• Open Access

Interannual variability of rainfall over the Arabian Peninsula using the IPCC AR4 Global Climate Models

Authors


Correspondence to: M. Almazroui, Center of Excellence for Climate Change Research/Department of Meteorology, King Abdulaziz University, P. O. Box 80234, Jeddah 21589, Saudi Arabia. E–mail: mansour@kau.edu.sa

ABSTRACT

The interannual rainfall variability derived from the 22 Global Climate Model (GCM) simulations of the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) for the duration 1979–2000 is analysed and compared with the gridded observed dataset over the Arabian Peninsula. The annual cycle of the rainfall derived from these models is validated for the entire Arabian Peninsula, and separately for its two sub-regions, named northern and southern Arabian Peninsula. The spatial patterns of the rainfall and the mean sea level pressure are analysed to judge the ability of the models to simulate the mean climatology of the Peninsula. This analysis reveals that out of the 22 IPCC AR4 GCM multi-model datasets, only one group (composed of 5 models) is relatively better than all the others in simulating the interannual variability of the wet season rainfall for the northern sub-region, and another group (also composed of 5 models) is likewise for the dry season rainfall climatology of the southern sub-region, compared with the gridded dataset. The above two groups of models tend to fall within one-sigma standard deviation of the mean seasonal rainfall derived from the gridded dataset. Moreover, only one model [CCCMA-CGCM3 (T47) from Canada] is found to be relatively better in simulating the rainfall climatology for both the wet and the dry seasons (i.e. for the northern and the southern sub-regions) simultaneously, compared with the observed data.

1. Introduction

The Arabian Peninsula is located in southwest Asia at the junction of Africa and Asia; it is an important part of the Middle East, and consists of Bahrain, Kuwait, Oman, Qatar, United Arab Emirates, Yemen and Saudi Arabia. Saudi Arabia alone covers almost 80% of the Peninsula (Almazroui et al., 2012). This region is perhaps more appropriately called the Arabian subcontinent because it lies on the Arabian tectonic plate. The north-eastern side of the Peninsula is bordered by the Arabian Gulf; on the western side the Red Sea is located, whereas the south-eastern side is bordered by the Arabian Sea. The Red Sea and the Arabian Gulf are the main sources of water vapour, given the descending motion within the Hadley cell in the region (Evans et al., 2004). The Sudan low, Mediterranean flow, Indian monsoon and position of the Intertropical Convergence Zone (ITCZ) all control the development of rainfall systems in this region (Atlas, 1984; Walters and Sjoberg, 1988; Zekai and Khalid, 2002; Chakraborty et al., 2006). A mountain range is present, running parallel to the Red Sea coast in the south-western region of the Peninsula, which also plays a role in the development of rainfall systems in the region (see Figure 1). Most of the rainfall in the south-western areas of the Peninsula occurs in this mountainous area. The highest elevations within this mountain range are in Yemen, although relatively significant heights extend into Saudi Arabia. Another prominent feature of the Peninsula is desert. The Rub Al-Khali is the world's largest continuous sand desert, which covers almost the entire south-eastern region of Saudi Arabia (Atlas, 1984; Edgell, 2006; Bishop, 2010). The Arabian Peninsula has very few lakes or permanent rivers, and most of the water storage areas lie under dry riverbeds, called ‘wadis’; the main source of water in these wadis is seasonal rainfall (Zekai and Khalid, 2002).

Figure 1.

The topography (m) of the Arabian Peninsula. The Box indicates the two sub-regions of the Arabian Peninsula, the upper is the northern and the lower is the southern sub-region, as discussed in the text.

Among all the atmospheric variables, rainfall plays a central role in determining the climate impact of an arid region (Edgell, 2006). In particular, variability in the rainfall is a crucial factor in determining the spatio-temporal extent of the arid climate. Thus, given the arid nature of the climate of the Arabian Peninsula, it is of paramount importance for a highly detailed representation of rainfall to be made available, as water supplies, agriculture, food production and power generation are directly affected by rainfall variability.

In general, the Arabian Peninsula's climate is classified as arid and semi-arid climate (Almazroui, 1998; Almazroui, 2012). The Peninsula has no dense forests, although desert-adapted wildlife is present throughout the region. The south-western region receives rainfall during almost every month of the year, whereas the northern region (above 22°N) receives most of its rainfall only during what may be termed the winter and spring seasons, although in the Arabian Peninsula, this would be the wet season (Al-Jerash, 1985). The rainfall in the south-western region is associated with monsoon activity in the Indian Ocean (Atlas, 1984). In the wet season, the rainfall mechanism is mainly associated with the migration of Mediterranean cyclones (from west to east), along with the availability of an upper trough and the presence of the active phase of the Sudan trough (Walters and Sjoberg, 1988; Abdullah and Almazroui, 1998; Zekai and Khalid, 2002).

