. With the use of satellites it has been possible to create a list of observed events but these events are very recent so it is impossible to know the medicanes frequency in the recent past. Some studies have fixed the frequency of the medicanes in one or two events by year (Romero and Emanuel, 2013; Cavicchia et al., 2014) or even less than one per year (Walsh et al., 2014; Tous and Romero, 2011). Frequency Distribution by Distribution by months (n/year) regions (%) (%) Mediterranean West Center East Aug Sep Oct Nov Dec Jan Miglietta 1,08 35, ,3 0,0 28,6 28,6 14,3 14,3 0,0 RCA3 0,74 27,6 51,7 20,7 0,0 3,5 13,8 27,6 27,6 27,6 ALADIN 0, ,4 21,6 0,0 5,4 18,9 21,6 29,7 24,3 CLM 10,21 33,2 43,2 23,6 3,5 8,5 15,6 19,6 24,9 27,9 HadRM3Q16 2,02 19,3 54,2 26,5 2,4 3,6 12,1 26,5 27,7 27,7 HadRM3Q3 2,24 29,3 43,5 27,2 1,1 7,6 10,9 27,2 28,3 25 RACMO 2,48 27, ,1 3,9 6,7 11, ,9 25 HIRHAM 1,56 24,6 47,5 27,9 1,6 4,9 11,5 21,3 32,8 27,9 REMO 1,67 27,7 44,6 27,7 3,1 0,00 9,2 30,8 23,1 33,8 RCA 0,67 42,3 42,3 15,4 0,00 0,00 19,2 30,8 23,1 26,9 PROMES 2,41 29,8 47,9 22,3 2,1 7, ,3 27,7 24,5 RCM Mean 2,49 30,4 45,6 24 2,6 6,5 13,9 23,3 26,5 27,2 Table 6.2: Medicanes annual frequency for Miglietta data, every RCM and RCMs mean for ERA40 period: total frequency, regional relative frequency and monthly relative frequency. In Miglietta et al. (2013) a list of the medicanes observed in different zones of the Mediterranean sea between 1999 and 2012 has been created and the difficulty of elaborating a medicanes catalogue has been expressed. Table 6.2 shows the medicanes frequency registered by such study and for the RCMs in the hole Mediterranean Sea, in the three different evaluated zones and also the monthly 128

139 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO frequency in the studied months (from August to January). The frequency of the medicanes studied by Miglietta et al. (2013) is 1,08 events year -1, agreeing with the mentioned frequencies calculated in other studies. According to Miglietta et al. (2013) the Central zone (between 10º and 24 º of longitude) is the region of the Mediterranean Sea with a larger number of detected medicanes, where half of them have been registered, while the Eastern zone has only 0,15 events year -1 (14,3 %). Walsh et al. (2014) also pointed out the central zone of the Mediterranean Sea as the zone with a mayor risk of medicanes development, but Cavicchia et al. (2014) indicated towards the western zone of the Mediterranean Sea as the zone with larger medicanes activity. According to Miglietta et al. (2013) the 85,7 % of the medicanes take place from September to December, and more than a half occur during the months of September and October. Nevertheless, Tous and Romero (2011) pointed to a different monthly distribution with the greater number of medicanes in December. The period to evaluate the ensemble of the ten ERA40-forced RCMs covers from 1961 to 2000, and only for the months from August to January. As the largest number of medicanes occurs during September and October, according to Miglietta et al. (2013), it is possible to compare the annual frequency with the results obtained by such study, taking into account that the number of simulated medicanes could be slightly greater if all the year is considered. Most of the regional models obtain a frequency of around 2 events by year, agreeing with previous studies (Romero and Emanuel, 2013; Cavicchia et al., 2014; Walsh et al., 2014; Tous and Romero, 2011). Only CLM presents a clear overestimation of that overall value, with more than 10 events by year. CLM is the model which simulates a larger number of cyclones in general, and more than 40 % of them are medicanes, while the medicanes percentage of the rest of the models is less than the 16 % of the cyclones. Regarding the distribution by regions, most of the models follow the observed distribution (with maximum value over the Central part of the Mediterranea Sea), except ALADIN, that detects the western zone as the one with larger number of medicanes, and RCA, with an identical number of events in the west and in the central zone. With respect to the time (monthly) distribution, most of the RCMs follow a similar behaviour, with the highest frequency in November, December and January, showing a more similar distribution to Tous and Romero (2011) than to Miglietta et al. (2013) medicanes results. Although Tous and Miglietta databases show a decreasing in the medicanes activity during January, Cavicchia et al. (2014) detects a peak activity in such month. The spread among observational results is then large enough to be able to give strong conclusions about the differences obtained with the RCMs. 129

