1. Introduction
The Arabian Peninsula (AP) is one of the driest and most water-limited environments in the world, where the availability of freshwater is of major regional concern (Osman-Elasha 2010; Barlow et al. 2016). There are no rivers with perennial streamflow, and water supplies in the Kingdom of Saudi Arabia are principally derived from rainfall, mined groundwater, and (more recently) desalination (Ouda 2013). Rapid socioeconomic development, expansion of urbanization, agricultural activities, and high population growth are intensifying the stress on water supplies in the region. The reported increase in drought episodes (Ragab and Prudhomme 2000; Kumar et al. 2017) along with the anticipated warmer future climate (Almazroui 2016; Attada et al. 2018, 2019a,b) will further stress the management of water resources. It is thus essential to understand in detail the spatiotemporal variability of rainfall over the AP, to enable its accurate prediction and design efficient strategies for mitigating water scarcity and associated risks.
The availability of accurate datasets is key to studying regional rainfall variability; however, observations and associated rainfall information over the AP are lacking. Global reanalysis datasets are a crucial source of information for regions with limited observed data records. However, global climate reanalyses are still coarse, with resolutions on the order of 50–100 km, not sufficient for investigating regions with complex topography, such as the western and eastern AP (Almazroui et al. 2015; Zittis and Hadjinicolaou 2017). In such areas, regional climate models with finer grid spacing are more appropriate to resolve the local-to-regional processes that interact with the large-scale circulations (e.g., Gao et al. 2017). Validated high-resolution simulations may provide the relevant information at sufficient spatial and temporal scales for data-sparse regions to enable studying and predicting regional rainfall variability.
Most (75%) of the AP annual rainfall falls in winter, from November through April, which is known as the wet season for the region (Almazroui 2011; Dasari et al. 2018). Convective rainfall predominates with high spatial variability over the region, as a result of the strong impact of complex terrain on the initiation and organization of convective processes (Kumar et al. 2015, and references therein). High-resolution modeling with a suitable cumulus parameterization could be used to provide a reliable characterization of regional convection processes. In this respect, Prein et al. (2015) presented a detailed review of the different aspects of high-resolution convection modeling and concluded that the choice of cumulus parameterization scheme (CPS) is an important factor in the simulation of convective precipitation. Cumulus convection has a major effect on the hydrological cycle through the release of latent heat, on the vertical transport of sensible heat, water vapor, and momentum (Han et al. 2016). It is therefore necessary to develop models that accurately represent the interactions between cumulus convection and these movements within a large-scale environment in order to obtain viable weather and climate simulations and subsequent predictions.
Identifying the most suitable CPS for a particular region is crucial for reliable simulation of rainfall. Among the many available CPS schemes, extensive tests have been conducted on the Grell scheme (Grell 1993; Grell and Dévényi 2002), which was originally based on Arakawa and Schubert (1974); the Betts–Miller–Janjić (BMJ) scheme (Betts and Miller 1986; Janjić 1994); and the Kain–Fritch (KF) scheme (Kain and Fritsch 1993; Kain 2004), which was developed based on Fritsch and Chappell (1980). Various sensitivity studies with respect to the CPS have also focused on reproducing climatological rainfall. For instance, Giorgi and Mearns (1999) suggested that the Grell scheme produces a realistic regional climate over the continental United States, although Liang et al. (2004) later reported the superiority of the KF for simulating North American regional climate rainfall.
Almazroui (2016) and Almazroui et al. (2016a) recently used a 50-km regional climate model (RegCM) to investigate the impact of different CPSs in the Middle East and North Africa (MENA) over a limited time period of 5 years. The study reported that rainfall over the AP is quite sensitive to the cumulus parameterization. Similar studies have also been conducted again over short time frames and using relatively coarse-resolution models (e.g., Evans et al. 2004; Almazroui 2016). The complex AP terrain may induce low-level convergence and upslope winds through valleys. This may significantly impact the stimulation and growth of deep convection (Bennett et al. 2011; Wang et al. 2016) and cannot be resolved with coarse resolution models. The sensitivity of convective precipitation over the complex terrain on the AP with respect to different CPSs has yet to be studied using a high-resolution model.
Several studies have investigated the sensitivity of rainfall simulations to CPSs in various regions. For instance, some studies highlighted the importance of choosing a suitable combination of parameterization schemes within the Weather Research and Forecasting (WRF) Model to simulate the rainfall features over the Indian region (Mukhopadhyay et al. 2010; Srinivas et al. 2013; Ratnam et al. 2017). Similar efforts have been conducted for Australia (Evans et al. 2012; Kala et al. 2015), Spain (Argüeso et al. 2011), Europe (Mooney et al. 2013), China (Yuan et al. 2012), South Africa (Crétat et al. 2012; Ratna et al. 2014), and the MENA region (e.g., Zittis et al. 2014; Ehsan et al. 2017).
This study investigates the sensitivity of WRF-simulated rainfall at seasonal scales over the AP with respect to the choice of CPSs based on a high-resolution (5 km) configuration capable of resolving the complex regional topography during the period 2001–16. The selected CPSs are analyzed in terms of their ability to effectively simulate the magnitude and spatial patterns of rainfall and associated physical processes, and are further tuned to enhance the precipitation simulations. The remainder of the paper is organized as follows. Section 2 describes the data and methodology, which also outlines the model configurations and the design of the numerical experiments. Sections 3 and 4 present and analyze the results. A summary of the main conclusions is offered in section 5.
2. Model, data, and methods
a. Model details and experimental configuration
We implemented a nonhydrostatic Advanced Research version of WRF Model (version 3.8.1; Skamarock et al. 2008) with terrain-following coordinates and a constant pressure surface at the top. The model configuration includes two two-way nested domains with respective horizontal resolutions of 15 and 5 km, each with 52 vertical sigma levels. The chosen model domain extends over 30°W–130°E in the zonal direction and 30°S–45°N in the meridional direction and is used to resolve the large-scale atmospheric features and internal dynamics of the system (e.g., Wang et al. 2004; Lucas-Picher et al. 2011; Raju et al. 2015a,b). The initial and 6-hourly boundary conditions are taken from the European Centre for Medium-Range Weather Forecasts Interim Reanalysis (ERAI) data available at a resolution of 0.75°. Sea surface temperature (SST) data are also prescribed from the ERAI dataset. For each winter season, simulations are conducted from 1 November to 1 April, with the first month used as a spinup period to remove spurious effects.
The sensitivity of the model to the following three CPSs is investigated: KF (Kain and Fritsch 1993; Kain 2004), BMJ (Betts and Miller 1986; Janjić 1994), and the scale-aware Grell–Freitas (GF; Grell and Freitas 2014).
KF is a simple mass-flux cloud model for moist updraft/downdraft. It includes a trigger function to initiate convection, compensating for circulation, and closure assumption.
BMJ is a convective adjustment–type scheme that was developed to adjust atmospheric instabilities (toward a reference profile derived from a climatology) by triggering deep convection when sufficient moisture is available.
GF is an ensemble scheme, in which multiple cumulus schemes and variants are run within boxes to obtain an ensemble-mean realization. The ensemble members use different parameters for updraft/downdrafts entrainment/detrainment. It is an updated Grell–Dévényi scheme (Grell and Dévényi 2002), such that the scale awareness is improved by introducing the method of Arakawa et al. (2011). This relaxes the assumptions of traditional parameterizations in which convection is contained within individual model grid columns when the fractional area covered by convection clouds is small.
b. Data and methods
Daily precipitation data with a spatial grid resolution of 0.25° × 0.25° were obtained from the Tropical Rainfall Measuring Mission (TRMM) version 7 (hereafter referred to TRMM; Huffman et al. 2007, 2010). This product combines precipitation estimates from various satellite systems (both infrared and radar) and a surface-gauge analysis on a grid at 3-hourly intervals. Almazroui (2011) compared the TRMM gridded rainfall data with rain gauge observations over the AP and concluded that the TRMM rainfall data is in good agreement with the observations, which was lately confirmed by Hasanean and Almazroui (2015) and Sultana and Nasrollahi (2018). We also evaluated the model-simulated temperature, specific humidity, geopotential height and horizontal wind vectors at different pressure levels against the National Aeronautics and Space Administration’s Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA-2; Gelaro et al. 2017), which is available on an 0.58° × 0.625° grid. Mean monthly cloud information from the Clouds and the Earth’s Radiant Energy System (CERES) database available at a spatial resolution of 1° × 1° was also used to assess the model-simulated cloud characteristics.