Almazroui (2011a) studied the rainfall climatology of Saudi Arabia by using the Tropical Rainfall Measuring Mission dataset, and defined the wet and dry season respectively from November to April and from June to September. Drought is one of the prominent climatic features of the Arabian Peninsula, which hampers livestock production and the agricultural economy of the region; therefore, the wet season is important for the Peninsula in defining drought in this region (Abdullah and Almazroui, 1998; Zekai and Khalid, 2002; Almazroui, 2006; Almazroui, 2011a). During the wet season, rain occurs in the northern Peninsula; there is essentially no rain in the southern Peninsula (below 22°N), except in the south-western area. Conversely, during the dry season, the rainfall that occurs in the southern Arabian Peninsula falls mostly in the south-western area. Thus, only the wet season in the northern Arabian Peninsula and the dry season in the southern Peninsula are considered in this study.

Nasraullah and Balling (1996) studied the temperature trends from the observed and gridded datasets for the Arabian Peninsula. They showed that for the period 1891–1990, the temperature increased by 0.63°C, whereas the station-based rainfall showed a statistically insignificant decrease for the period 1940–1989. Almazroui (2011b) studied the simulation of extreme rainfall events over the Jeddah region of Saudi Arabia using a Regional Climate Model (RCM), and found that domain size and domain centre are important in simulating extreme events in the region. Hemming et al. (2010) studied the projections of two Global Climate Models (GCMs) and one RCM over the Middle East in order to determine the availability of water resources. They studied the uncertainties in the water availability projections over the Middle East, including the Arabian Peninsula, for the period 2021–2050 using 22 models from the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) GCM simulations. They suggested that efforts should be made to reduce the model number by focusing on large-scale processes and their teleconnections to the Middle East climate, and on localized processes affecting the rainfall in the region.

Climate models are gradually proving to mimic the (coupled) state of the atmosphere quite well (Räisänen, 2007; Edwards, 2011). Climate modelling groups around the world have performed a comprehensive, systematic, well-designed and well-coordinated set of 20th and 21st century climate change experiments for the IPCC AR4 (IPCC, 2007; Kripalani et al., 2007). The IPCC AR4 coupled climate model datasets of the 20th century experiments, forced by changing greenhouse gas concentrations, as well as those for the control simulations, are available as ensembles through the Program for Climate Model Diagnosis and Intercomparison (PCMDI) website (http://www-pcmdi.llnl.gov). The coupled IPCC AR4 GCM performance has been studied over different regions of the globe. It has been found that for a particular region, the models chosen for simulating the current climate conditions perform differently (Annamalai et al., 2007; Kripalani et al., 2007; Carolina and Gabriel, 2009; Errasti et al., 2011). Furthermore, different methodologies lead to differing selections of models, even for one particular region (Annamalai et al., 2007; Kripalani et al., 2007). The general consensus is that although a model may be good in one region, it may not to be good in simulating the climate conditions of other regions. Therefore, it is always deemed efficacious to validate the models on the basis of their performance in simulating the current climate conditions of the particular region under consideration. If the model biases are low and they are able to simulate well the current climate conditions of the region, then it may be assumed that their projections are also reliable for the region (Errasti et al., 2011). Therefore, this article is focused on the evaluation of the rainfall climatology obtained from the IPCC AR4 22 GCM simulations over the Arabian Peninsula as a whole as well as over its two sub-regions (northern and southern) on the seasonal scale (wet and dry, respectively).

2. Datasets and methodology

2.1. Datasets

The IPCC AR4 22 GCM datasets are available through the PCMDI website. The 1° × 1° spatial resolution re-gridded datasets used in this analysis were obtained from the Earth System Physics Section, the Abdus Salam International Centre for Theoretical Physics (ICTP), Italy, to compare the model's simulated rainfall with the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) gridded dataset; this dataset is abbreviated as CMP for the rest of the analysis. The GCM model characteristics and the acronyms used in this study are presented in Table 1. The CMP rainfall dataset with 2.5° × 2.5° spatial resolution is used as an (observed) gridded dataset (Xie and Arkin, 1997). Sea level pressure from the NCEP/NCAR is also used as an observed dataset for the region (Kalnay et al., 1996; Kistler et al., 2001); this dataset is abbreviated as NCEP-SLP. The CMP data is also re-gridded to 1° × 1° spatial resolution in order to validate the GCMs' rainfall over the Arabian Peninsula.

