ORIGINAL PAPER

Global warming impact on the dominant precipitationprocesses in the Middle East

Jason P. Evans

Received: 21 May 2008 /Accepted: 5 May 2009 /Published online: 19 May 2009# Springer-Verlag 2009

Abstract In this study, the ability of a regional climatemodel, based on MM5, to simulate the climate of theMiddle East at the beginning of the twenty-first century isassessed. The model is then used to simulate the changesdue to global warming over the twenty-first century. Theregional climate model displays a negative bias in temper-ature throughout the year and over most of the domain. Itdoes a good job of simulating the precipitation for most ofthe domain, though it performs relatively poorly over thesoutheast Black Sea and southwest Caspian Sea. Usingboundary conditions obtained from CCSM3, the model wasrun for the first and last 5 years of the twenty-first century.The results show widespread warming, with a maximum of~10 K in interior Iran during summer. It also found somecooling in the southeast Black Sea region during spring andsummer that is related to increases in snowfall in the region,a longer snowmelt season, and generally higher soilmoisture and latent heating through the summer. Theresults also show widespread decreases in precipitationover the eastern Mediterranean and Turkey. Precipitationincreases were found over the southeast Black Sea,southwest Caspian Sea, and Zagros mountain regionsduring all seasons except summer, while the Saudi desertregion receives increases during summer and autumn.Changes in the dominant precipitation-triggering mecha-nisms were also investigated. The general trend in thedominant mechanism reflects a change away from the directdependence on storm tracks and towards greater precipita-tion triggering by upslope flow of moist air masses. The

increase in precipitation in the Saudi desert region istriggered by changes in atmospheric stability brought aboutby the intrusion of the intertropical convergence zone intothe southernmost portion of the domain.

1 Introduction

The landscape of the Middle East has been altered byhuman activity for most of the Holocene. The rate of thesemodifications has accelerated in the last century, and today,rapid population growth, political conflict, and waterscarcity are common throughout the area. All of thesefactors increase the region’s vulnerability to potentiallynegative impacts of climate change while decreasing thelikelihood of successful region-wide adaptation strategiesemerging.

While much of the region has a Mediterranean-typeclimate, the region spans many climate zones from thehyper-arid deserts of Saudi Arabia to the cool highlandclimates of various mountain ranges such as the Taurus andZagros mountains. The region is surrounded by large inlandwater bodies with extensive areas being impacted bycoastal effects. In several places, such as the southerncoasts of the Black and Caspian Seas, there are largemountains in close proximity to coastlines providing a highlikelihood of upslope flow-driven precipitation. The com-plex relationship between landscape and climate means thatdifferent precipitation processes dominate in different partsof the region (Evans et al. 2004). Previously, Evans et al.(2004) used a regional climate model to identify thedominant precipitation-triggering processes in areas of localprecipitation extrema. In this study, a similar technique isused to identify potential changes in the dominant mech-anisms caused by global warming.

Theor Appl Climatol (2010) 99:389–402DOI 10.1007/s00704-009-0151-8

J. P. Evans (*)Climate Change Research Centre, Faculty of Science,University of New South Wales,Sydney, NSW 2052, Australiae-mail: jason.evans@unsw.edu.au

Simulations of the global climate, including thoseperformed for the recent Intergovernmental Panel onClimate Change Fourth Assessment Report (IPCC AR4)report, are conducted with horizontal resolutions thatpreclude the simulation of small-scale circulation featuresthat are an important part of the regional climate. The use ofa regional climate model (RCM) provides the ability toresolve detailed features of topography and the landsurface, as well as to separate different zones of precipita-tion extrema. RCMs have become more widely used duringthe last decade, and reviews of previous regional modelingstudies and approaches can be found in McGregor (1997)and Giorgi and Mearns (1999).

Perhaps due to the relative data sparsity of the MiddleEast, a relatively small amount of research focusing onmodern regional climatic phenomena exists. Several studiesof the medium term change in local climate regime basedon ground station data have been performed for severalsubregions (Ben-Gai et al. 1998; Ben-Gai et al. 1999; Smallet al. 2001; Cohen et al. 2002; Kostopoulou and Jones2005) as well as the region as a whole (Nasrallah andBalling 1995; Zhang et al. 2005). Global climate model(GCM), observational analyses (model with observationsassimilated), and ground-based data have been used toinvestigate large-scale atmospheric phenomena and tele-connections influencing the region (Reddaway and Bigg,1996; Saaroni et al. 1998; Nazemosadat and Cordery 2000;Saaroni and Ziv, 2000; Kutiel and Benaroch 2002; Paz etal. 2003; Krichak and Alpert 2005; Turkes and Erlat 2005;Chakraborty et al. 2006; Ghasemi and Khalili 2006). Somehigher resolution studies investigating weather events usingregional models have also been carried out (Alpert et al.1997; Krichak et al. 2000). RCMs have also been used toinvestigate the impact of using “correct” land coverconditions, as defined by satellite data, on the regionalclimate (Zaitchik et al. 2005) as well as to investigateregional drought feedbacks (Zaitchik et al. 2007a), moun-tain effects (Zaitchik et al. 2007b), the impact of irrigation(Evans and Zaitchik 2008), and the influence of watervapor source regions and transport on precipitation (Evansand Smith 2006).

Growing acceptance of the reality of global warming hasrecently led to an increase in the publication of studiesrelated to global warming impacts. Many GCM simulationswere performed as part of the IPCC AR4, and the resultingmodel output has been made available through the EarthSystem Grid (http://www.earthsystemgrid.org/home/home.htm) for use in impact studies. Some of these studies focuson global changes in various phenomena that are importantfor the Middle East such as changes in storm tracks(Bengtsson et al. 2006; Lambert and Fyfe 2006), temper-ature (Min and Hense 2006), and drought (Wang 2005).Evans (2009) presents the changes in climate predicted for

the Middle East over the twenty-first century by anensemble of 18 GCMs using the Special Report onEmission Scenarios (SRES) A2 emission scenario whichis the scenario closest to a “business as usual” scenario inthe SRES family. That study found an increase intemperature for the region of almost 4 K by late century,along with significant changes in precipitation that includea decrease in an area stretching from Turkey and theEastern Mediterranean across to Northeastern Iran and anincrease over much of the Persian Gulf and Saudi Arabia.Various aspects of the performance of CCSM3, the GCMused to provide boundary conditions in this study, underdifferent emission scenarios can be found in Meehl et al.(2006).

This paper uses an RCM to analyze the impact ofglobal warming on the precipitation triggering processesin the Middle East. Using a technique similar to thatgiven in Evans et al. (2004) and simulations of futureclimate under increasing greenhouse gas conditions, thispaper investigates changes in precipitation and precipi-tation-triggering mechanisms in the region. The climatemodels used are described in Section 2 and the simu-lations are evaluated against current observations inSection 3. Section 4 presents the changes due to globalwarming, with the main conclusions being summarized inSection 5.