140 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO As it has been said (section 2.II), the axial symmetry (B parameter) of the cyclones has not been used to filter the medicanes nevertheless eight from the ten analysed RCMs present more than the 93% of the medicanes with axial symmetry. CLM is the model with the less percentage of symmetric medicanes (79,40%) follow by RCA3 (86,21%). 3.II. Climate Change Projections (up to 2050 or 2100) Figure 6.2 shows the intensity extremes of the medicanes through the percentile 95 (p95, represented by colour bars) and the maximum value (represented by squares) of the daily maximum wind speed in the centre of the medicanes for every RCM and every period ( , and for some of the models). The figure shows a large variability among individual models, not only in the value of the p95, but also in the tendency for future climate. This variability makes it difficult to find a pattern among simulations, even nested in the same GCM. Nevertheless, some patterns can be seen among this spread results: CLM-HCQ0 (M3), HadRM3Q3HCQ3 (M5), HIRHAM-BCM (M7), HIRHAM-HCQ0 (M8) and RCA-HCQ3 (M12) are the models that show an increasing tendency on this maximum wind intensity (p95). The mean of the 13 RCMs exhibit also an increase of the p95 wind intensity of the medicanes in future climate conditions. As not all the RCMs have results for the last 50 years period this result could look misleading, so the mean have been also calculated using just the 9 RCMs with values for the three periods (going up to 2100). The result (not shown in the figure) is very similar, with a p95 of 26.06, and m s-1 for each period, showing an also clear rising tendency. Previous studies, using one only RCM, also conclude that the intensity of the medicanes (Cavicchia et al., 2014) or the Mediterranean cyclones (Lionello et al., 2002) will increase at the end of the century. 130

141 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO Figure 6.2: Wind Speed p95 (m s-1) for all the RCMs and RCMs mean (MEAN) for the periods (red bar), (green bar) and (blue bar). Squares represent maximum wind speed for each model and period. Note that some RCMs only have simulated until 2050 so there is no blue bar neither blue square. In the figure M1=RCA3-ECHAM, M2=ALADIN-ARPEGE, M3=CLMHCQ0, M4=HadRM3Q16-HCQ16, M5=HadRM3Q3-HCQ3, M6=RACMO-ECHAM, M7=HIRHAMBCM, M8=HIRHAM-HCQ0, M9=REMO-ECHAM, M10=RCA-BCM, M11=RCA-ECHAM, M12=RCAHCQ3 and M13=PROMES-HCQ0. Maximum wind speed has also been depicted in Figure 6.2 (represented by colour squares) for each RCM and each period. This result reveals that the RCMs nested in ECHAM-GCM (M1, M6, M9 and M11), together with HadRM3 models (M4 and M5) and UCLM-PROMES (M13) give smaller extreme wind values than the rest of the models. By other hand CLM-HCQ0 (M3) is the model with the greatest value of maximum wind speed. 131

142 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO Figure 6.3: Medicanes decadal frequency for each model for the whole simulated period ( ). Note that some models finish its simulations in Figure 6.3 shows the annual frequency of the medicanes (from August to January) from 1951 to 2100 for every RCM in the Mediterranean Sea, averaged over each decade. Although the figure reflects an important variability among the changes in the number of medicanes with time, a first overview seems to show a decrease in the number of such events along the XXI st century, for example, PROMES-HCQ0 or HadRM3Q16-HCQ16 simulate a decrease in the number of medicanes while RCAECHAM or RCA-HCQ3 do hardly show variation in the annual frequency along the years. A test to assessment the statistical significance of the slope of the annual frequency evolution (do not show in the figure) has been made, using annual values to adjust them to a simple linear regression. The result is that eleven of the thirteen simulations have a negative slope although this slope is significant at the 95 % confidence level only in seven cases. RCA-BCM and RCA-HCQ3 are the runs with positive slope, but only the case of RCA-BCM has a statistical significant tendency. These different results among models reinforces the need of considering an ensemble of RCMs when studying these processes. 132