Table 1. The IPCC AR4 GCMs used for the Arabian Peninsula rainfall interannual variability analysis
IPCC ID of the modelCountryAcronymAtmospheric resolutionKey references
MRI-CGCM2.3.2JapanMR2.8° × 2.8°Yukimoto et al. (2001)
ECHAM5/MPI-OMGermanyEC1.9° × 1.9°Jungclaus et al. (2006)
BCC-CM1ChinaBC1.9° × 1.9°Dong (2001)
CCCMA-CGCM3.1(T47)CanadaCC2.8° × 2.8°Flato et al. (2000)
CSIRO-Mk3.0AustraliaCS1.9° × 1.9°Gordon et al. (2002)
MIROC3.2(hires)JapanMH1.1° × 1.1°Hasumi and Emori (2004)
MIROC3.2(medres)JapanMM2.8° × 2.8°Hasumi and Emori (2004)
GFDL-CM2.1USAG12.0° × 2.5°Delworth et al. (2006)
GFDL-CM2.0USAG02.0° × 2.5°Delworth et al. (2006)
MIUB-ECHO-GGermany/KoreaMI3.9° × 3.9°Legutke and Voss (1999)
GISS-AOMUSAGA3° × 4°Russell et al. (1995)
GISS-ERUSAGR4° × 5°Schmidt et al. (2006)
GISS-EHUSAGH4° × 5°Schmidt et al. (2006)
IAP-FGOALS-g1.0ChinaIF2.8° × 2.8°Yu et al. (2004)
UKMO-HadGEM1UKHG1.3° × 1.9°Johns et al. (2006)
UKMO-HadCM3UKHC2.5° × 3.75°Jones et al. (2004)
INM-CM3.0RussiaIN4° × 5°Diansky and Volodin (2002)
BCCR-BCM2.0NorwayBB1.9° × 1.9°Furevik et al. (2003)
NCAR-CCSM3USANC1.4° × 1.4°Collins et al. (2006)
NCAR-PCMUSANP2.8° × 2.8°Washington et al. (2000)
IPSL-CM4FranceIC2.5° × 3.75°Marti et al. (2005)
CNRM-CM3FranceCN1.9° × 1.9°Salas-Mélia et al. (2005)

The available period of simulation of the IPCC GCMs for the 20th century is 100 years or more. However, in this study, a 22-year analysis period (1979–2000) is selected in accordance with the availability of common data from within the models and the observations. Most of the models have more than one ensemble member, and therefore the ensemble mean for rainfall and mean sea level pressure is used in the analysis, where applicable.

2.2. Methodology

The wet (November to April) and dry (June to September) seasons are adopted from Almazroui (2011a). As each region in the Peninsula has different climatic characteristics, on this basis, the whole Arabian Peninsula (12–32°N, 35–60°E) is divided into two sub-regions: the northern Arabian Peninsula (22–32°N, 35–60°E) and the southern Arabian Peninsula (12–22°N, 35–60°E) (see Figure 1). The Arabian Peninsula as a whole and its two sub-regions (northern and southern) are considered for the analysis of rainfall climatology. The sub-region of the southern Arabian Peninsula includes some parts of Sudan, Eritrea and Ethiopia, whereas the northern Arabian Peninsula sub-region includes some parts of Iran. The northern and southern sub-regions are considered, respectively, for the wet and the dry season rainfall analyses. The ability of the GCMs in simulating the interannual variability of the rainfall over the Peninsula is studied by applying widely used statistical methods (Wilks, 2006; Gleckler et al., 2008; Errasti et al., 2011). The intention is to judge how well the models are able to simulate the mean present rainfall climatology on the annual and on the seasonal basis in comparison with the CMP rainfall. The following stepwise methodology is adopted to validate the rainfall of the 20th century modelling over the Peninsula:

  1. The annual cycle is analysed for the entire Peninsula, the northern and the southern sub-regions in order to judge the mean monthly behaviour of the models as compared with the gridded CMP rainfall. The correlation is calculated to determine how well the models are able to simulate the annual cycle of the rainfall in the regions in comparison with the CMP rainfall. The ‘Student's t-test’ is used to estimate the significance of the correlation at the 95% significance level.
  2. The root mean square error (RMSE) is calculated to identify the error/bias in each model in comparison with the CMP rainfall. The smaller the error/bias is, the better the model performance is, in simulating the mean present climate in comparison with the CMP dataset (Annamalai et al., 2007).
  3. The seasonal mean and its standard deviation are calculated to determine the variability of the modelled rainfall. The one-sigma standard deviation criterion is used to identify the relatively better performing models in simulating the seasonal variability of the rainfall (in the given region) in comparison with the CMP dataset.
  4. Coefficient of variation (CV) is defined as the ratio of standard deviation to the mean of the sample. Scatter plots of the seasonal mean rainfall versus the CV are plotted to analyse the variability of the models in terms of CV variation (Kripalani et al., 2007).
  5. The spatial plots of the rainfall and the mean sea level pressure of the models are obtained to identify the spatial behaviour of the models in comparison with the corresponding observed dataset for the region of interest. These are good indicators in simulating the interannual variability of the rainfall in the region on the seasonal scale.

3. Results and discussion

3.1. Mean rainfall climatology

This section describes the mean annual and seasonal (wet and dry) rainfall climatology over the Arabian Peninsula. The annual cycle for the wet season is focused on the northern sub-region, whilst the dry season is focused on the southern sub-region.