2 Climate model and simulation description

This study examines the changes in local precipitationprocesses in the Middle East caused by global warming. Toaccomplish this, the output from a GCM simulation isdownscaled using an RCM and the output from these RCMsimulations is used to investigate locally dominantprecipitation-triggering mechanisms.

2.1 Global climate model (CCSM3)

The community climate system model version 3 (CCSM3)is a coupled climate model developed and maintainedthrough the National Center for Atmospheric Research(NCAR) in the USA and described in Collins et al. (2006).The model couples components that model the atmosphere,ocean, sea ice, and land surface. The simulation used in thisstudy came from a T85 version of CCSM3 and is describedin detail in Meehl et al. (2006). In short, the model hasatmospheric grid points roughly every 1.4° latitude andlongitude and 26 levels in the vertical. The ocean ismodeled on a nominal 1° grid with 40 levels in the vertical.No flux adjustment is used. Atmospheric composition is inaccordance with the SRES A2 emission scenario for thetwenty-first century.

390 J.P. Evans

2.2 Regional climate model (MM5-Noah)

The Pennsylvania State University/NCAR mesoscale mod-eling system MM5 is described in Dudhia (1993) and Grellet al. (1994). MM5 is a limited-area non-hydrostatic modelthat uses a terrain-following vertical coordinate system. Ithas two-way nesting capabilities and flexible physicsoptions. In this study, MM5 was implemented with theReisner mixed-phase explicit moisture scheme (Reisner etal. 1998), the MRF planetary boundary layer scheme (Hongand Pan 1996), the rapid radiative transfer model radiationscheme (Mlawer et al. 1997), and the Grell scheme forconvective precipitation (Grell et al. 1994).

MM5 is operationally linked with the Noah land surfacemodel (LSM). Noah is a direct descendent of the Oregon StateUniversity LSM (Mahrt and Ek 1984; Mahrt and Pan 1984;Pan and Mahrt 1987), a sophisticated land surface model thathas been extensively validated in both coupled anduncoupled studies (Chen and Mitchell 1999; Chen andDudhia 2001). The Noah LSM simulates soil moisture, soiltemperature, skin temperature, snowpack depth and waterequivalent, canopy water content, and the energy flux andwater flux terms of the surface energy balance and surfacewater balance. In its MM5-coupled form, Noah has adiurnally dependent Penman potential evaporation (Mahrtand Ek 1984), a four-layer soil model (Mahrt and Pan 1984),a primitive canopy model (Pan and Mahrt 1987), modestlycomplex canopy resistance (Jacquemin and Noilhan 1990),and a surface runoff scheme (Schaake et al. 1996).

MM5 has been applied successfully at grid cell reso-lutions ranging from greater than 100 km to less than 1 kmand is used for both weather forecasts and climate research(Zaitchik et al. 2005; Evans et al. 2005). Here, we apply themodel at 27-km horizontal resolution and 23 vertical levelsover a domain which includes much of the Middle East andthe surrounding water bodies. Figure 1 shows the model

domain excluding the rows and columns which are directlyinfluenced by the boundary conditions. Previously, RCMsimulations over the same domain were performed usingRegCM2 (Evans et al. 2004) and MM5 (Evans and Smith2006) for the early 1990s. These simulations showedreasonable agreement with observations, with the MM5performing best.

In this study, three MM5 simulations are discussed.Two of these simulations cover the period 2000 through2004. One of these runs uses initial and boundaryconditions from the National Centers for EnvironmentalPrediction (NCEP)/NCAR reanalysis (NNRP; Kistler etal. 2001) and the other from the CCSM3 simulationperformed for the IPCC AR4 using the SRES A2 emissionscenario. The third simulation covers the period 2095through 2099, with initial and boundary conditionscoming from the same CCSM3 SRES A2 simulation.The SRES A2 scenario produces the largest increase ingreenhouse gases, and hence warming, by the end of thecentury of the commonly reported emission scenarios.Atmospheric greenhouse gas concentrations in recentyears have exceeded all the emission scenarios in theSRES family; hence, this scenario is chosen as it is closerto the likely future emissions growth without a determinedinternational effort to reduce them.

In each case, the first year of the MM5 simulation isconsidered “spin-up,” and relevant statistics are calculatedusing the remaining 4 years of the model run. Givenunlimited computing resources, an ensemble of longersimulations would have been preferable. However, theresources available limited the length of simulation to5 years. Examination of the annual precipitation from theCCSM3 simulation through the twenty-first century showsthat the first and last 5 years do not represent anomalousperiods, being within 10% of the respective 20-year meansand even closer to the 10-year means.

Fig. 1 RCM domain including topography from GTOPO (a), MM5 (b), and CCSM3 (c)

Global warming impact on the precipitation processes in the Middle East 391

Examination of the CCSM3 simulation-derived winterNorth Atlantic Oscillation, using the sea level pressure atthe relevant grid points, shows these last 5 years to berepresentative of the last 20 years of simulation in terms ofthe mean value, 0.44 and 0.34 respectively, and indistribution of years with some strong NAO positive yearsand some weak NAO negative years. Similarly, the first5 years are representative of the first 20 years with meanvalues of −0.78 and −0.70, respectively. Thus, the changein NAO simulated by the 5-year RCM simulations iscomparable to the change computed from longer periods ofthe CCSM simulation. It must be remembered that thesesimulations represent one possible response to globalwarming, thus allowing the investigation of potentialchanges in precipitation-triggering mechanisms. They arenot considered a definitive prediction of the future change.

3 Model evaluation

Performance of the models is evaluated against observa-tions during the first 5 years of the twenty-first century inthis section. Both the global and regional model outputs areevaluated. It is important to note that the RCM is confinedto be exactly the same as the global model at the RCMboundary. Far from the boundary, the RCM can differ fromthe global model due to small-scale topographic featuresand related circulations that are not resolved by the globalmodel. Large-scale features in the RCM are heavilyinfluenced by the driving global model and are unable todiffer substantially. Thus, the quality of an RCM simulationis necessarily dependent on the quality of the GCMsimulation providing the initial and boundary conditions.

3.1 Observational data

Surface climatology was established using data availablefrom the United Nations Food and Agriculture Organization(FAO; FAOCLIM version 1.2). This dataset includesmonthly values for standard climate quantities, averagedover a period from 1940 to 1970. There are almost 1,000stations in our area of interest. Data from these stationswere used to create monthly temperature and precipitationclimatologies interpolated to the MM5 27-km grid. In orderto account for the large changes in interstation distances, aCRESSMAN technique using a variable radius of influencewas used. The interpolation scheme includes a −5°C/kmtemperature lapse rate correction, but no correspondingcorrection to precipitation. The spatial distribution ofprecipitation shows a strong increase of precipitationnorthward and outward from the deserts of Saudi Arabia,eastern Jordan, western Iraq and southeastern Syria, towardthe mountains, and the Mediterranean Sea.