143 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO Figure 6.4: Medicanes frequency monthly distribution for every RCM and RCM mean (dashed line) and for every period: (top), (middle) and (bottom). The monthly distribution of medicanes frequency (see Figure 6.4) indicates that the dispersion among models could decrease at the end of the century, although again an important variability among runs is showed. The figure do not show any relevant tendency in the monthly frequency of the medicanes. A Mann-Whitney test (von Storch and Zwiers, 1999) has been carried out for the assessment of the statistical significance of the differences between future and current climate as it was done by Lionello et al. (2008a), but due to the limited number of medicanes by month and by year, the test does not produce any significant results. Previous studies about Mediterranean cyclones using one only RCM obtained different results: Lionello (2005) concluded that there were not a clear tendency in the Gulf of Venezia, Lionello et al. (2008b) affirmed that the cyclone activity would decrease in November and 133

144 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO December studying the storm surges, Lionello et al. (2008a) found that the cyclone frequency would decrease in December and January and increase in September but these results are dependent on the zone of the Mediterranean and on the thresholds used to classify the cyclones. On the contrary, Lionello and Giorgi (2007) concluded that the cyclone activity would increase in winter. Here, as seen in Figure 6.4, any significant conclusion can be obtained as even the mean of the RCMs (dashed line in the figure) do not show any significant variation under climate change conditions. The total number of medicanes, divided by the available RCMs number (13 up to 2050 and 9 up to 2100), is represented in Figure 6.5, showing the regional distribution for each period. Left panel shows the distribution following the regions defined in Figure 6.1. The decrease in the number of medicanes at the end of the century is clearly depicted over the hole Mediterranean Sea, independent of the region. As Miglietta et al. (2013) have pointed out, the Central Zone of the Mediterranean Sea is the region which presents a larger number of medicanes and this result is going to last under climate change conditions. Although the regional distribution looks to be constant along the years, the central region is going to suffer a larger decrease than the others in future climate, resulting in the last period ( ) than only the 45% of the total number of medicanes will have place in such region versus the 62% in control run ( ). Right panel of Figure 6.5 shows a different regional distribution, dividing the Mediterranean Sea in two zones, northern and southern of 36ºN. The greater number of medicanes is gathered in the north of Mediterranean Sea, but, as it can be seen in the left panel, the number of events will decrease under climate change conditions. The south part of the Mediterranean Sea is the only place where the number of medicanes is not going to experiment any change in the future. According to Figure 6.5 (left and right panels) the regions with a larger number of medicanes in control run will suffer a more pronounced decrease in the future so the number of medicanes will seem to be distributed in a more uniform mode along the Mediterranean Sea at the end of the century. 134

145 6MEDICANES: DESCRIPCIÓN Y PROYECCIONES DE CAMBIO CLIMÁTICO Figure 6.5: Regional distribution of total medicanes number, divided by the 13 RCMs available up to 2050 and by 9 RCMs available up to 2100, in each region (left: west, central and east according to Figure 6.1; right: north, for latitude greater than 36ºN, and south, for latitude less than 36ºN) and each period: (red), (green) and (blue). In Figure 6.6 all the studied medicanes, no matter from which RCM simulation come from, with maximum daily value among the whole life of the medicane of surface wind speed larger than 25 m s -1 (threshold between strong gale force and storm force in Beaufort Scale) are represented. Green colour indicates maximum daily wind speed between 25 and 33 m s -1 (minimum threshold of Saffir-Simpson Hurricane Wind Scale) while red colour represents the more intense medicanes (maximum daily wind speed larger than 33 m s-1). The figure allows to inspect the regional spatial distribution of the medicanes and changes in the number and intensity in future climate conditions. The zones with the major number of medicanes are Ligurian Sea (North of Corsica), Tyrrhenian Sea, Adriatic Sea and Aegean Sea in the first ( ) and second ( ) modelled periods. But in the far future climate ( ) the number of medicanes seem to be distributed all along the Mediterranean Sea in a more homogeneous way, together with the overall reduction already mentioned for this period, as it has been seen in Figure 6.5. As it was also seen in Figure 6.2, the intensity of the strongest medicanes would be increased in future climate (more red points on that last period). This Figure 6.6 allows to detect that the more intense medicanes would be formed in the south and east of the Mediterranean Sea, in the region where the number of events are more scarce for any of the periods analysed. 135