The mean annual rainfall climatology obtained from the gridded CMP dataset averaged over the period 1979–2000 is displayed in Figure 2. The northern sub-region receives most of its rainfall from November to April in any given year; this is critical for the water resources and agricultural production of the region. The maximum for the rainfall (as in Figure 2(a)) is observed in the central, north-eastern and south-western areas of the Peninsula. In this connection, the rainfall during the wet and dry seasons is discussed in detail next.

Figure 2.

The rainfall (mm/day) climatology for the Arabian Peninsula obtained from the CMP data for (a) annual, (b) wet season and (c) dry season, averaged over the period 1979–2000.

During the wet season (November to April) most of the rainfall occurs in the north-to-northeastern side of the Peninsula, and the system penetrates to the centre of the Peninsula (Figure 2(b)). In this sub-region, the rainfall is mainly associated with the storm activities of the northern hemisphere, whereas in the southern sub-region, the rain-bearing systems penetrate from the Ethiopia region, crossing the Red Sea to reach the south-western areas of the Peninsula (Walters and Sjoberg, 1988). This is the tropical belt, where the rainfall is mostly affected by the position of the ITCZ. It is also evident from Figure 2(b) that most of the rainfall during the wet season occurs over the northern sub-region, relative to the southern sub-region in the Peninsula.

During the dry season (June to September), most of the rainfall in the Peninsula occurs over the south and south-western regions (Figure 2(c)). The rainfall during this season is associated with the monsoon activity in the Indian Ocean, particularly in the Arabian Sea, and it occurs in the south-to-southwestern areas of the Peninsula (Zekai and Khalid, 2002). The maximum rainfall activity in the domain as a whole is observed over South Sudan and Ethiopia. During the dry season, the northern and south-eastern regions of Saudi Arabia and other (Gulf) countries are dry. As most parts of the Arabian Peninsula are dry except for the south and south-western region of Saudi Arabia and Yemen, this season is considered the dry season in the Peninsula.

3.2 Annual cycle

The annual cycle of rainfall is one of the primary features of rainfall climatology (see, for instance, Räisänen, 2007). It is to be expected that models should at least capture some of the basic features of the annual cycle. The annual cycle of the rainfall from the GCM datasets and the CMP dataset for the Peninsula as a whole, the northern and the southern sub-regions are discussed next.

3.2.1. Whole Arabian Peninsula

The annual cycle for the whole Arabian Peninsula is the area-weighted average rainfall for the 22 years derived from the models is shown in Figure 3(a) and compared with the CMP rainfall. According to the CMP rainfall, the Arabian Peninsula receives about 144 mm rainfall on an annual basis with a peak during the month of March. During this peak month Peninsula receives about 20 mm of the rainfall. The November to April period (considered as the wet season) is dominant in the annual cycle of the whole Arabian Peninsula; this period receives about 60% of the annual rainfall in the Peninsula. During the summer season (dry season) most of the rainfall occurs over Sudan, South Sudan, Ethiopia and the south-western region of Saudi Arabia as well as over Yemen (see Figure 2(c)). Overall, some models simulate well the wet season rainfall climatology, and some of them simulate well the dry season rainfall climatology. However, a few of them greatly overestimate the dry season rainfall compared with the gridded CMP dataset, establishing the necessity for identifying the best models for simulating the rainfall climatology over the study area. Because of the large variations in the modelled rainfall for the Arabian Peninsula as a whole as well as for the northern and the southern sub-regions, the rainfall scales of the annual cycles are different in Figure 3.

Figure 3.

The annual cycle of the IPCC AR4 GCMs compared with the CMP rainfall for (a) the whole Arabian Peninsula, (b) the northern sub-region and (c) the southern sub-region. Solid black line (with filled diamonds) is for the CMP rainfall.

None of the models in Figure 3(a) is able to simulate the annual cycle of the whole Peninsula particularly well except MR (see model acronyms in Table 1). Its correlation with the CMP dataset is statistically significant at the 95% confidence level (shown in solid line with upright triangles). Because of the different rainfall amounts in the wet/dry seasons in the northern/southern sub-regions, most of the models are unable to capture the rain in these two seasons simultaneously on an annual basis. Therefore, the following analysis is focused on the northern and the southern sub-regions of the Peninsula separately. The models are ranked according to their relative performance in simulating the climatic features of these two regions, and are discussed next.

3.2.2. Northern Arabian Peninsula

The annual cycle of the rainfall for the northern sub-region, obtained from the GCM dataset is compared with the CMP dataset (Figure 3(b)). The CMP rainfall shows that the March is the peak rainy month in the northern sub-region, and it receives on average about 28 mm rainfall during this month. The northern sub-region receives on average about 135 mm rainfall on an annual basis for the period 1979–2000. On average, wet season receives about 118 mm rainfall, which is about 87% of the annual total in the northern sub-region. The dry season in the northern sub-region receives on average about 6 mm rainfall during this period, which is only 4% of the annual total rainfall in this sub-region.