The FAO dataset used above provides good spatialcoverage, but has the disadvantage of being collected sometime earlier than the current experiments as well asproviding no interannual information. To address theseissues, data from ground stations were extracted from theglobal summary of monthly observations collected at theClimate Prediction Center (CPC) of the NCEP. There werearound 1,000 stations reporting in and nearby to the regionbetween 2000 and 2005, though there was no month whenall stations reported, with at most 539 and as few as 373stations reporting in a month. On average, the reportingstation density over the domain means each stationrepresents ~26,000 km2. While this dataset offers monthlyvalues for the entire period modeled, the station density isaround half that offered by the FAO dataset. In particular,almost no observations from Iraq are present in the CPCdata, causing significant errors in the climatology where theprecipitation maximum in northern Iraq captured by theFAO dataset is missing from the CPC data. This is also seenin the temperature bias where areas of Iraq display biasesgreater than 1 K. It should be noted that some of thedifference between these datasets is due to the fact that theycover different periods; however, the difference in stationdensity and distribution remains the major cause.

3.2 Statistical measures

Many different statistical measures have been used previ-ously to test the performance of climate models quantita-tively. Willmott et al. (1985) and Legates and McCabe(1999) provide an analyses of the suitability of several ofthese measures as well as suggest some of their own. In thispaper, the model performance is evaluated against obser-vations using several statistics including the bias

Bias ¼ M� OFAO ð1Þ

where M is the mean of the modeled values and OFAO is themean of the FAO climatology. The root mean square error(RMSE) is given by:

RMSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

N

XNi¼1

Oi �Mið Þ2vuut ð2Þ

where N is the number of observed, O, and modeled, M,values being compared. Here, N is the number of MM5 gridcells in the domain. Climatological values are used incalculating RMSE.

In order to quantitatively evaluate the spatial agreementbetween model and observations, Walsh and McGregor(1997) define the pattern correlation (3), ρp, betweenobserved and simulated fields simply as the correlation of

392 J.P. Evans

a series of data points from the observed field withcorresponding values from the modeled field at a fixedtime; in this study, seasonal means are used.

rp ¼P

Oi � O� �

Mi �M� �

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPOi � O

� �2r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP

Mi �M� �2q : ð3Þ

The anomaly correlation, ρa, is similar to the patterncorrelation except that fields are replaced by anomaliesfrom climatology. The anomaly correlation provides a morerigorous test of whether the model can capture the spatialpattern of interannual variations. Here, the sums arecalculated over the number of MM5 grid cells within thedomain.

Finally, since the region in question is quite complexwith many climate zones and precipitation regimes present,we have divided the domain into precipitation extremasubregions for further analyses (Fig. 2). These regionsconsist of the southeast Black Sea coast, the southwestCaspian Sea coast, the eastern Mediterranean coast, easternpart of the Fertile Crescent (essentially the headwaters ofthe Tigris River), the southern Zagros Mountains, and theSaudi desert. Each of these areas demonstrates substantialclimatological differences and thus provides quite a strongtest of the regional models abilities.

3.3 Model evaluation

The bias and RMSE for seasonal mean temperature andprecipitation are given in Tables 1 and 3. The non-parametric paired Wilcoxon signed-rank test (Wilcoxon

1945) was used to test the statistical significance of theseresults. This test does not assume that the variables arenormally distributed. In this test, the null hypothesis is thatthe distribution of the differences between the two datasetsis symmetric around zero, that is, the datasets come fromthe same distribution. Using the climatology from the FAOdataset as the baseline, an observational bias and RMSEcan be calculated between it and the CPC observationaldata. Similar bias and RMSE calculations are made for eachof the present day model results. These model bias andRMSE are then tested for significance against the observa-tional series. For all seasons and all models, the nullhypothesis could not be rejected at the 95% significancelevel, that is, the modeled seasonal temperature andprecipitation could not be distinguished from observationswithin significance bounds.

The bias and RMSE for seasonal mean temperature isgiven in Table 1. All the models consistently display anegative temperature bias except for CCSM which has amodest positive bias during summer. In general, the MM5cold bias is larger than that of the driving global model,particularly in autumn. Spatially, the NNRP bias isdistributed throughout the region with the largest biases,both positive and negative, occurring in the mountainousregions of Turkey and Iran. The spatial bias pattern forMM5/NNRP resembles that of NNRP, with the largestdifference occurring in autumn when it has a consistentnegative bias over the mountainous regions, while NNRPhas interspersed zones of positive bias as well. Thesedifferences can be related to differences in topography asseen in Fig. 1. The NNRP biases tend to cancel each otherout, while the MM5/NNRP biases are more consistent, andthis is reflected in the lower RMSEs for MM5/NNRP in allseasons but autumn.

CCSM is not constrained by observations like the NNRPis. It represents a possible realization of the weather giventhe global “boundary” conditions, largely the solar insola-tion and atmospheric composition; as such, it cannot beexpected to perform as well as the NNRP when evaluatedagainst observations. This also applies to MM5 when it isdriven at the boundary by the CCSM. Given this, theCCSM performs remarkably well, with both the bias andRMSE being similar to those of the NNRP. As with MM5/NNRP, the MM5/CCSM tends to have a more negative biasthan its driving model, especially in autumn. In terms of theRMSE, the CCSM tends to perform slightly worse than theNNRP, within 0.5 K, while the MM5/CCSM tends toperform slightly worse again.

The pattern and anomaly correlations, given by Eq. 3, ofseasonal mean temperature are reported in Table 2. MM5/NNRP is consistently better than NNRP alone, as expectedgiven MM5s higher spatial resolution. A larger improve-ment is seen in the anomaly correlation than in the pattern

Fig. 2 Study domain showing the focus regions: 1 Southeast BlackSea coast; 2 southwest Caspian Sea coast; 3 eastern Mediterranean Seacoast; 4 eastern Fertile Crescent; 5 Zagros Mountains; 6 Saudi desert

Global warming impact on the precipitation processes in the Middle East 393

correlation, that is, once the model biases are removed,downscaling with MM5 produces a larger improvement inthe temperature patterns than that simulated by NNRPalone. CCSM produces pattern and anomaly correlationssimilar to those of the NNRP and does so without theobservational nudging. MM5/CCSM also produces similarcorrelations except for the transition season anomalycorrelations where it performs worse.

Precipitation displays a high degree of spatial variability,with some coastal areas receiving over 1,000 mm annually,while the southern desert regions receive less than 200 mm.Capturing this variability is challenging for both the observa-tional network and models. The CPC observational datasethas relatively low density in this region, with each stationrepresenting, on average, more than 26,000 km2. Of course,the stations are not evenly distributed, with stationsrepresenting an area of just over 5,000 km2 on theMediterranean coast but more than 75,000 km2 in the Saudidesert. Practically no observations exist over the waterbodies or in Iraq during 2000 through 2004; hence, theseareas are excluded from the statistics below. The seasonalprecipitation bias and RMSE are presented in Table 3. Interms of bias, no model is consistently better than the othermodels. The global models, NNRP and CCSM, do wellduring winter when the precipitation is dominated by large-scale systems. MM5 tends to produce less precipitation thanits driving model. In terms of RMSE, the MM5/NNRPperforms the best. Again, considering the lack of anyobservational nudging in CCSM, both the CCSM andMM5/CCSM perform quite well.