Statistical skill scores for mean daily rainfall (mm day−1) during the winter season (DJFM) over the AP and its different subregions for the period 2001–16 from model simulations with different convection schemes and observations.
Statistical skill scores for mean daily 2-m mean, maximum, and minimum temperatures (K) during the winter season (DJFM) over the AP and its different subregions for the period 2001–16 from model simulations with different convection schemes and observations.
3. Results and discussion
We first evaluate the sensitivity of the model simulated rainfall to different CPSs with respect to the TRMM observations. We then analyze the circulation, temperature, moisture, and cloud distributions to understand the dynamic and thermodynamic responses of the model rainfall to the selected convective schemes.
a. Evaluation of seasonal rainfall
Figure 1 shows a comparison of the spatial distribution of winter [December–March (DJFM)] TRMM observed total seasonal rainfall with WRF simulations with the different CPSs, KF, BMJ, and GF over the period 2001–16. High rainfall bands are located over the NAP, the Arabian Gulf, and the Mediterranean region. A considerable amount of rainfall is also observed in the narrow zones over the southwestern AP followed by the central and southern parts of the Sarawat mountain ranges (Fig. 1a). The high rainfall in the NAP is mainly related to the passage of Mediterranean cyclonic storms (midlatitude westerlies). The alignment of the mountains along the coast of the Mediterranean Sea also influences the precipitation distribution in the NAP by creating a pronounced lee effect with rapidly decreasing rainfall toward the northeast. It is also noticeable that precipitation decreases from north to south, with a minimal (or no) rain, observed over the SAP (referred to as a dry zone), particularly over the Rub Al-Khali (the world’s largest desert) region.
Spatial distribution of (a)–(d) mean total winter rainfall (mm) and (e)–(h) its standard deviation (mm day−1) from TRMM, KF, BMJ and GF schemes. (j)–(l) Mean rainfall biases between model simulations and observations (significant at 95% confidence level).
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
These observed rainfall features are simulated reasonably well with KF (Fig. 1b) and BMJ (Fig. 1c). However, GF (Fig. 1d) simulates an extremely dry area over the entire region of Saudi Arabia, except the eastern Mediterranean and the southern Red Sea. Although BMJ and KF underestimate the rainfall compared to observations over the Northeastern AP, KF produces spatial patterns of rainfall that are more realistic than those of BMJ and GF. KF also interestingly produces major precipitation zones over the AP: one located in NEAP and the other over the south-central Red Sea (with 80–150 mm), which is known as the Red Sea convergence zone (RSCZ). Northerly and southerly winds converge in this region and enhance convection, (e.g., De Vries et al. 2013; Viswanadhapalli et al. 2016; Dasari et al. 2018), and this effect is more realistically resolved by KF and BMJ compared to the observed rainfall. The spatial correlation coefficients between the observed rainfall and model simulations (KF, BMJ, and GF) suggest that the superiority of KF, with a higher correlation coefficient 0.71 compared to 0.66 for BMJ and 0.19 for GF.
To achieve good fidelity of the WRF Model with different CPSs, the model should not only capture the mean fields, but also generate variances that are consistent with those of the observations. We therefore compared the standard deviations (SD) of rainfall as they result from the model with KF, BMJ, GF, and TRMM observations (Figs. 1e–h). TRMM (Fig. 1e) shows the highest (>2 mm) SD over the NEAP and eastern Mediterranean regions. This seasonal mean rainfall variability is reproduced best with KF (Fig. 1f) and BMJ (Fig. 1g). The highest rainfall variability occurs over the NEAP compared to the other subregions, as reported in earlier studies (Kang et al. 2015; Abid et al. 2016). The weaker SD in GF is similar to the seasonal average, which has a lower magnitude (Fig. 1d). KF and BMJ reproduce better the details of the rainfall variability in the southern Red Sea where RSCZ-induced rainfall is predominant. Overall, KF exhibits a spatial variability pattern and amplitude that is more in agreement with TRMM than the other two CPSs.
The biases between the observed and simulated rainfall are shown in Figs. 1j–l for KF, BMJ, and GF, respectively. All three schemes produce negative biases over the NEAP and positive biases over the SAP. KF shows a dry bias of approximately 0.8–1 mm day−1, whereas BMJ and GF exhibit significant dry biases of around 1.5–1.8 mm day−1 and more than 2 mm day−1, respectively. These dry biases are reflected in the higher RMSEs for all schemes and are more pronounced over the NEAP for BMJ and GF. The regional averaged RMSEs of rainfall over the AP are 0.29, 0.31, and 0.37 for KF, BMJ, and GF, respectively (Table 1). Overall, the analyses of mean rainfall patterns, SDs, and biases indicate that the model-simulated precipitation sensitive to the CPSs over the AP, with the KF outperforming the other two CPSs.
b. Seasonal evolution of rainfall
The time series of daily rainfall climatology from TRMM rainfall over the AP, NAP, and SAP are presented in Fig. 2 for each CPS. Based on the TRMM observations, the amount of precipitation and rainfall episodes are relatively highest in NAP (Figs. 2a–c). All CPSs simulated these variations in the seasonal evolution of rainfall, but with lower magnitudes than in the observations. The seasonal variability of rainfall from GF is significantly dampened compared to BMJ and KF, which well reproduce the seasonal cycle as observed in TRMM for AP, NAP, and SAP subregions. TRMM also suggests that the largest rainfalls occur during December and March over the AP and NAP, whereas over SAP the high rainfall is recorded during February and early March. With the exception of a few episodes, the simulated rainfall with KF, BMJ, and GF clearly exhibits a significant dry bias, throughout the winter season over the AP and NAP, and a wet bias over the SAP. All three CPSs depict the north–south rainfall gradients, with higher rainfall over NAP and lower over SAP, in agreement with the TRMM observations. The excess amount of rainfall over the NAP is attributed to the passage of midlatitude synoptic storms during winter (Almazroui et al. 2013; Barlow et al. 2016). Out of the three CPSs, the KF-simulated rainfall seasonal cycle closely follows the TRMM rainfall patterns, with daily peaks over the AP and its subregions.
Seasonal cycle of daily rainfall climatology over (a) AP, (b) NAP, and (c) SAP subregions from TRMM, KF, BMJ, and GF cumulus parameterization schemes.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
To validate the model skill in simulating rainfall, different verification scores, namely, the equitable threat score (ETS), bias score (BS), and false alarm rate (FAR) are computed over a wide range of rainfall thresholds based on the contingency table suggested by Bhomia et al. (2019). Figure 3 shows the ETS, BS, and FAR verification score at different rainfall thresholds varying from 1 to 15 mm over NAP. ETS first increases and then decreases for the higher rainfall thresholds of KF and BMJ. For KF, ETS has higher values at all rainfall thresholds compared to BMJ and GF. Note that KF and BMJ show lower skills for higher rainfall thresholds (above 12 mm) whereas GF has the poorest performance. A gradual increase of BS is seen with increased rainfall thresholds. KF shows higher BS compared to BMJ for all the thresholds. FAR is increased rapidly with increased rainfall thresholds, and all CPSs has no/minimal skill for high thresholds. KF has low FAR values compared to BMJ and GF. Overall, the KF has a better rainfall skill compared to the others two CPSs. The impact parameter (Wilks 2006; Raju et al. 2018; Kumar and Shukla 2019) is also estimated to quantify the improvement/degradation of KF in simulating rainfall over GF and BMJ. The analysis (not shown) confirms that the KF has a better skill in simulating rainfall.