In this northern sub-region, the month of November indicates the onset of the rainy season with a monthly average rainfall of about 14 mm, and peaks during the month of March, as mentioned earlier. The rainfall drops to about 13 mm in the month of April, indicating the departure of the rainy season in the northern sub-region of the Peninsula. This annual cycle of the rainfall in the northern sub-region displays typical Gaussian behaviour. During the dry season, the northern sub-region receives almost no or very little rain.

The correlation between the annual cycle for the rainfall of the CMP dataset and the following 12 models is statistically significant at the 95% confidence level: MR, EC, CC, MH, HC, G0, G1, IC, GR, HG, IF and GA; these models are shown as solid lines in Figure 3(b). These models are able to simulate the annual rainfall climatology quite well in comparison with the annual cycle of the CMP rainfall in the northern sub-region. Although some models underestimated the rainfall amount, compared with the CMP rainfall, overall, the 12 models mentioned above capture well the variability in the annual rainfall, and reproduce the behaviour of the annual rainfall, i.e. the peak rainfall in March, and the dry period during June to September in the northern sub-region. These 12 models are ranked as category-1 models for the northern sub-region of the Peninsula.

The other 10 models (MM, BC, BB, CS, GH, MI, NC, NP, IN and CN) were unable to simulate the annual rainfall climatology for the northern sub-region. The correlation of these models with the CMP rainfall climatology is either weak or negative. This shows that these models failed to capture the basic features of the annual rainfall climatology of this sub-region. These are ranked as category-2 models for the northern sub-region of the Peninsula.

Later on, the analysis of the wet season rainfall in the northern sub-region is focused only on the category-1 models, as they are able to simulate the annual climatology quite well in comparison with the CMP rainfall climatology for this sub-region.

3.2.3. Southern Arabian Peninsula

In Figure 3(c), the annual cycle of the rainfall for the southern sub-region from the models and the CMP dataset is displayed for the period 1979–2000. This southern sub-region covers some areas of Sudan, Eritrea and Ethiopia, which essentially determines the maximum rainfall in this region during the dry season, as evident in Figure 2(c) also. The maximum rainfall in the Sudan/Eritrea/Ethiopia region during the dry season is due to the monsoon activity in the Indian Ocean. The average annual rainfall for the period 1979–2000 in the southern sub-region is about 153 mm. During this dry season, it receives on the average about 73 mm rainfall, which is about 48% of the annual total rainfall, whereas for the wet season it receives only 37% of the annual total rainfall in this sub-region of the Peninsula.

The annual cycle of the CMP dataset indicates that August is the month of maximum rainfall in the southern sub-region; it shows on average about 25 mm of rainfall for the month of August. The month of June shows the onset of the rainy season and September is the departure month for the rainy season in the southern sub-region. It is also evident from the results that the northern sub-region is almost rainless during the dry season (see Figure 2(c)). Only the south-western region of the Peninsula receives rainfall during this season, whereas all other areas of the Arabian Peninsula are almost totally dry during this season.

The peak rainfall in the entire domain during these dry months is over the Sudan/Eritrea/Ethiopia region (see Figure 2(c)). The rainfall over the south-western area in the dry season is due to the monsoon activity in the Indian Ocean, as this system penetrates into south-western Peninsula and contributes to the rainfall in the region. However, most of the models greatly overestimate the dry season rainfall in the southern sub-region of the Peninsula.

On the annual scale, only 14 models (MR, EC, CC, MH, MM, HC, G0, G1, NC, CN, HG, BB, IF and NP) out of the 22 are able to produce statistically significant correlated cycles, in comparison with the CMP rainfall annual cycle, in the southern sub-region. The correlation of the annual cycles from these models is statistically significant at the 95% confidence level (thick solid lines, Figure 3(c)). This implies that these models are relatively better in simulating the annual rainfall in comparison with the observed data. During the dry season, the behaviour of the rainfall annual cycle in the southern sub-region can be summarized as being similar to monsoon, with a peak in the month of August. Some of these models underestimate the amount of rainfall, whereas some overestimate the rainfall. However, importantly, they are able to capture the annual cycle quite well. These models are ranked as category-1 for the southern sub-region of the Peninsula.

The remaining eight models (BC, CS, GR, GH, GA, MI, IN and IC) are not able to capture the annual cycle of the rainfall in the southern sub-region in a statistically significant fashion. These models are unable to capture the basic features of the climatology, i.e. annual cycle for the rainfall in the southern sub-region, and are ranked as category-2. Henceforth, only the category-1 models are taken into consideration for the analysis of the dry season rainfall in the southern sub-region of the Peninsula.