The seasonal precipitation pattern and anomaly correla-tions can be found in Table 4. MM5/NNRP tends to

produce the best pattern correlations, significantly improv-ing the spatial distribution produced by NNRP. CCSM,however, does particularly well during spring in allstatistical measures. Once model biases are removed, theanomaly correlation shows that MM5/NNRP produces aneven greater improvement over NNRP than was seen in thepattern correlation.

The monthly mean precipitation for each of thesubregions is shown in Fig. 3. The annual cycle ofprecipitation is captured well by MM5/NNRP in theMediterranean Sea, Fertile Crescent, and Zagros Mountainssubregions. The general cycle is also captured in theCaspian Sea region, though the summer dryness is over-estimated and the autumn peak is underestimated. Thisunderestimation of precipitation may be related to poorlymodeled topography and coastlines, which mean themodels fail to capture sea breeze-induced low-level flowreversal that can cause upslope flow triggering precipita-tion. The Black Sea subregion contains a relatively smallseasonal cycle that is not reproduced by any of the models.While the timing of the precipitation is not well constrainedby the observational datasets, it is clear that the maximumoccurs during late autumn or winter. NNRP simulates aprecipitation cycle almost 6 months out of phase with thatobserved. The other models are better able to capture thetiming of this maximum, though they simulate significantlylower summer precipitation than is observed. These differ-ences are almost certainly related to the differences intopography shown in Fig. 1. Firstly, no models capture thetall, narrow mountain range along the southern shore of theBlack Sea in this region, which can enhance the formationof a sea breeze and may be responsible for a significant

Table 2 Pattern and anomaly correlation for temperature (2001–2004)

Season Pattern correlation Anomaly correlation

MM5/NNRP NNRP MM5/CCSM CCSM MM5/NNRP NNRP MM5/CCSM CCSM

DJF 0.95 0.91 0.89 0.87 0.31 0.18 0.14 0.16

MAM 0.97 0.92 0.90 0.90 0.42 0.34 −0.08 0.21

JJA 0.94 0.86 0.78 0.79 0.34 0.18 0.17 0.18

SON 0.96 0.91 0.88 0.89 0.27 0.18 −0.09 0.05

Table 1 Bias and RMSE for temperature (2001–2004)

Season Bias (K) RMSE (K)

MM5/NNRP NNRP MM5/CCSM CCSM MM5/NNRP NNRP MM5/CCSM CCSM

DJF −2.0 −2.3 −1.5 −0.9 2.9 3.7 3.4 3.4

MAM −1.9 −1.9 −3.0 −2.7 2.5 3.2 4.1 3.8

JJA −1.5 −0.4 −0.2 0.9 2.7 2.8 3.6 3.3

SON −2.7 −1.0 −2.6 −1.3 3.2 2.7 3.9 3.1

394 J.P. Evans

proportion of the summer precipitation. Secondly, theglobal models fail to separate the Caucuses Mountainsand hence do not have the wide low valley over Georgiaand Azerbaijan that can significantly alter low level windpatterns.

4 Global warming predictions

This section focuses on changes predicted to happen overthe coming century due to global warming. In particular,the focus is on the results obtained from the early and latetwenty-first century MM5/CCSM runs. The relative perfor-mance of the model in various parts of the domain shouldbe kept in mind when translating these results to the realworld. Evans (2009) presents predictions made by anensemble of GCMs for this domain and provides contextwithin which these results can be interpreted.

As can be seen in Fig. 4, the temperature is not predictedto increase uniformly across the domain; much of thisspatial distribution can be explained by changes in moistureavailability. Winter demonstrates the smallest changes, withmuch of the domain temperature increasing by less than2 K, which is not significant at the 0.99 level using aStudent’s t test. There are, however, patches within this areaof small change where the change is significant at the 0.9level. Most of the change outside this area, including overthe water bodies, is significant at the 0.99 level. Only thewesternmost part of the domain experiences increases ofmore than 4 K. All the water bodies experience similarsurface temperature increases of around 3 K.

A similar pattern is present in spring, with an extradegree of warming and substantial reduction in the areadisplaying an insignificant change at the 0.9 level. Increasesin temperature along the Mediterranean and Red Sea coasts,as well as a decrease along the east Black Sea coast, arenow significant changes at the 0.99 level. The coastalwarming is related to increased evapotranspiration anddecreased precipitation producing more rapid drying of theland and hence moving the energy balance towards sensibleheat. The decreasing temperatures along the Black Seacoast are due to the increased precipitation in winter (andspring), leading to greater snow accumulation and a longersnowmelt season.

The largest temperature increases are seen in summerover most of the domain, with an increase of around 6 Kbeing common. This increase is significant at the 0.99 levelfor the entire domain except the Anatolia plateau in Turkeyand the Caucuses Mountains region. The largest tempera-ture increases of around 10 K are seen over the Iranianplateau, while the largest temperature decreases (around4 K) are predicted to occur along the southeast Black Seacoast. This temperature decrease is again related to anincreased snowpack which is accumulated over winter andpersists through the spring. In these model simulations,there is such a large increase in snowfall that it melts late inthe spring and significant soil moisture persists throughsummer despite the generally warmer conditions. TheIranian plateau is quite dry to begin with, and hence,increases in temperature cannot be moderated by increasesin evapotranspiration, and instead all the energy isconverted to sensible heat.

Table 4 Pattern and anomaly correlation for precipitation (2001–2004)

Season Pattern correlation Anomaly correlation

MM5/NNRP NNRP MM5/CCSM CCSM MM5/NNRP NNRP MM5/CCSM CCSM

DJF 0.74 0.71 0.46 0.57 0.47 0.31 0.27 0.28

MAM 0.69 0.59 0.48 0.71 0.33 0.08 0.30 0.44

JJA 0.80 0.78 0.12 0.13 0.65 0.14 0.77 0.76

SON 0.86 0.62 0.46 0.56 0.49 0.07 0.20 0.11

Table 3 Bias and RMSE for precipitation (2001–2004)

Season Bias (mm) RMSE (mm)

MM5/NNRP NNRP MM5/CCSM CCSM MM5/NNRP NNRP MM5/CCSM CCSM

DJF 16 0 −37 −6 74 83 112 97

MAM 17 42 −6 2 57 90 70 51

JJA −16 21 −18 20 41 63 58 62

SON −23 −11 −8 22 64 93 106 100

Global warming impact on the precipitation processes in the Middle East 395

While not as large a change as summer, the majority of thedomain experiences temperature increases significant at the0.99 level. The Iranian plateau remains the zone with the largestincreases in temperature during autumn. An increase in

precipitation over the Zagros Mountains alleviates this warmanomaly over western Iran. The southern portion of the domainwarms more than the northern section, except for the northeastcorner which experiences significant warming in all seasons.