Verification skill scores for the simulated rainfall from KF, BMJ, and GF at different rainfall thresholds over the NAP.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
c. Assessment of spatial distribution of near surface temperatures
The presence of complex mountains to the west of AP is generally difficult to handle with numerical models, and may result in temperature bias which ultimately impacts the simulation of precipitation. The accurate representation of steep land–sea thermal gradients is one of the basic requirements for a model to simulate realistic rainfall distributions. To assess the simulated near surface temperature distributions in the model, we plot the mean seasonal winter daily mean temperature (2mT), maximum temperature (Tmax), and minimum temperature (Tmin) are plotted in Fig. 4 at 2-m height for the period 2001–16 from MERRA-2 and the model with the three different CPSs. The mean 2mT from MERRA-2 (Fig. 4a) indicates low temperatures (<288 K) over the NAP, moderate temperatures (288–296 K) over central and western AP, and higher temperatures (>296 K) over SAP and the southern Red Sea (including Sudan and northern Ethiopian regions). KF (Fig. 3b) and BMJ (Fig. 4c) schemes simulate well the high temperature observed over the Rub Al-Khali desert region, and the north–south temperature gradients over the AP and the Red Sea (high temperatures over the southern Red Sea and low temperatures over the northern Red Sea). GF (Fig. 4d) underestimates the near surface 2mT patterns compared to MERRA-2. The temperatures in the southeastern AP are higher than in the southwestern AP, due to the local topography. The lowest temperatures (<275 K) are confined to the NEAP region in all CPSs, and these are in good agreement with MERRA-2. All three CPSs simulate the lowest temperatures over the mountainous region, suggesting the effectiveness of a high-resolution WRF model in reproducing the lowest temperatures, namely, by resolving local topography and their effects on temperatures (e.g., Viswanadhapalli et al. 2016). The comparative statistics between MERRA-2 and the model-simulated 2mT, Tmax, and Tmin for the entire AP and subregions are outlined in Table 2. The spatial correlations between MERRA-2 and WRF with KF, BMJ, and GF are 0.96, 0.96, and 0.93, respectively. Over the NAP (SAP), these correlations are 0.96 (0.92), 0.95 (0.93), and 0.93 (0.91) with KF, BMJ, and GF, respectively; and for the NEAP, the three schemes provide even higher correlation coefficients of 0.97, 0.96, and 0.95, with KF being relatively higher than BMJ and GF.
Spatial distribution of winter season (a)–(d) mean surface temperature (K), (e)–(h) maximum temperature (K), and (i)–(l) minimum temperature (K) averaged over the period 2001–16 from MERRA-2, KF, BMJ, and GF.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
To quantify the ability of WRF to describe mean temperatures, we further conducted different statistical skill score analyses over the AP and its subregions and these skill scores are statistically significant at 95% confidence level with the Student’s t test. The observations exhibit the highest variability over the NEAP, NAP, AP, and SAP with the values of 2.74, 2.59, 2.12, and 1.68 K, respectively. All three schemes produce higher temperature variability over the AP, with KF (2.97 K) performing relatively better than BMJ (3.07 K) and GF (3.30 K). Similar results were also obtained in other subregions of the AP (Table 2). BMJ and GF exhibit strong cold biases of approximately 2.1 and 3.1 K over the AP, whereas the mean bias of KF is around 1.4 K, indicating the superiority of KF in simulating mean temperature patterns.
The salient characteristics of winter mean daily Tmax, such as the significant north–south gradient (higher temperatures over the SAP than the NAP) superimposed with coastal effects and localized orographic features observed in MERRA-2 (Fig. 4e), are well simulated by all CPSs, despite being slightly underestimated. MERRA-2 shows that the highest Tmax (>300 K) occurs over Sudan and the SAP. KF (Fig. 4f), BMJ (Fig. 4g), and GF (Fig. 4h) show low Tmax over the NEAP and high Tmax over the SAP, including the Rub Al-Khali region as in MERRA-2. Relatively, lower Tmax values are noticeable over the eastern side of the Red Sea, suggesting the influence of topography on the maximum temperature distribution in the WRF model. Overall, all CPSs underestimate the Tmax patterns over the AP, although they are able to simulate the north–south Tmax gradient. In terms of spatial distribution, KF simulates a realistic distribution of Tmax similar to that of MERRA-2; however, those of BMJ and GF are not as accurate, with GF significantly underestimating Tmax. Moreover, only KF successfully simulates the three distinct climate regimes (Attada et al. 2019a,b) over the AP that are observed in MERRA-2. In general, this meridional temperature gradient is mainly modulated by western disturbances originating in the Mediterranean region during winter (Viswanadhapalli et al. 2016; Dasari et al. 2018; Attada et al. 2019b). The pattern correlations between the model simulations and MERRA-2 over the AP reveal higher values for KF (0.91) than BMJ (0.89) and GF (0.84). Higher pattern correlations are also obtained for NEAP: 0.94, 0.94, and 0.92 using KF, BMJ, and GF, respectively. The SDs of Tmax are similar in magnitude to those of mean temperatures (Table 2), and Tmax has stronger negative biases compared to mean 2mT, with values of approximately −2.8, −3.5, and −4.1 K over the entire AP for KF, BMJ, and GF, respectively. For the mean temperature, GF leads to higher RMSEs than KF and BMJ over the AP and its subregions.
The comparison of simulated daily minimum temperatures (Tmin) with MERRA-2 (Figs. 4i–l) suggests reasonable agreement for the north–south gradient over the AP and the high minimum over the Red Sea, southeastern AP, and the Arabian Gulf. The simulations also produce lower temperatures over Ethiopia and western Yemen, consistent with those of Almazroui (2012) using the RegCM model. The CPSs leads to significant differences when simulating minimum temperatures over the AP. Although all schemes underestimate minimum temperatures compared to MERRA-2, KF performs better, in terms of the regional distribution of temperatures, than BMJ and GF. The spatial distribution of the mean bias of Tmin (not shown) shows a strong cold bias over the entire AP, in agreement with the findings of Viswanadhapalli et al. (2016). Furthermore, higher correlations with MEERA are obtained with KF simulated Tmin patterns over the AP and its subregions (Table 2). The SD of Tmin suggests that all CPSs exhibit a higher SD over NEAP than the other subregions, but are lower compared to MERRA-2. KF has less RMSE over the AP (1.4 K), NAP (1.2 K), SAP (1.6 K), and NEAP (1.5 K) compared to BMJ and GF (Table 2).
d. Seasonal cycle of daily mean, maximum, and minimum temperatures
Figure 5 depicts the seasonal cycles of daily mean temperature, Tmax, and Tmin over the AP, NAP, and SAP over the period 2001–16. The seasonal cycle of daily temperatures from MERRA-2 over the AP (Fig. 5a), NAP (Fig. 5b), and SAP (Fig. 5c) indicates peak temperatures during the last week of February. This seasonal evolution of temperatures is well reproduced by WRF using all CPSs. Overall, the temperature evolutions are similar in all climatic zones and are well captured (with some deviations) compared to MERRA-2. Over the AP and SAP, KF is better at producing mean temperatures, while GF is slightly better over the NAP. All CPSs simulate the peak temperatures earlier than MERRA-2.
Seasonal cycle of daily mean surface temperature (K), maximum temperature (K), and minimum temperature (K) climatology over (a) AP, (b) NAP, and (c) SAP subregions from MERRA-2, KF, BMJ, and GF cumulus parameterization schemes averaged over the period 2001–16.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
The seasonal cycle of Tmax (Figs. 5d–f) is similar to that of mean temperature, but varies between 292 and 302 K. All three schemes capture the evolution of Tmax over the AP with notable underestimation compared to MERRA-2. KF performs better over the AP than over NAP and SAP, for which it produces cold biases. In the case of NAP, GF shows the best phase of Tmax evolution, while KF and BMJ depict colder biases. KF seems to not perform as well as BMJ and GF in simulating the maximum temperature evolution. In MERRA-2, the seasonal evolution of Tmin (Figs. 5h,i) varies between 284 and 286 K over the AP, between 281 and 283 K over the NAP, and between 289 and 294 K over the SAP. All three schemes simulate these evolutions of Tmin over the AP subregions with considerable discrepancies. They also underestimate the seasonal cycle compared to MERRA-2, with KF performing relatively better than BMJ and GF.
e. Monthly variations in rainfall and temperature biases
The subregional average precipitation and temperature biases computed for the individual months of December, January, February, and March between the model simulations with different CPSs and TRMM observations are presented in Fig. 6. Monthly variations in the rainfall biases of KF are smaller than those of BMJ and GF for all regions. Over the AP, the rainfall bias ranges between −0.1 and 0.21 mm day−1 with KF, between −0.1 and 0.30 mm day−1 with BMJ, between −0.19 and 0.35 mm day−1 with GF. All CPSs simulate the wet bias in the month of February and March. Strong wet biases are obtained with BMJ over NAP during the month of December and while dry bias with KF and GF. The wet bias in all CPSs over the SAP is observed during February and March.