3.3. Interannual variability

To verify the models' performance in simulating the mean rainfall climatology over the Arabian Peninsula, we focus on a statistical analysis of the wet and dry seasons, respectively (i.e. for the northern and the southern sub-regions), using the category-1 models only. The RMSE is calculated for the wet seasons in the northern Peninsula and for the dry seasons in the southern sub-region during the period 1979–2000 (Figure 4). The modelled RMSE of the rainfall is calculated with respect to the CMP rainfall. The horizontal dashed line represents the mean values of the RMSE for all the category-1 models with respect to the CMP rainfall in each sub-region. This line shows how well the models are able to simulate the seasonal mean rainfall in comparison with the CMP rainfall. Only the models that satisfied the annual cycle test are examined for the interannual variability of the rainfall in the northern and southern sub-regions during the wet and dry seasons, respectively.

Figure 4.

The rainfall RMSE of the IPCC AR4 GCMs with respect to the CMP-gridded dataset averaged over the period 1979–2000 for (a) wet season in the northern sub-region and (b) dry season in the southern sub-region. The horizontal dashed line represents the multi-model mean of the RMSE from the models used in each sub-region.

It is clear from Figure 4(a) that only 5 models (MR, CC, MH, GR and IF) out of 12 have relatively less RMSE compared with the multi-model mean RMSE of the category-1 models (dashed horizontal line, Figure 4(a)). The category-1 multi-model mean RMSE for the rainfall is 12.60 mm. The MR model has the greatest difference in RMSE with respect to the multi-model mean; it is 4.62 mm (a 37% difference). The MH model has the smallest difference in RMSE with respect to its multi-model mean RMSE; it is 0.62 mm (a 5% difference). It shows that MH shows a largest RMSE among the five models. These results indicate low rainfall variability in the MR, and high variability in the MH. All of these models have less bias in their simulation of the rainfall, compared with the CMP dataset, and they all accurately reproduced the seasonal mean climatology for the wet season in the northern sub-region of the Peninsula.

Figure 4(b) shows that the models MR, EC, CC, G0, G1, HG and HC have less RMSE, compared with the other models, with respect to the CMP rainfall for the dry season in the southern sub-region, and are below the mean value of RMSE from the models used (dashed horizontal line). During the dry season in the southern sub-region, the aforementioned seven models have less bias on the interannual scale compared with the CMP rainfall. These models are thus relatively better in simulating the rainfall climatology, i.e. they are close to the CMP rainfall. The MR model shows the lowest RMSE (a relative difference of 64%), whereas G1 and EC show the largest RMSE (a relative difference of 77%), relative to the multi-model mean RMSE for the dry season rainfall climatology of the southern sub-region of the Peninsula.

For the remaining GCMs displayed in Figure 4(a) and (b), the RMSE differences are above the mean value of RMSE from the models used (dashed horizontal line). This shows that the biases in their interannual variability are relatively high, compared with the selected models in the northern and southern sub-regions.

In Figure 5, the variability analysis has been performed by comparing the one-sigma standard deviation of the rainfall (mm) from the models with the CMP rainfall for the wet and dry seasons in the northern and southern sub-regions, respectively. The variability of only 5 models (MR, CC, MH, GR and IF) out of the category-1 12 models is reasonably comparable to the CMP rainfall for the wet season in the northern sub-region (Figure 5(a)). For the wet season, the CMP rainfall mean value along with the one-sigma standard deviation is 19.78 ± 6.53 mm. The MR (15.80 ± 2.25 mm) is within the lower range of this standard deviation spread, indicating that the biases in it are the lowest. The CC, GR and IF models show an underestimation of the rainfall, whereas MH (26.60 ± 7.72 mm) shows a large overestimation of the rainfall in comparison with the CMP rainfall.

Figure 5.

The seasonal rainfall mean including one-sigma standard deviation for the models compared with the CMP dataset for (a) the wet season in the northern sub-region and (b) the dry season in the southern sub-region. The CMP (first label along x-axis) is used as the acronym for the CMAP rainfall; the model acronyms are listed in Table 1.

Similarly, the dry season rainfall analysis for the southern sub-region is shown in Figure 5(b). In this season and region, only five models (EC, CC, G1, HC and HG) reasonably fall within the CMP rainfall one-sigma standard deviation spread either side of the mean. For the dry season, the CMP rainfall mean value along with the one-sigma standard deviation is 18.26 ± 5.60 mm. The results show that the mean as well as the variability of the EC rainfall (19.47 ± 6.00 mm) are comparable with the CMP rainfall, and this shows that EC has less bias in comparison with the CMP dataset, whereas the models represented by the acronyms G1, HC, CC and HG show more localized variability during the dry season, close to the mean, in comparison with the CMP rainfall. The CC model (26.16 ± 3.38 mm) shows an overestimation of the rainfall in comparison with the CMP rainfall data. The variability in G1 (18.24 ± 3.49 mm) is within the range of the CMP rainfall. The HC model (11.01 ± 2.34 mm) shows less variability in the rainfall during the dry season; this indicates less bias in the model during the dry season. Hence, an ensemble comprised of five models is relatively better in simulating the wet season rainfall climatology in the northern Arabian Peninsula, and another ensemble of five models is better for the dry season in the southern Peninsula.