Fig. 3 Monthly averaged pre-cipitation for each subregion(2001–2004)

Fig. 4 MM5/CCSM modeledchange in seasonal mean tem-perature (2095–2000). The 0.9(0.99) significance levels is in-dicated by the thin (thick)dotted line

396 J.P. Evans

Figure 5 shows the seasonal change in precipitation.Most of the increases seen are substantial, being around100% of the current precipitation, and even some of thedecreases approach 100%. The largest decreases occur overthe eastern Mediterranean and Turkey in all seasons, inagreement with the GCM ensemble discussed in Evans(2009). In these MM5 simulations, most of this areaexperiences a decrease that is significant at the 0.9 levelduring winter, while only parts of the area is decreasingsignificantly during other seasons. During winter, severalareas experience precipitation increases that are significantat the 0.99 level in these simulations. Of note are theeastern Black Sea across to the northern Caspian Sea and anarea of Iran south of the Caspian Sea across to southernIraq. The GCM ensemble has the greatest uncertaintyassociated with the southeastern Black Sea and ZagrosMountains. Thus, while this model predicts significantincreases in precipitation in these zones, not all GCMs did,and a higher level of uncertainty is associated with thesechanges. During autumn, the GCM ensemble predicts asignificant increase in precipitation in the Zagros Moun-tains region in agreement with MM5, although only aportion of this area is significant at the 0.9 level. Another

feature of the change in precipitation present in the regionalclimate model and the GCM ensemble is an increase overSaudi Arabia during summer and autumn which isassociated with the intertropical convergence zone (ITCZ)having moved northward, into the domain, during thecentury.

The monthly precipitation climatology for early and latetwenty-first century conditions, for each subregion, can beseen in Fig. 6. The Black Sea region experiences a largeincrease in winter precipitation, with around five times asmuch precipitation falling by late twenty-first century. Themodel underestimated the present-day precipitation in thisarea by around half (Fig. 3), so this change represents asubstantial increase over present-day observed precipitationof around 2.5 times. This is by far the largest changepredicted by the model for any subregion, and it must beremembered that it is an area of strong disagreement amongGCMs (see Evans 2009) and therefore high uncertainty.The Caspian Sea subregion also shows large increases,particularly in winter. While substantial increases inprecipitation are seen in spring and autumn, the precipita-tion maximum has moved to winter by late century. TheMediterranean subregion experiences an overall decrease in

Fig. 5 MM5/CCSM modeledchange in seasonal mean pre-cipitation (2095–2000). The0.9 (0.99) significance level isindicated by the thin (thick)dotted line

Global warming impact on the precipitation processes in the Middle East 397

precipitation, particularly during spring. The FertileCrescent subregion changes little except for an increasein the late summer/early autumn precipitation whichcurrently is the driest time of year. In the ZagrosMountains subregion, the summer remains dry, butincreases are seen at other times of year, particularlyduring late autumn/early winter. The Saudi desertsubregion is predicted to have significant precipitationduring the summer/autumn transition. In each case, thechanges predicted could have serious implications foragriculture and urban water supply.

Compared to the climate change predictions given bythe ensemble of GCM simulations using the SRES A2emissions scenario (Evans 2009), this RCM-predictedchange displays much more spatial variation within thedomain. While domain-wide mean changes are similar,significant differences occur in some locations. The absenceof the Caucuses Mountains in the GCMs means that theRCM produces significantly different results in areasaffected by them, including the Eastern Black Sea. Here,the RCM produces more precipitation, much of which issnow, that produces a slight cooling compared to the overallwarming throughout the domain. Another significantdifference is the large area of decreasing precipitationsimulated by the GCMs that extends over the EasternMediterranean, Turkey, Syria, Northern Iraq, and North-eastern Iran. In the RCM, this area of decrease is confinedto the Eastern Mediterranean and Turkey, while NorthernIraq and Northeastern Iran experience an increase inprecipitation due to water vapor transported into the areaby the Zagros mountain barrier jet (Evans 2008). Thisphenomenon occurs on a scale too small to be resolved bythe GCMs.

4.1 Monthly indicators of precipitation processes

This section investigates the extent to which storm tracklocation, topography, and atmospheric stability explain themodeled precipitation using a methodology based on thatfound in Evans et al. (2004). While the climate changepredicted by these simulations must be interpreted withcaution, they come from only a single possible realizationafter all, they do provide the opportunity to examine possiblechanges in precipitation-triggering mechanisms that could bebrought about by global warming. The results presented inthis section are based purely on model output and hence anyconnections with reality are only cautiously made given thelimitations of the model simulation as evaluated in previoussections. To identify the intensity of passing storms, twoproxies are defined on a monthly basis. The first is thestandard deviation of the daily 500-hPa geopotential height(sdgp), while the second is the standard deviation of the daily500-hPa kinetic energy (sdke). The kinetic energy present isobtained through the magnitude of the horizontal wind field(KE=1/2 mv2), which is then band-pass-filtered to isolateevents with duration between 2 and 7 days. A symmetric setof weights similar to that found in Trenberth (1991) is usedto perform the filtering. The standard deviation of this band-pass-filtered KE is equivalent to the eddy kinetic energy.These storm track indices correlate well with each otherthrough most of the domain. To identify the influence oftopography, we define a topographic index, ψ, as:

y ¼~lwv �rh ð4Þ

where~lwv is the flux of water vapor and h is the topographicelevation. By setting any negative values to zero, this index

Fig. 6 MM5/CCSM modeledmonthly mean precipitation forearly and late twenty-firstcentury

398 J.P. Evans

provides a measure of the flux of water vapor movingupslope and ignores the water vapor moving downslope. Anatmospheric stability index, φ, is calculated as the differencebetween the equivalent potential temperature of a near-surface (lowest sigma level) air parcel, qse, and the equivalentpotential temperature of an air parcel at 500 hPa, q500e .

ϕ ¼ qse � q5e00 ð5Þφ is zero when the near-surface potential convective

energy, as indicated by qse, is equal to the potentialconvective energy aloft indicated by q500e . Negative valuesof φ are also set to zero, as these represent stable conditionsthat do not contribute to precipitation. If φ is positive, thenmore potential convective energy exists at low levels,indicating potential instability of the atmosphere. Allindicators are calculated as a mean value for the specifiedregion for each month.

The four indicators used here represent phenomenawhich can act together to produce precipitation. In orderto find what combination of indicators produces the bestpredictor of monthly precipitation, a multivariable stepwiselinear regression with a 0.05 significance cutoff wasperformed for each region, that is, variables are added andremoved from the regression based on whether theirsignificance, measured using an F statistic, is better thanthe 0.05 cutoff. Hence all variables reported in Table 5 aresignificant at least to the 0.05 level. Note that due to theway the indices have been defined, only positive coef-ficients have physical meaning and are allowed. Theresulting best model for each subregion can be found inTable 5 along with the corresponding correlation coeffi-

cient. The best statistical models are shown for MM5/NNRP, as this best represents current precipitation, as wellas MM5/CCSM and MM5/CCSM95 which show thesimulated models for early and late twenty-first century,respectively. The results shown in Table 5 for MM5/NNRPcan be compared to those shown in Table 7 of Evans et al.(2004), though those results are for a different RCMsimulating a time period 10 years earlier. Taking intoaccount slight changes in the definitions of φ and ψ, thedominant precipitation mechanisms identified are remark-ably similar, suggesting little change in the dominantmechanisms over this time frame.