Subregional average bias of (a)–(c) rainfall and (d)–(f) 2-m air temperature from KF, BMJ, and GF over the (a) AP, (b) NAP, and (c) SAP during winter.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
The average mean temperature bias (Figs. 6d–f) over the AP and its subregions for individual months indicates that the CPSs in WRF produce a cold bias. A stronger cold bias from about −2 to −4 K in GF, from about −0.5 to −3 K in BMJ, and from about −0.2 to −2 K with KF is obtained in all months. Overall, the results indicate that the KF leads to better simulations of mean surface temperatures. Similar biases are also obtained for maximum and minimum temperatures. From Table 2, the regional temperatures error statistics suggest lower errors and highest correlations with KF.
f. Assessment of circulation patterns
Figure 7 shows the spatial distribution of seasonal mean winter wind flow and geopotential height patterns at 850 hPa from MERRA-2, and WRF with KF, BMJ, and GF. The results shows the salient winter circulation patterns of AP, such as the strong anticyclonic circulation pattern (clockwise rotation) between the central to SAP (referred to as the Arabian anticyclone), the strong westerly winds passing through the Mediterranean Sea toward the NAP, the more pronounced wind circulation from the Arabian Gulf to the central AP and NAP, and the RSCZ over the central Red Sea (with its eastern plank toward the AP and its western plank that has moved toward the Sudan region). The geopotential height 850 hPa also indicates the presence of the Arabian anticyclone (high geopotential heights) over the eastern AP, which is an important modulator of rainfall in the region (e.g., Dasari et al. 2018).
Winter seasonal mean (left) low-level (850 hPa) and (right) upper-level wind speed (shaded; m s−1), direction (vectors), and geopotential height (m) from (a),(e) MERRA-2, (b),(f) KF, (c),(g) BMJ, and (d),(h) GF averaged over 2001–16.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
The Arabian anticyclonic pattern is well simulated by KF (Fig. 7b) and BMJ (Fig. 7c), although slightly shifted westward in BMJ (Fig. 7a), while GF (Fig. 7d) misses its location as compared to MERRA-2. BMJ- and KF-simulated winds over the Gulf of Aden are in good agreement with MERRA-2, but GF overestimates these winds. KF yields a more realistic simulation of the Arabian anticyclone, RSCZ, and westerly winds. It also shows the southerly flow from the Arabian Sea toward land in agreement with MERRA-2 flow patterns. The midlevel winds (500 hPa) during winter (not shown) from MERRA-2 show a strong anticyclonic circulation over the southern Red Sea and the Sudan region, and these are better simulated by KF and BMJ compared to GF. Over the northern AP, strong midtropospheric westerlies are observed in both KF and BMJ, and MERRA-2, which act as waveguides for the Mediterranean westerly systems to generate rainfall over the eastern AP and NEAP.
The seasonal mean distribution of sea level pressure (SLP) during winter (not shown) exhibits low pressure systems over east Africa (the Sudan low) and the south western AP (including the southern Red Sea), and a high pressure system over the NEAP. This meridional pressure gradient (~5 hPa) plays an important role in the generation of the AP winter rainfall. Rainfall in the southwestern AP is developed by the penetration of the low pressure system emanating from the Sudan low and the Red Sea low, which interacts with the southwestern AP mountains and trigger rainfall (e.g., Chakraborty et al. 2006; Dasari et al. 2018). These winter pressure patterns are well simulated by KF, BMJ, GF compared to MERRA-2, while the Sudan lows are better simulated by KF.
The upper tropospheric winds (200 hPa) from MERRA-2, KF, BMJ, and GF (Figs. 7e–h) show the presence of the subtropical westerly jet (SWJ) over the AP, which has highest regional wind speeds of approximately 45 m s−1. This jet is often referred to as the Middle East jet stream and is an important dynamical precipitation factor in the AP, acting as a waveguide for westerly disturbances (e.g., Athar 2014; Kumar et al. 2016; Dasari et al. 2018; Attada et al. 2019a). The position and intensity of the upper tropospheric circulation are well simulated by KF and BMJ, whereas GF simulates a northward-shifted SWJ compared to MERRA-2. The upper tropospheric geopotential height patterns indicate that the north–south gradient in geopotential height over the AP is better simulated by KF (Fig. 7f) compared to BMJ and GF.
In the upper troposphere, synoptic transients (western disturbances) are pronounced during winter, and these have a significant impact on AP winter rainfall (Yadav et al. 2013; Kang et al. 2015; Almazroui et al. 2016b; Attada et al. 2019a; Dasari et al. 2019, unpublished manuscript). These eastward-moving systems are a result of baroclinic and barotropic energy sources that are generally guided by upper tropospheric jet streams centered between 25° and 35°N (e.g., Hoell et al. 2015). We thus investigated the sensitivity of synoptic transients during winter to the CPSs over the period 2001–16. The synoptic variability is shown in terms of 2–8-day filtered upper-level zonal winds. Meridional winds during winter are a good indicator for upper level synoptic transient activity (Fig. 8) (Barlow et al. 2016). In MERRA-2 (Fig. 8a), the mean synoptic transients during winter in the zonal and meridional wind components are pronounced over the NAP and Arabian Gulf during the entire study period. These transients are relatively low in the SAP compared to the NAP. KF (Fig. 8b) and BMJ (Fig. 8c) are better able to produce synoptic transients as compared to MERRA-2, while the locations of these transients in GF (Fig. 8d) are shifted northward, associated with the northward shift in the GF-simulated subtropical westerly jet.
Spatial distribution of winter mean upper-tropospheric (200 hPa) synoptic transients in the zonal (shaded) and meridional wind components (contours) from MERRA-2 and three different cumulus parameterization schemes. Red contours indicate the wind maxima (above 40 m s−1) of upper-level (200 hPa) zonal winds (m s−1).
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
We further analyzed the storm tracks to examine the influence of Mediterranean storms on the AP winter rainfall. Figure 9 plots the storm tracks based on the local vorticity maxima at 850-hPa level (Flaounas et al. 2014) as extracted from WRF simulations with KF, BMJ, and GF and compared with the corresponding tracks from MERRA-2. Both model simulations and reanalysis fields show that most of the storm tracks originate in the Mediterranean Sea and propagate eastward before dissipating over the northern AP. These storm passages confirm their important contribution to the rainfall over the AP. The simulation of these storm tracks with KF is relatively in closer agreement with MERRA-2 than BMJ and GF. Note that the storms simulated by WRF that originate over the Red Sea region and propagate northward are not observed in the reanalysis. These convective storms, triggered by the RSCZ that form over the central Red Sea move inland into AP. The horizontal length scales of these storms are about 3–5 km and require a high-resolution model to properly simulate these features. Our high-resolution configuration is able to reproduce these small-scale convective activities and their propagation toward the AP. Overall, the intrusion of midlatitude synoptic transients toward the AP, in conjunction with the low-level northward advection of warm and moist air from the Red Sea and Arabian Sea, prompts the dynamic and thermodynamic instabilities to enhance rainfall during winter (e.g., Chakraborty et al. 2006; Kumar et al. 2015; De Vries et al. 2016; Dasari et al. 2019), and this is realistically produced by KF.