The scatter plots of the seasonal mean rainfall (mm) versus the CV of the selected category-1 models for the wet season in the northern sub-region and for the dry season in the southern sub-region are shown in Figure 6. This figure is meant to display the spread in the static CV for the above models. The vertical sides of the two rectangles are based on the one-sigma spread of the seasonal mean CMP rainfall. The CV for the wet season varies between 13 and 37%, whereas it varies between 13 and 31% for the dry season. This indicates that, for the dry season, the models have less spread in CV; they are in better agreement with the CMP rainfall deviation. In Figure 6(a), only the mean and variability of MR, CC, MH, GR and IF are relatively in better agreement with the CMP rainfall during the wet season for the northern sub-region. In Figure 6(b), EC coincides with CMP; this shows that its interannual variability is in best agreement with CMP, better than any other selected model. The model G0 is not included, as its one-sigma standard deviation spread does not fall within the corresponding CMP spread, though it is inside the rectangle.

Figure 6.

The scatter plot of the seasonal mean rainfall (mm) versus CV (in %). The rectangles display the approximate CV variation of the selected category-1 GCMs for (a) the wet season in northern Arabian Peninsula and (b) the dry season in the southern sub-region.

3.4. Spatial distribution

On the basis of the temporal annual cycles and the seasonal variability analysis of the models in comparison with the CMP rainfall, it is found that a group comprising 5 models (out of the 22) is in relatively better agreement in simulating the wet season rainfall in the northern sub-region, and another group comprising 5 models is in relatively better agreement for the dry season rainfall climatology for the southern sub-region. To further verify the performance of these two groups of models in each season and region, the spatial distribution of the model seasonal rainfall and the mean sea level pressure patterns are analysed and compared with the corresponding NCEP-SLP datasets.

3.4.1. Wet season rainfall climatology

The spatial distribution of the wet season rainfall climatology for the selected five models is shown in Figure 7. The MR, CC, MH, IF and GR models perform relatively better in simulating the seasonal rainfall, compared with the CMP dataset. The MH model is able to capture the pattern of the north-easterly penetration of the rainfall quite well. It is also able to capture well the penetration of the rainfall system across the Red Sea into the south-western Peninsula. Overall, the spatial pattern obtained from MH is similar to the CMP rainfall during the wet season in the Peninsula. The IF model is able to capture the north-easterly pattern of the rainfall quite well compared with the CMP data. It also able to capture the rainfall in the Red Sea but it seems to be weak in simulating the rainfall in the south-western Peninsula during the wet season. The MR model simulates well the rainfall in the northern Peninsula, whereas it underestimates the rainfall in the central Peninsula. It is unable to capture the penetration of the rainfall over the Red Sea and in the south-western Peninsula, whilst it simulates well the maximum rainfall over Ethiopia (with some overestimation). The CC model is able to capture the north-eastern pattern of the rainfall quite well, compared with the CMP dataset; it simulates well the rainfall over Yemen too. The GR model is good in simulating the rainfall in the south-western areas as well as the penetration into the north-western region, whereas it is weak in capturing the system in the north-eastern Peninsula.

Figure 7.

The wet season rainfall (mm/day) climatology for the Arabian Peninsula using (a) CMP, (b) MR, (c) IF, (d) CC, (e) MH and (f) GR model data, averaged over the period 1979–2000.

Overall, the selected five models, on the basis of their interannual variability, are reasonably good in simulating the spatial distribution of the rainfall in the Peninsula during the wet season. It is observed that the MH performance is relatively better among all the selected models in this season.

3.4.2. Wet season mean sea level pressure

The mean sea level pressure climatology for the wet season was obtained for the selected five models averaged over the period 1979–2000 (Figure 8). All the models are able to simulate the low pressure zone over East Africa, called the Sudan low, which is the key location for the maximum rainfall during the wet season in the Arabian Peninsula (Abdelmola, 2009). Most of the models are able to simulate the pressure gradient quite well, i.e. high pressures over the north-eastern Arabian Peninsula and low pressures over the south-western Peninsula. This pressure gradient may be a contributing factor to the Arabian Peninsula receiving heavy rainfall in the northern and the central regions during the wet season. The rainfall in the south-western region of the Peninsula is caused by the penetration of the low pressure systems emanating from the Sudan low, which tend to interact with the mountains in this region, contributing to the rainfall there (see, for instance, Chakraborty et al., 2006).

Figure 8.

The mean sea level pressure (hPa) climatology for the wet season over the Arabian Peninsula using (a) NCEP-SLP, (b) MR, (c) IF, (d) CC, (e) MH and (f) GR. The mean sea level pressure data are averaged over the period 1979–2000. The contour interval is 2 hPa.