Storm track is the dominant precipitation predictor forthe Black Sea subregion, both currently and late in thetwenty-first century. Late in the twenty-first century, thecoefficients of the storm track indices have increasedsignificantly, indicating that for the same variation ingeopotential or kinetic energy more precipitation is pro-duced. This is caused by the presence of more water vaporin the systems as they pass, producing higher precipitationtotals, as is expected in a globally warmed world. It is alsoconsistent with a change in storm track noted for this area(see Bengtsson et al. 2006; Lambert and Fyfe 2006; Evans2009) such that the number of low pressure systemsapproaching this subregion by crossing the Taurus Moun-tains from the south (low water vapor) decreases, while thenumber approaching from across the Black Sea (high watervapor) increases. It is worth noting that the precipitationincrease in this region is not seen in the GCM ensemblepresented in Evans (2009). This is at least partly due to thelack of a Caucuses Mountain range in the GCMs (see

Region RCM Precipitation model R2

Black Sea MM5/NNRP 0.85+0.93sdgp 0.35

MM5/CCSM −20.23+0.23sdgp+0.39sdke 0.44

MM5/CCSM95 −79.09+1.05sdgp+1.21sdke 0.70

Caspian Sea MM5/NNRP −18.98+1.01sdgp 0.43

MM5/CCSM −8.80+0.52sdgp 0.23

MM5/CCSM95 −64.74+0.70sdgp+562.78ψ 0.27

Mediterranean Sea MM5/NNRP −15.84+0.77sdke 0.39

MM5/CCSM −36.52+0.59sdgp+72.14ψ 0.43

MM5/CCSM95 −9.19+160.86ψ 0.21

Fertile Crescent MM5/NNRP −39.36+0.84sdke+96.50ψ 0.65

MM5/CCSM −14.02+0.20sdgp+0.15sdke 0.34

MM5/CCSM95 −33.29+0.37sdgp+191.97ψ 0.31

Zagros Mountains MM5/NNRP −18.68+111.63ψ 0.49

MM5/CCSM −10.37+0.25sdgp+12.20ψ 0.39

MM5/CCSM95 −10.83+69.14ψ 0.26

Saudi desert MM5/NNRP −3.00+6.91φ+0.08sdke 0.47

MM5/CCSM −0.63+0.03sdgp 0.16

MM5/CCSM95 −20.35+8.09φ+0.39sdgp 0.77

Table 5 Dominant precipitation-triggering mechanisms statisticalmodel

Global warming impact on the precipitation processes in the Middle East 399

Fig. 1c). This means that the path with the lowest energyrequirements for air parcels approaching the eastern BlackSea from the west is a deviation to the north, around thetopography that is present. The RCM has a clear CaucusesMountain range that splits air parcels approaching from thewest such that some deviate north of the range while therest deviate south. The parcels forced south become trappedby converging tall mountain ranges and are forced toascend, causing precipitation.

The Caspian Sea subregion precipitation is dominated bymovement of storms through the area during the earlytwenty-first century. By late century, the dominant mecha-nism is upslope flow which accounts for almost two thirdsof the achieved R2. The storm tracks remain a significantinfluence, though diminished in their importance.

The storm track indices dominate the present-dayprecipitation mechanisms for the Mediterranean subregion.According to MM5/CCSM, some significant contribution isprovided by the upslope flow as well. By late century, thedominant mechanism is the upslope flow, though it has alower R2. Again, the previously mentioned change in stormtracks away from the eastern Mediterranean plays animportant role.

The Fertile Crescent subregion contains much of theheadwaters for the Euphrates and Tigris rivers. Present-dayprecipitation is dominated by storm track and upslope flowconditions. These remain the dominant mechanisms late inthe twenty-first century, though the upslope flow becomesmore important and accounts for around 90% of theachieved R2. This dominance of the upslope flow intriggering precipitation is due largely to an increase in thewater vapor transport into the area by a mountain barrier jetforming above the slopes of the Zagros mountains (Evans2008). This strong southeasterly flow becomes upslopeflow where the northwest/southeast-oriented Zagros moun-tains meet the west/east-oriented mountains of Anatolia inTurkey.

The Zagros Mountain subregion precipitation is domi-nated by upslope flow at all times. MM5/CCSM found asignificant contribution also came from the sdgp stormtrack indices. As with the Caspian Sea, the MediterraneanSea, and Fertile Crescent subregions, the Zagros Mountainssubregion is affected by the movement of storm tracksaway from the eastern Mediterranean, and instead of beingdirectly influenced by the passage of storm centers, all ofthese subregions become more dependent on the transportof water vapor to the subregion and the subsequent liftingproduced by mountains within each region.

The Saudi desert subregion precipitation is dominated byatmospheric stability and storm track activity. MM5/CCSMfinds no relationship with stability, though it has a very lowR2 and predicts a total of less than 10 mm of precipitationannually. By late century, the atmospheric stability plays

the largest role, being responsible for over 90% of the R2

achieved. This large increase in importance of the atmo-spheric stability is the cause of the late summer/earlyautumn precipitation peak and is related to the influence ofthe ITCZ as it moves northward during the century.

5 Conclusions

In this study, the ability of a regional climate model, basedon MM5, to simulate the climate of the Middle East at thebeginning of the twenty-first century is assessed. The RCMdisplays a negative bias in temperature throughout the yearand over most of the domain. It does, however, bettercapture the spatial variability present than its driving globalmodel, as is expected due to its higher spatial resolution.While the RCM does a good job of simulating theprecipitation for most of the domain, it performs relativelypoorly over the southeast Black Sea and southwest CaspianSea. These areas contain steep topography, high mountains,and extensive shorelines, and it appears that the 27-kmresolution of the RCM is not high enough to capture thesevariations. This highly varied terrain also increases theuncertainty associated with the interpolation of pointobservations such that the RCM is being evaluated againstmore uncertain observational fields.

The RCM was evaluated against observations whendriven by boundary conditions from the NCEP/NCARreanalysis (NNRP) and the CCSM3 GCM. The RCMdriven by NNRP performed best as expected since thereanalysis is constrained by observations. Without observa-tional nudging, the CCSM-driven simulation, while worse,performs well, being only slightly worse than thereanalysis-driven simulation in most cases. Over theEastern Black Sea region, the (relatively small) seasonalcycle is simulated better in CCSM than NNRP.