Storm tracks associated with AP winter rainfall for the period 2001–16 from (a) reanalysis, (b) KF, (c) BMJ, and (d) GF. Here we present the tracks cover the whole lifetime of the storms from their formation to dissipation.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
g. Analysis of moisture distribution and its dynamics
The distribution of moisture and its dynamics are key factors determining the variability of rainfall and characterizing the vertical distribution of specific humidity is crucial for understanding moist convective processes over the AP (Chakraborty et al. 2006; Babu et al. 2011, 2016; Kang et al. 2015; Dasari et al. 2018). Analysis of different datasets suggests that moisture budgets over the Mediterranean Sea and the Red Sea, have strong links with AP winter rainfall (Jin et al. 2011; Şahin et al. 2015; Dasari et al. 2018; Zolina et al. 2017). Zolina et al. (2017) pointed that the moisture transportation in the surface layer is dominated by breezes driven by SST, and the advection of moisture above the boundary layer is controlled by regional circulation patterns. This section analyzes the characteristics of the mean specific humidity at different tropospheric levels during winter over the AP as simulated by the WRF model with the different CPSs and from MERRA-2 (Fig. 10).
Spatial distribution of seasonal mean low-level (850 hPa), midlevel (500 hPa), and upper-level (200 hPa) specific humidity (contours; g kg−1) during winter from MERRA-2, KF, BMJ, and GF.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
The spatial distribution of low-level (850 hPa) specific humidity (Fig. 10a) from MERRA-2 exhibits highest values of approximately 10 g kg−1 over the southern Red Sea, about 5–6 g kg−1 over the SAP (between 10° and 23°N), and below 3 g kg−1 over the NAP. KF (Fig. 10b) and BMJ (Fig. 10c) reproduce these regional changes in the specific humidity distribution over the AP, while GF underestimates them over the southern Red Sea and SAP regions, in conjunction with the weaker winds (Fig. 10d). All three schemes show the north–south gradient in the lower tropospheric moisture over the Red Sea, but GF provides lower values, particularly over the Arabian Gulf and NEAP. MERRA-2 (Fig. 10e) shows a narrow zone of specific humidity at a pressure level of 500 hPa from east Africa to the northeastern AP through southwestern AP. The highest specific humidity is reached over the southern Red Sea and Sudan regions, whereas the lowest specific humidity is found over the Arabian and Mediterranean regions. KF (Fig. 10f) exhibits a clear maximum specific humidity extending from the equatorial regions and eastern Africa toward the AP, which is typical of tropical plumes over the region (Ziv 2001; Rubin et al. 2007; Tubi and Dayan 2014). These tropical plumes are primarily confined to the winter and contribute to the light to heavy widespread rainfall across arid desert regions like the AP. These plumes follow the southward penetration of midlatitude troughs that are associated with an intensified thermal wind and longer jet streaks (e.g., Tubi and Dayan 2014). BMJ (Fig. 10g) also simulates these atmospheric plumes of specific humidity, but significantly underestimates their magnitude. GF (Fig. 10h) fails to simulate the midtropospheric specific humidity band. Moisture availability in the GF is therefore meager, which results in a dry bias in the precipitation simulation in the NAP. The comparison between the simulated upper tropospheric (200 hPa) specific humidity with MERRA-2 (Figs. 10i–l) indicates an increased moisture content in KF compared to BMJ and GF. This is due to the higher values of extended specific humidity plumes from the equatorial regions and eastern Africa toward the AP in KF. It also shows that most of the moisture is confined to the SAP and southern Red Sea regions compared to the NAP during winter.
To investigate the moisture source that triggers moist convection and associated rainfall over the AP, the composite winter means of vertically integrated (from the surface to 400 hPa) moisture transport from the model simulations and MERRA-2 are analyzed and presented in Fig. 11, where the vectors represent the resultant moisture transport components of zonal and meridional moisture components. MERRA-2 shows that the moisture fluxes occur predominantly over the Arabian Sea and Red Sea and are driven by the Arabian anticyclone (Fig. 11a). Furthermore, the subtropical jet is associated with an anticyclonic flow over the south of the AP, which advects moisture from the Red Sea and the Arabian Sea. MERRA-2 suggests that moisture originates in the Arabian Sea, Gulf of Aden, and the southern Red Sea as a result of the formation of the Arabian Anticyclone and the effect of the Indian winter monsoon flow (Dasari et al. 2018). It can be discerned that a significant amount of moisture is transported by the westerly winds from the eastern Mediterranean toward the NAP region. Compared to MERRA-2, KF (Fig. 11b) provides a more realistic representation of moisture transport and the location of maximum moisture transport (more than 130 kg m−1 s−1) over the southern Red Sea and the Gulf of Aden. The model simulations with different CPSs confirm that the Red Sea is a major contributor of moisture for the AP precipitation (Zolina et al. 2017; Dasari et al. 2018; Sandeep and Ajayamohan 2018). The BMJ (Fig. 11c) simulates a similar vertical integrated moisture transport structure to KF and MERRA-2, but with a weaker magnitude. The BMJ also shows that moisture from the southern Red Sea is advected toward eastern Africa. In contrast, GF (Fig. 11d) fails to simulate the locations of maximum moisture transport, and both GF and BMJ display weaker moisture transport flux vectors compared to KF and MERRA-2, which leads to a dry bias in rainfall (Fig. 6). Our analysis of the vertically integrated horizontal moisture fluxes suggests that the availability of higher moisture during winter provides a favorable condition for generating rainfall over the AP and is also associated with weather disturbances migrating from the Mediterranean region. Therefore, proper representation of the sources of moisture in the model is essential to properly resolve the mechanisms for developing moist convection and the associated dynamics of precipitation over the AP. KF and BMJ successfully reproduce these features while GF fails to do so.
Vertically integrated moisture transport during winter season from MERRA-2, KF, BMJ, and GF for the period 2001–16.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
4. Vertical structures of dynamic and thermodynamic profiles
This section evaluates the three-dimensional representation of the atmosphere in the model to understand the winter dynamics. Specifically, it focuses on the vertical profiles of temperature and moisture that are interrelated with convective processes, which are essential for initiating convective activity (e.g., Raju et al. 2015a; Martínez-Castro et al. 2017). The representation of these profiles in the model is determined by the convective schemes and is connected with the precipitation formation process.
Seasonally averaged vertical profiles of different variables were averaged over the NAP (with respect to the highest precipitation in a subregion) from MERRA-2, KF, BMJ, and GF, and the results are presented in Fig. 12. In general, the vertical distribution of temperature decreases with height in MERRA-2 and the model with the three CPSs (Fig. 12a). However, KF and BMJ agree better with MERRA-2, albeit for cold biases in the lower troposphere, while GF exhibits a strong cold bias in the lower troposphere and a warm bias in the middle to upper troposphere. Overall and compared to MERRA-2, the temperature distribution in KF is slightly better than that of BMJ and is far superior to that of GF. Warm temperature biases in the upper troposphere are systematically stronger in GF, consistent with weak/scanty rainfall amounts.
Area averaged winter mean vertical profiles of (a) temperature, (b) specific humidity, (c) zonal wind, (d) relative vorticity, and (e) apparent heat source over the NAP subregion from MERRA-2 and model simulations for the period 2001–16.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
The seasonally averaged vertical profile of specific humidity over the NAP (Fig. 12b) shows high magnitudes at the surface (about 6.5 g kg−1) and a gradual decrease with height thereafter. The vertical variations in specific humidity are well simulated by the model with all CPSs. KF exhibits higher moisture in the lower troposphere compared to MERRA-2, whereas BMJ is dry at the surface and in the mid- to upper troposphere; however, its results are in good agreement with MERRA-2 in the lower troposphere. GF configuration exhibits a dry bias of approximately 2 g kg−1 in the entire troposphere over the AP, which is further corroborated by the underestimation of rainfall. Compared to the other schemes, the vertical profile configuration of KF is overall closer to that of MERRA-2.
The vertical distribution of zonal winds shows lower tropospheric weak westerlies and mid-to-upper tropospheric strong westerlies over the NAP (Fig. 12c). The highest zonal wind speed (45 m s−1) occurs at 200 hPa over the NAP and is associated with the subtropical westerly jet. This vertical zonal wind structure in KF agrees better with that of MERRA-2 than BMJ and GF, where BMJ simulates stronger zonal wind speeds at 200 hPa and GF underestimates the zonal wind in the entire troposphere. These results suggest that the BMJ (GF) simulated zonal wind is strongly (weakly) driven by the subtropical jet. The mean vertical profile of the relative vorticity from MERRA-2 (Fig. 12d) shows a cyclonic circulation (positive values) in the upper troposphere (from 600 to 100 hPa) and an anticyclonic circulation (negative values) in the surface to the middle troposphere (from surface to 600 hPa). In KF, a low-level anticyclonic vorticity and cyclonic vorticity aloft is noticeable, in agreement with the observations. The relative vorticity profile is also reproduced by BMJ, but with considerable discrepancies compared to MERRA-2, while the results of GF are completely offset from the observations, except at the surface.