3.4.3. Dry season rainfall climatology

The spatial distribution of the dry season rainfall climatology obtained from the selected five models averaged over the period 1979–2000 is compared with the CMP dataset, and is shown in Figure 9. In this season, all the selected models are reasonably good in simulating the spatial distribution of the rainfall, compared with the CMP dataset. The identified models are able to simulate the maximum rainfall over Ethiopia and South Sudan fairly well, relative to the observed patterns. The CC model is able to capture the rainfall relatively better over Oman and Yemen, compared with the CMP dataset, whereas the other models slightly underestimate the rainfall in the eastern side of the Peninsula. During the dry season, the region is under the ITCZ belt, which causes low pressure, facilitating rainfall. The rainfall in this region is also associated with the monsoon activity in the Indian Ocean (Walters and Sjoberg, 1988). Importantly, the dryness of the northern Arabian Peninsula during the dry season is well captured by the selected models.

Figure 9.

The dry season rainfall (mm/day) climatology for the Arabian Peninsula using (a) CMP, (b) EC, (c) G1, (d) CC, (e) HC and (f) HG model data, averaged over the period 1979–2000.

Overall, on the basis of the temporal and spatial analyses, these identified five models are able to simulate the structure of the rainfall quite well, in comparison with the CMP dataset during the dry season.

3.4.4. Dry season mean sea level pressure

The mean sea level pressure climatology obtained from the selected five models averaged over the period 1979–2000 during the dry season of the Arabian Peninsula is compared with the NCEP-SLP dataset, and is shown in Figure 10. In this season, the high pressure zone over the south-western region of the Peninsula is simulated well by the models, in comparison with the observed SLP data. The low pressure zone in the northern Arabian Peninsula is also captured well by the models, therefore the pressure gradient in the models are quite well, compared with the observed SLP data. The subsidence motion of the Hadley cell during the dry season over the Peninsula, which suppresses convection in the region, may contribute towards this high pressure zone (Evans et al., 2004).

Figure 10.

The mean sea level pressure (hPa) climatology for the dry season over the Arabian Peninsula using (a) NCEP-SLP, (b) EC, (c) G1, (d) CC, (e) HC and (f) HG model data, averaged over the period 1979–2000. The contour interval is 2 hPa.

4. Summary and conclusions

The rainfall from the IPCC AR4 GCM simulations is analysed for the period 1979–2000 for the two seasons (wet and dry) over the entire Arabian Peninsula and its two sub-regions (northern and southern). Out of the 22 available GCMs, one group comprising 5 models (MR, CC, MH, GR and IF) is found to perform better than all the others in simulating the wet season rainfall variability, compared with the CMP dataset, for the northern sub-region (above 22°N and 35–60°E). These models fall within (or overlap considerably) the one-sigma standard deviation spread of the CMP wet season mean rainfall (19.78 ± 6.53 mm). The spatial pattern of rainfall and mean sea level pressure during the wet season is quite well captured by these models in comparison with the observed dataset.

Similarly, another group comprising of 5 models (EC, G1, CC, HC and HG) is found to perform relatively better than the other 17 for the dry season rainfall climatology in the southern sub-region (below 22°N and 35–60°E), when compared with the CMP dataset. These models also fall within (or overlap considerably) the one-sigma standard deviation of the CMP dry season mean rainfall (18.26 ± 5.60 mm). These models are also able to capture the rainfall pattern in the southwest of Saudi Arabia, Yemen, Oman and Sudan/Eritrea/Ethiopia. Furthermore, these models are able to capture the seasonal mean sea level pressure patterns quite well, including the low and high pressure zones. The south-western Arabian Peninsula receives rainfall in both the wet and the dry seasons, which are well captured by the selected models.

It is found that only one model, i.e. CC [namely CCCMA-CGCM 3.1 (T47) from Canada] is common to both groups; it is able to simulate the rainfall climatology for both the northern and the southern sub-regions of the Peninsula, both on the temporal and the spatial scales. Therefore, in terms of ensemble mean, the aforementioned two groups of models are found good in simulating the interannual rainfall variability for the present climate of the Arabian Peninsula for its two seasons (wet and dry). This study will facilitate for the future work thus reducing the uncertainties in the rainfall projections for the Arabian Peninsula.

Acknowledgements

This study is supported by King Abdulaziz University and King Abdulaziz City for Science and Technology (KACST), Riyadh, Saudi Arabia, under the Strategic Priorities for Environmental Technology Program (Grant No. 8-ENV125-3). The authors thank the PCMDI for providing the IPCC AR4 simulation datasets through their website. The authors are thankful to the Earth System Physics (ESP) section of the Abdus Salam ICTP, Italy, for providing the re-gridded IPCC AR4-based GCM datasets. The CMAP and the NCEP mean sea level pressure datasets were obtained from the NCEP/NCAR websites.

Ancillary