The RCM was also used to investigate the effects ofglobal warming. Using boundary conditions obtained fromCCSM3, the RCM was run for the first and last 5 years ofthe twenty-first century. The results show widespreadwarming that is smaller in winter (~2 K) and stronger insummer (~6 K). There is a maximum warming of ~10 K ininterior Iran in summer. This region is quite dry andincreases in temperature are not moderated by increases inevapotranspiration. It also found some cooling in thesoutheast Black Sea region during spring and summer thatis related to increases in snowfall in the region, a longersnowmelt season, and generally higher soil moisture andlatent heating through the summer. The results also showwidespread decreases in precipitation over the easternMediterranean and Turkey. Precipitation increases werefound over the southeast Black Sea, southwest Caspian Sea,and Zagros mountains regions during all seasons except

400 J.P. Evans

summer, while the Saudi desert region receives increasesduring summer and autumn.

The dominant precipitation-triggering mechanisms werealso investigated for regions of precipitation extrema. Thegeneral trend in the dominant mechanism reflects a changeaway from the direct dependence on storm tracks andtowards greater precipitation triggering by upslope flow ofmoist air masses. This change is in agreement with thedecreasing trend in storm tracks for this domain in anumber of GCM analyses. The increase in the absoluteamount of low-level precipitable water due to increasingtemperatures, largely due to increased evaporation over thewater bodies, produces high precipitation amounts if it istriggered by an external mechanism such as lifting forcedby topography, hence the increased importance of upslopeflow triggering.

In the Black Sea region, the storm track is the dominantprecipitation predictor both currently and late in thetwenty-first century. The change in coefficients indicatesthat late in the century, the same storm produces moreoverall precipitation than today. In the Caspian Sea region,storm tracks are the dominant predictor currently. By latecentury, they remain a predictor but only a secondary onecompared to the upslope flow. Given the proximity ofcoastlines and mountains in these regions, a higherresolution simulation that was better able to capture seabreeze phenomena would probably increase the impor-tance of upslope flow further.

The Fertile Crescent region relies on both storm tracksand upslope flow for the triggering of precipitationcurrently and at the end of the century. Currently, stormtracks are the most important mechanism, but by the end ofthe century, the upslope flow dominates. This is causedlargely by an increase in the number of events dominatedby water vapor transport into the area by a mountain barrierjet forming above the slopes of the Zagros mountains.Precipitation in the Zagros mountains region is dominatedby upslope flow at all times. By late in the century, thisregion also experiences an increase associated with anincrease in these mountain barrier jet-related events.

The increase in precipitation in the Saudi desert region istriggered by changes in atmospheric stability brought aboutby a small northward movement of the ITCZ, allowing it tointrude into the southernmost portion of the domain. Thisdecrease in stability combined with an increase in low-levelwater vapor produces precipitation increases mostly in thetransition seasons.

Acknowledgments I would like to thank Oak Ridge National Lab(ORNL), in particular Dr. David Erickson, for making the CCSM3data available for the creation of boundary conditions for MM5. Iacknowledge the modeling groups for providing their data foranalysis, the Program for Climate Model Diagnosis and Intercom-parison (PCMDI) for collecting and archiving the model output, and

the JSC/CLIVAR Working Group on Coupled Modelling (WGCM)for organizing the model data analysis activity. The multi-model dataarchive is supported by the Office of Science, US Department ofEnergy and made available through the Earth System Grid. TheNational Center for Atmospheric Research (NCAR) supplied theNCEP/NCAR reanalysis data and computing capacity to performthe MM5 runs. I also thank members of the SWAP team at YaleUniversity: Ron Smith; Roland Geerken; Frank Hole; and LarryBonneau for continuing fruitful discussions. This study wascarried out as part of the research project “The Water Cycle ofthe Tigris–Euphrates Watershed: Natural Processes and HumanImpacts.” (NNG05GB36G) supported and financed by NASA.

References

Alpert P, Shafir H, Issahary D (1997) Recent changes in theclimate at the Dead Sea—a preliminary study. Clim Change37:513–537

Ben-Gai T, Bitan A, Manes A, Alpert P, Rubin S (1998) Spatial andtemporal changes in rainfall frequency distribution patterns inIsrael. Theor Appl Climatol 61:177–190

Ben-Gai T, Bitan A, Manes A, Alpert P, Rubin S (1999) Temporal andspatial trends of temperature patterns in Israel. Theor ApplClimatol 64:163–177

Bengtsson L, Hodges KI, Roeckner E (2006) Storm tracks and climatechange. J Clim 19:3518–3543

Chakraborty A, Behera SK, Mujumdar M, Ohba R, Yamagata T(2006) Diagnosis of tropospheric moisture over Saudi Arabia andinfluences of IOD and ENSO. Mon Weather Rev 134:598–617

Chen F, Mitchell K (1999) Using GEWEX/ISLSCP forcing data tosimulate global soil moisture fields and hydrological cycle for1987–1988. J Meteorol Soc Jpn 77:1–16

Chen F, Dudhia J (2001) Coupling an advanced land surface-hydrology model with the Penn State–NCAR MM5 modelingsystem. Part II: preliminary model validation. Mon Weather Rev129:587–604

Cohen S, Ianetz A, Stanhill G (2002) Evaporative climate changes atBet Dagan, Israel, 1964–1998. Agric For Meteorol 111:83–91

Collins WD, Bitz CM, Blackmon ML, Bonan GB, Bretherton CS,Carton JA, Chang P, Doney SC, Hack JJ, Henderson TB, KiehlJT, Large WG, McKenna DS, Santer BD, Smith RD (2006) Thecommunity climate system model version 3 (CCSM3). J Clim19:2122–2143

Dudhia J (1993) A nonhydrostatic version of the Penn State/NCARmesoscale model: validation tests and simulation of an Atlanticcyclone and cold front. Mon Weather Rev 121:1493–1513

Evans JP (2008) Changes in water vapor transport and the production ofprecipitation in the Eastern Fertile Crescent as a result of globalwarming. J Hydrometeor 9:1390–1401. doi:10.1175/2008JHM998.1

Evans JP (2009) 21st century climate change in the Middle East. ClimChange 92:417–432. doi:10.1007/s10584-008-9438-5

Evans JP, Smith RB (2006) Water vapor transport and the productionof precipitation in the Eastern Fertile Crescent. J Hydrometeorol7:1295–1307

Evans JP, Zaitchik BF (2008) Modeling the large-scale water balanceimpact of different irrigation systems. Water Resour Res 44:W08448. doi:10.1029/2007WR006671

Evans JP, Smith RB, Oglesby RJ (2004) Middle East climatesimulation and dominant precipitation processes. Int J Climatol24:1671–1694

Evans JP, Oglesby RJ, Lapenta WM (2005) Time series analysis ofregional climate model performance—art. no. D04104. J Geo-phys Res—Atmospheres 110:4104–4104

Global warming impact on the precipitation processes in the Middle East 401

Ghasemi AR, Khalili D (2006) The influence of the Arctic Oscillationon winter temperatures in Iran. Theor Appl Climatol 85:149–164