The time-averaged vertical distributions of diabatic heating over the north Arabian Peninsula region from MERRA-2 and WRF simulations with KF, BMJ and GF are shown in Fig. 12e. MERRA-2 shows maximum diabatic heating in the lower (upper) troposphere below 900 hPa (above 150 hPa), whereas strong diabatic cooling with two maxima in the middle troposphere (between 900 and 150 hPa), indicating the dominance of radiative cooling. KF, BMJ and GF simulated similar vertical structures of diabatic heating as those of MERRA-2, but not in terms of magnitudes. As compared to BMJ and GF, the vertical profile of apparent heat source from KF is in better agreement with MERRA-2. GF shows a large deviation in the vertical profile compared to MERRA-2, with a maximum surface heating and strong diabatic cooling in the upper troposphere. Discrepancies in representing the diabatic heating profile by BMJ and GF lead to discrepancies in instability, and in turn precipitation biases. These profiles are qualitatively similar to those reported in earlier studies (e.g., Shay-El and Alpert 1991).
a. Evaluation of cloud distribution
In this section, we evaluate the efficiency of different convective schemes in representing different cloud types. Clouds are evidently important for providing the precipitation distribution (e.g., Diaz et al. 2015) and cloud processes are often poorly represented in numerical models (e.g., Randall et al. 2003; Stevens and Bony 2013). We present different cloud levels (low, middle, and high clouds) during winter from CERES observations along with those from the model simulations using the three different CPSs (Fig. 13).
Spatial distribution of seasonal mean (a)–(d) low-level cloud cover, (e)–(h) mid-level cloud cover, and (i)–(l) high-level cloud cover during winter from observations and model simulations for the period 2001–16.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
Observations (Fig. 13a) indicate that a high percentage (more than 25%) of low-level clouds (which have a cloud-top height below 700-hPa level) are located over the southern and central Red Sea, Mediterranean Sea, Arabian Gulf, and the Gulf of Aden. Other parts of the Red Sea and the NAP show a 10%–20% coverage of low-level clouds; 10%–15% of low-level clouds are distributed over the NAP and NEAP regions, and low-level cloud coverage is limited over the land regions of the SAP. This indicates that most of the low-level clouds over the Red Sea are associated with the RSCZ, which is a shallow system that creates maritime stratocumulus clouds, and this is also observed over the Arabian Gulf and the Mediterranean Sea. KF (Fig. 13b) and BMJ (Fig. 13c) are able to well simulate the low-level cloud distribution, slightly underestimated, over the Red Sea and AP. Although GF captures the correct low-level cloud over the RSCZ and Arabian Gulf regions, it fails to simulate the low-level clouds in the NAP (Fig. 13d). All the schemes fail to reproduce the observed cloud structure in the SAP.
The observed middle clouds over the region with cloud-top heights between the 350- and 700-hPa levels (Fig. 13e) show maximum cloud coverage over the NAP region (>10%–15%), while the SAP is not covered by these altostratus cloud types. KF (Fig. 13f) and BMJ (Fig. 13g) provide a proper representation of midlevel clouds over the NAP, but with excess coverage compared to MERRA-2, whereas GF fails to produce these midlevel clouds and confines them to the far north of the domain. Overall, KF simulates a north–south distribution of midlevel clouds that is more in agreement with the observations than the other two CPSs. High-level clouds from MERRA-2 (Fig. 13i) show less amount of cirrus clouds compared to low- and midlevel clouds during winter, whereas KF (Fig. 13j) and BMJ (Fig. 13k) simulated more high-level clouds compared to the observations. This is more noticeable over the SAP for GF, which simulates high values of cloud coverage. GF results over the NAP well match the observed high-level clouds during winter. However, the locations of high clouds are better depicted by KF and BMJ, as compared to MERRA-2. Overall, KF and BMJ outperform GF in simulating the low- and midlevel clouds, but they struggle with the simulation of high-level clouds during winter over the AP.
b. Vertical distribution of cloud microphysical properties
The vertical structures of cloud hydrometeors have a large impact on precipitation processes (e.g., Rajeevan et al. 2013), and are thus investigated here. We focus in particular on the cloud microphysical properties over the NAP, which receives the largest amount of rainfall.
The vertical profiles of liquid hydrometeors (cloud and rainwater) and solid hydrometeors (graupel, ice, and snow) over the NAP are presented in Fig. 14. Because of the lack of data, the validation of the model-simulated hydrometeors is only conducted for cloud water and ice mixing ratio profiles using MERRA-2. The results suggest an increase in cloud water from the surface to 700 hPa, and thereafter a decrease with height in both MERRA-2 and the model simulations. The main cloud deck (maximum peak of cloud water mixing ratio) is located at 700 hPa in KF and BMJ, is in agreement with the observations (Fig. 14a). KF shows slightly higher values of cloud water than BMJ, while GF fails to produce cloud water, leading to a significant underestimation and shift of the maxima to lower levels at around 900 hPa.
Spatial and temporal means of vertical profiles of cloud hydrometeors provided by the different schemes, corresponding to the winter season and computed for northern AP region: (a) cloud water, (b) rainwater, (c) graupel, (d) ice, and (e) snow.
Citation: Journal of Hydrometeorology 21, 5; 10.1175/JHM-D-19-0114.1
The vertical profile of the rainwater maxing ratio (Fig. 14b) suggests that the maximum amount of rainwater is available at a pressure level of 750 hPa (slightly below cloud water) in both KF and BMJ. Raindrops are the only precipitating hydrometeor at the lowest level of the atmosphere as can be seen in Fig. 14b for both BMJ and KF. The rainwater mixing ratio produced from GF is different than that of KF and BMJ. For the graupel mixing ratio (Fig. 14c), BMJ simulates the maximum peak at 650 hPa reasonably well, whereas GF fails to achieve this. The ice mixing ratio (Fig. 14d) has a maximum peak at 300 hPa (above the freezing level), which is more underestimated in BMJ than in KF. GF fails to distribute the ice mixing ratio over the NAP. All CPSs leads to a significant underestimation of cloud ice compared to MERRA-2. The ice hydrometeor profile is the key microphysical processes in the formation of precipitation. As the ice crystals grows, they become heavier than snow particles before they start falling, which leads to growth of graupel by accretion of supercooled water and then melt just above the surface to form rainfall (e.g., Gao et al. 2016). Although KF underestimates this process, it performs slightly better than BMJ and GF. The failure of BMJ and GF in reproducing this important process could be one of the reasons for their simulated dry rainfall biases over the AP. The vertical profile of the snow mixing ratio (Fig. 14e) indicates that the upper troposphere (450 hPa) is characterized by the maximum amount of snow, with KF exhibiting a higher snow mixing ratio than BMJ and GF. It is thus assumed that the sources of systematic model errors (in Fig. 14) are related to the cloud modeling in the different convective schemes, including the model vertical resolution.
Overall, liquid hydrometers are formed below the freezing level where warm precipitation processes occur, and ice, graupel, and snow are distributed beyond the freezing level and are mainly related to cold precipitation processes over the AP. Therefore, an improved representation of the vertical structure of cloud hydrometeors is necessary for providing realistic model simulations of AP winter rainfall; this is not actualized in GF, which results in a poorer rainfall simulation skill than BMJ and KF.
5. Summary and conclusions
This study evaluated the performance of the WRF model and its sensitivity to three CPSs [Kain–Fritsch (KF), Betts–Miller–Janjić (BMJ) and the scale-aware Grell–Freitas (GF)] for seasonal scale simulations of AP winter rainfall during the period 2001 to 2016, and then elucidated the associated regional dynamics. We used the WRF Model configured on two two-way nested domains with respective horizontal resolutions of 15 and 5 km to capture the detailed rainfall distribution and associated underlying processes. The model simulated variables were validated against satellite observations and reanalysis datasets, before investigating the sensitivity of the three CPSs.
Our results suggest that the model-simulated seasonal scale AP winter rainfall is sensitive to the CPSs. KF appears to produce realistic geographic distributions, and its simulated seasonal climatology of precipitation and air temperature are in good agreement with the observations compared to BMJ and GF. All CPSs exhibit, however, dry biases in rainfall and cold biases in mean, maximum, and minimum 2-m temperatures. Overall, KF depicts higher spatial correlations with fewer errors for temperature (including maximum and minimum) and precipitation compared to BMJ and GF. Furthermore, the standard deviation of temperature and precipitation are also better reproduced by KF, while BMJ produces better variability than GF (on par with KF) over some parts of the AP. The analysis of daily mean regional precipitation indicates that BMJ and GF fail to well reproduce the seasonal evolution of rainfall compared to the observations and KF. Precipitation over the AP is better captured by KF albeit with a slight underestimation.
The Arabian anticyclone, which is one of the main characteristics of low-level circulation, is well captured by KF and BMJ, but its position is shifted in GF. Strong westerly winds passing through the Mediterranean Sea toward the NAP and the winds blowing from the Arabian Gulf to the central and NAP regions are better simulated by KF than by BMJ and GF. In the case of upper tropospheric circulation, KF and BMJ simulate well the SWJ (in terms of location and strength) as compared to MERRA-2. The position of SWJ is important and acts as a waveguide for westerly disturbances and associated precipitation in the AP. Overall, KF is better able to represent the eastward moving storm systems (large scale synoptic transients and storm tracks) that are guided by the SWJ. The proper representation of moisture sources in KF enables the development of moist convection and associated precipitation dynamics in the AP; both BMJ and GF generally fail to simulate these structures.
The simulated vertical profiles of several atmospheric variables, such as temperature, specific humidity, zonal wind, and relative vorticity were also evaluated, suggesting that the KF exhibits higher fidelity with the observed atmospheric structures compared to BMJ and GF, which leads to better vertical thermodynamic structures and realistic convective precipitation. The discrepancies between the different schemes reveal that the proper simulation of different cloud types and associated cloud hydrometer responses enables KF to better simulate the rainfall variability over the AP.
This study examined the differences between the three CPSs in terms of simulating the AP winter rainfall, but did not attempt to determine which processes within the schemes produce the differences outlined here. Liang et al. (2004) suggested that the KF incorporates detailed cloud microphysics and entrainment and detrainment between clouds and environment, which are not described in the two other convective schemes. Moreover, subgrid-scale cloud–radiation interactions within the KF have been found to be important (Alapaty et al. 2012; Herwehe et al. 2014) in the simulation of precipitation. The analysis of heat source (diabatic heating) suggests that KF more accurately simulates the thermodynamic feedback to rainfall. This further improves the representation of the vertical structure of cloud hydrometeors, which in turn better resolves the precipitation distribution. Further, the superiority of the KF can also be explained by its appropriate treatment of convective available potential energy as a triggering function, and its treatment of deep convection with strong updrafts, downdrafts, and environmental mass fluxes that adjust precipitation. It should also be noted that the difficulties in accurately simulating AP precipitation could be caused by deficiencies in other related physical processes, such as the subtropical westerly jet, synoptic transients, and cloud microphysics (Dai and Trenberth 2004). Based on the results of our study, GF seems to be relatively less suitable for simulation of AP rainfall with WRF.
Our study investigated the sensitivity of winter rainfall over the AP with respect to the convective parameterization schemes within a high-resolution (5 km) regional modeling framework. Several other studies advocated for the use of CPSs at this resolution, suggesting improved simulations compared to fully explicit simulations (e.g., McMillen and Steenburgh 2015; Lind et al. 2016). Convective resolving models were not investigated yet for predicting the AP rainfall; this will be investigated in our future work. Note that the treatment of dust in the model may play an important role in the simulation of AP rainfall through the aerosol–radiative feedback mechanisms. The complex interaction processes between aerosols and rainfall will also be investigated in our future work.
Acknowledgments
This research work was supported by the Office of Sponsored Research (OSR) at King Abdulla University of Science and Technology (KAUST) under the “Virtual Red Sea Initiative” (Grant REP/1/3268-01-01). All simulations were conducted on the KAUST Super Computational facility SHAHEEN supported by the KAUST Supercomputing Laboratory (KSL). The authors thank three anonymous reviewers for their constructive and insightful comments.
REFERENCES
Abid, M. A., F. Kucharski, M. Almazroui, and I. Kang, 2016: Interannual rainfall variability and ECMWF-Sys4-based predictability over the Arabian Peninsula winter monsoon region. Quart. J. Roy. Meteor. Soc., 142, 233–242, https://doi.org/10.1002/qj.2648.
Alapaty, K., J. A. Herwehe, T. L. Otte, C. G. Nolte, O. R. Bullock, M. S. Mallard, J. S. Kain, and J. Dudhia, 2012: Introducing subgrid-scale cloud feedbacks to radiation for regional meteorological and climate modeling. Geophys. Res. Lett., 39, L24809, https://doi.org/10.1029/2012GL054031.
Almazroui, M., 2011: Calibration of TRMM rainfall climatology over Saudi Arabia during 1998–2009. Atmos. Res., 99, 400–414, https://doi.org/10.1016/j.atmosres.2010.11.006.
Almazroui, M., 2012: Dynamical downscaling of rainfall and temperature over the Arabian Peninsula using RegCM4. Climate Res., 52, 49–62, https://doi.org/10.3354/cr01073.
Almazroui, M., 2016: RegCM4 in climate simulation over CORDEX-MENA/Arab domain: Selection of suitable domain, convection and land-surface schemes. Int. J. Climatol., 36, 236–251, https://doi.org/10.1002/joc.4340.
Almazroui, M., M. Adnan Abid, H. Athar, M. Nazrul Islam, and M. Azhar Ehsan, 2013: Interannual variability of rainfall over the Arabian Peninsula using the IPCC AR4 global climate models. Int. J. Climatol., 33, 2328–2340, https://doi.org/10.1002/joc.3600.
Almazroui, M., R. Dambul, N. Islam, and P. J. Jones, 2015: Atmospheric circulation patterns in the Arab region and its relationships with Saudi Arabian surface climate: A preliminary assessment. Atmos. Res., 161–162, 36–51, https://doi.org/10.1016/j.atmosres.2015.03.014.
Almazroui, M., M. N. Islam, A. K. Al-Khalaf, and F. Saeed, 2016a: Best convective parameterization scheme within RegCM4 to downscale CMIP5 multi-model data for the CORDEX-MENA/Arab domain. Theor. Appl. Climatol., 124, 807–823, https://doi.org/10.1007/s00704-015-1463-5.
Almazroui, M., S. Kamil, K. Ammar, K. Keay, and A. O. Alamoudi, 2016b: Climatology of the 500-hPa Mediterranean storms associated with Saudi Arabia wet season precipitation. Climate Dyn., 47, 3029–3042, https://doi.org/10.1007/s00382-016-3011-0.
Arakawa, A., and W. H. Schubert, 1974: Interaction of a cumulus cloud ensemble with the large-scale environment, Part I. J. Atmos. Sci., 31, 674–701, https://doi.org/10.1175/1520-0469(1974)031<0674:IOACCE>2.0.CO;2.
Arakawa, A., J.-H. Jung, and C.-M. Wu, 2011: Toward unification of the multiscale modeling of the atmosphere. Atmos. Chem. Phys., 11, 3731–3742, https://doi.org/10.5194/acp-11-3731-2011.
Argüeso, D., J. M. Hidalgo-Muñoz, S. R. Gámiz-Fortis, M. J. Esteban-Parra, J. Dudhia, and Y. Castro-Diez, 2011: Evaluation of WRF parameterizations for climate studies over Southern Spain using a multistep regionalization. J. Climate, 24, 5633–5651, https://doi.org/10.1175/JCLI-D-11-00073.1.