Giorgi F, Mearns LO (1999) Introduction to special section: regionalclimate modeling revisited. Journal of Geographical Research104:6335–6352

Grell GA, Dudhia J, Stauffer DR (1994) A description of the fifthgeneration Penn State/NCAR mesoscale model (MM5). Vol.NCAR/TN-398 +STR, NCAR Tech. Note, National Center forAtmospheric Research, 117 pp

Hong SY, Pan HL (1996) Nonlocal boundary layer vertical diffusionin a medium-range forecast model. Mon Weather Rev 124:2322–2339

Jacquemin B, Noilhan J (1990) Sensitivity study and validation of aland surface parameterization using the HAPEX-MOBILHY dataset. Bound-Lay Meteorol 52:93–134

Kistler R, Kalnay E, Collins W, Saha S, White G, Woollen J, ChelliahM, Ebisuzaki W, Kanamitsu M, Kousky V, van den Dool H,Jenne R, Fiorino M (2001) The NCEP–NCAR 50-year reanaly-sis: monthly means CD-ROM and documentation. Bull AmMeteorol Soc 82:247–267

Kostopoulou E, Jones PD (2005) Assessment of climate extremes inthe Eastern Mediterranean. Meteorol Atmos Phys 89:69–85

Krichak SO, Alpert P (2005) Signatures of the NAO in theatmospheric circulation during wet winter months over theMediterranean region. Theor Appl Climatol 82:27–39

Krichak SO, Tsidulko M, Alpert P (2000) November 2, 1994, severestorms in the southeastern Mediterranean. Atmos Res 53:45–62

Kutiel H, Benaroch Y (2002) North Sea–Caspian Pattern (NCP)—anupper level atmospheric teleconnection affecting the EasternMediterranean: Identification and definition. Theor Appl Clima-tol 71:17–28

Lambert SJ, Fyfe JC (2006) Changes in winter cyclone frequenciesand strengths simulated in enhanced greenhouse warming experi-ments: results from the models participating in the IPCCdiagnostic exercise. Clim Dyn 26:713–728

Legates DR, McCabe GJ (1999) Evaluating the use of "goodness-of-fit"measures in hydrologic and hydroclimatic model validation. WaterResour Res 35:233–241

Mahrt L, Ek M (1984) The influence of atmospheric stability onpotential evaporation. J Clim Appl Meteorol 23:222–234

Mahrt L, Pan HL (1984) A two-layer model of soil hydrology. Bound-Lay Meteorol 29:1–20

McGregor JL (1997) Regional climate modelling. Meteorol AtmosPhys 63:105–117

Meehl GA, Washington WM, Santer BD, Collins WD, Arblaster JM,Hu AX, Lawrence DM, Teng HY, Buja LE, Strand WG (2006)Climate change projections for the twenty-first century andclimate change commitment in the CCSM3. J Clim 19:2597–2616

Min SK, Hense A (2006) A Bayesian assessment of climate changeusing multimodel ensembles. Part I: global mean surfacetemperature. J Clim 19:3237–3256

Mlawer EJ, Taubman SJ, Brown PD, Iacono MJ, Clough SA (1997)Radiative transfer for inhomogeneous atmosphere: RRTM, avalidated correlated-k model for the longwave. J Geophys Res102:16663–16682

Nasrallah HA, Balling RC (1995) Impact of desertification ontemperature trends in the Middle East. Environ Monit Assess37:265–271

Nazemosadat MJ, Cordery I (2000) On the relationships betweenENSO and autumn rainfall in Iran. Int J Climatol 20:47–61

Pan HL, Mahrt L (1987) Interaction between soil hydrology andboundary-layer development. Bound-Lay Meteorol 38:185–202

Paz S, Tourre YM, Planton S (2003) North Africa–West Asia (NAWA)sea-level pressure patterns and their linkages with the EasternMediterranean (EM) climate. Geophys Res Lett 30:1999

Reddaway JM, Bigg GR (1996) Climatic change over the Mediterra-nean and links to the more general atmospheric circulation. Int JClimatol 16:651–661

Reisner J, Rasmussen RJ, Bruintjes RT (1998) Explicit forecasting ofsupercooled liquid water in winter storms using the MM5mesoscale model. Q J Roy Meteorol Soc 124B:1071–1107

Saaroni H, Ziv B (2000) Summer rain episodes in a Mediterraneanclimate, the case of Israel: climatological–dynamical analysis. IntJ Climatol 20:191–209

Saaroni H, Ziv B, Bitan A, Alpert P (1998) Easterly wind storms overIsrael. Theor Appl Climatol 59:61–77

Schaake JC, Koren VI, Duan QY, Mitchell K, Chen F (1996) A simplewater balance model (SWB) for estimating runoff at differentspatial and temporal scales. J Geophys Res 101:7461–7475

Small EE, Sloan LC, Nychka D (2001) Changes in surface airtemperature caused by desiccation of the Aral Sea. J Clim14:284–299

Trenberth KE (1991) Storm tracks in the Southern-Hemisphere. JAtmos Sci 48:2159–2178

Turkes M, Erlat E (2005) Climatological responses of winterprecipitation in Turkey to variability of the North AtlanticOscillation during the period 1930–2001. Theor Appl Climatol81:45–69

Walsh K, McGregor J (1997) An assessment of simulations of climatevariability over Australia with a limited area model. Int JClimatol 17:201–223

Wang G (2005) Agricultural drought in a future climate: results from15 global climate models participating in the IPCC 4thassessment. Clim Dyn 25:739–753

Wilcoxon F (1945) Individual comparisons by ranking methods.Biometrics 1:80–83

Willmott CJ, Ackleson SG, Davis RE, Feddema JJ, Klink KM,Legates DR, Odonnell J, Rowe CM (1985) Statistics for theevaluation and comparison of models. J Geophys Res—Oceans90:8995–9005

Zaitchik BF, Evans J, Smith RB (2005) MODIS-derived boundaryconditions for a mesoscale climate model: application to irrigatedagriculture in the Euphrates basin. Mon Weather Rev 133:1727–1743

Zaitchik BF, Evans JP, Geerken RA, Smith RB (2007a) Climate andvegetation in the Middle East: inter-annual variability anddrought feedbacks. J Clim 20:3924–3941

Zaitchik BF, Evans JP, Smith RB (2007b) Regional impact of anelevated heat source: the Zagros Plateau of Iran. J Clim 20:4133–4146

Zhang XB, Aguilar E, Sensoy S, Melkonyan H, Tagiyeva U, AhmedN, Kutaladze N, Rahimzadeh F, Taghipour A, Hantosh TH,Albert P, Semawi M, Ali MK, Al-Shabibi MHS, Al-Oulan Z,Zatari T, Khelet IA, Hamoud S, Sagir R, Demircan M, Eken M,Adiguzel M, Alexander L, Peterson TC, Wallis T (2005) Trendsin Middle East climate extreme indices from 1950 to 2003. JGeophys Res—Atmospheres 110:D22104

402 J.P. Evans