Athar, H., 2014: Trends in observed extreme climate indices in Saudi Arabia during 1979–2008. Int. J. Climatol., 34, 1561–1574, https://doi.org/10.1002/joc.3783.
Attada, R., R. K. Yadav, R. K. Kunchala, H. P. Dasari, O. Knio, and I. Hoteit, 2018: Prominent mode of summer surface air temperature variability and associated circulation anomalies over the Arabian Peninsula. Atmos. Sci. Lett., 19, e860, https://doi.org/10.1002/asl.860.
Attada, R., H. P. Dasari, A. Parekh, J. S. Chowdary, S. Langodan, O. Knio, and I. Hoteit, 2019a: The role of the Indian summer monsoon variability on Arabian Peninsula summer climate. Climate Dyn., 52, 3389–3404, https://doi.org/10.1007/s00382-018-4333-x.
Attada, R., H. P. Dasari, J. S. Chowdary, Y. Ramesh Kumar, O. Knio, and I. Hoteit, 2019b: Surface air temperature variability over the Arabian Peninsula and its links to circulation patterns. Int. J. Climatol., 39, 445–464, https://doi.org/10.1002/joc.5821.
Babu, C. A., A. A. Samah, and H. Varikoden, 2011: Rainfall climatology over Middle East Region and its variability. Int. J. Water Resour. Arid Environ., 1, 180–192.
Babu, C. A., P. R. Jayakrishnan, and H. Varikoden, 2016: Characteristics of precipitation pattern in the Arabian Peninsula and its variability associated with ENSO. Arabian J. Geosci., 9, 186, https://doi.org/10.1007/s12517-015-2265-x.
Barlow, M., B. Zaitchik, S. Paz, E. Black, J. Evans, and A. Hoell, 2016: A review of drought in the Middle East and southwest Asia. J. Climate, 29, 8547–8574, https://doi.org/10.1175/JCLI-D-13-00692.1.
Bennett, L. J., and Coauthors, 2011: Initiation of convection over the Black Forest mountains during COPS IOP15a. Quart. J. Roy. Meteor. Soc., 137, 176–189, https://doi.org/10.1002/qj.760.
Betts, A. K., and M. J. Miller, 1986: A new convective adjustment scheme. Part II: Single column tests using GATE wave, BOMEX, ATEX and arctic air-mass data sets. Quart. J. Roy. Meteor. Soc., 112, 693–709, https://doi.org/10.1002/qj.49711247308.
Bhomia, S., P. Kumar, and C. M. Kishtawal, 2019: Evaluation of the weather research and forecasting model forecasts for Indian summer monsoon rainfall of 2014 using ground based observations. Asia-Pac. J. Atmos. Sci., 55, 617–628, https://doi.org/10.1007/s13143-019-00107-y.
Chakraborty, A., S. K. Behera, M. Mujumdar, R. Ohba, and T. Yamagata, 2006: Diagnosis of tropospheric moisture over Saudi Arabia and influences of IOD and ENSO. Mon. Wea. Rev., 134, 598–617, https://doi.org/10.1175/MWR3085.1.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, https://doi.org/10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Crétat, J., B. Pohl, Y. Richard, and P. Drobinski, 2012: Uncertainties in simulating regional climate of Southern Africa: Sensitivity to physical parameterizations using WRF. Climate Dyn., 38, 613–634, https://doi.org/10.1007/s00382-011-1055-8.
Dai, A., and K. E. Trenberth, 2004: The diurnal cycle and its depiction in the Community Climate System Model. J. Climate, 17, 930–951, https://doi.org/10.1175/1520-0442(2004)017<0930:TDCAID>2.0.CO;2.
Dasari, H. P., S. Langodan, Y. Viswanadhapalli, B. R. Vadlamudi, V. P. Papadopoulos, and I. Hoteit, 2018: ENSO influence on the interannual variability of the Red Sea convergence zone and associated rainfall. Int. J. Climatol., 38, 761–775, https://doi.org/10.1002/joc.5208.
Dasari, H. P., D. Srinivas, S. Langodan, R. Attada, R. K. Kunchala, V. Yesubabu, K. Omar, and I. Hoteit, 2019: High-resolution assessment of solar energy resources over the Arabian Peninsula. Appl. Energy, 248, 354–371, https://doi.org/10.1016/j.apenergy.2019.04.105.
De Vries, A. J., E. Tyrlis, D. Edry, S. O. Krichak, B. Steil, and J. Lelieveld, 2013: Extreme precipitation events in the Middle East: Dynamics of the Active Red Sea Trough. J. Geophys. Res. Atmos., 118, 7087–7108, https://doi.org/10.1002/jgrd.50569.
De Vries, A. J., S. B. Feldstein, M. Riemer, E. Tyrlis, M. Sprenger, M. Baumgart, M. Fnais, and J. Lelieveld, 2016: Dynamics of tropical–extratropical interactions and extreme precipitation events in Saudi Arabia in autumn, winter and spring. Quart. J. Roy. Meteor. Soc., 142, 1862–1880, https://doi.org/10.1002/qj.2781.
Diaz, J. P., A. González, F. J. Expósito, J. C. Pérez, J. Fernández, M. García-Díez, and D. Taima, 2015: WRF multi-physics simulation of clouds in the African region. Quart. J. Roy. Meteor. Soc., 141, 2737–2749, https://doi.org/10.1002/qj.2560.
Ehsan, M. A., M. Almazroui, A. Yousef, O. B. Enda, M. K. Tippett, F. Kucharski, and A. K. Alkhalaf, 2017: Sensitivity of AGCM-simulated regional JJAS precipitation to different convective parameterization schemes. Int. J. Climatol., 37, 4594–4609, https://doi.org/10.1002/joc.5108.
Evans, J. P., R. B. Smith, and R. J. Oglesby, 2004: Middle East climate simulation and dominant precipitation processes. Int. J. Climatol., 24, 1671–1694, https://doi.org/10.1002/joc.1084.
Evans, J. P., M. Ekström, and F. Ji, 2012: Evaluating the performance of a WRF physics ensemble over South-East Australia. Climate Dyn., 39, 1241–1258, https://doi.org/10.1007/s00382-011-1244-5.
Flaounas, E., V. Kotroni, K. Lagouvardos, and I. Flaounas, 2014: CycloTRACK (v1.0)—Tracking winter extratropical cyclones based on relative vorticity: Sensitivity to data filtering and other relevant parameters. Geosci. Model Dev., 7, 1841–1853, https://doi.org/10.5194/gmd-7-1841-2014.
Fritsch, J. M., and C. F. Chappell, 1980: Numerical prediction of convectively driven mesoscale pressure systems. Part I: Convective parameterization. J. Atmos. Sci., 37, 1722–1733, https://doi.org/10.1175/1520-0469(1980)037<1722:NPOCDM>2.0.CO;2.
Gao, W., C.-H. Sui, J. Fan, Z. Hu, and L. Zhong, 2016: A study of cloud microphysics and precipitation over the Tibetan Plateau by radar observations and cloud-resolving model simulations. J. Geophys. Res. Atmos., 121, 13 735–13 752, https://doi.org/10.1002/2015JD024196.
Gao, Y., R. Leung, C. Zhao, and S. Hagos, 2017: Sensitivity of U.S. summer precipitation to model resolution and convective parameterizations across gray zone resolutions. J. Geophys. Res. Atmos., 122, 2714–2733, https://doi.org/10.1002/2016JD025896.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Giorgi, F., and L. O. Mearns, 1999: Introduction to special section: Regional climate modeling revisited. J. Geophys. Res., 104, 6335–6352, https://doi.org/10.1029/98JD02072.
Grell, G. A., 1993: Prognostic evaluation of assumptions used by cumulus parameterizations. Mon. Wea. Rev., 121, 764–787, https://doi.org/10.1175/1520-0493(1993)121<0764:PEOAUB>2.0.CO;2.
Grell, G. A., and D. Dévényi, 2002: A generalized approach to parameterizing convection combining ensemble and data assimilation techniques. Geophys. Res. Lett., 29, 1693, https://doi.org/10.1029/2002GL015311.
Grell, G. A., and S. R. Freitas, 2014: A scale and aerosol aware stochastic convective parameterization for weather and air quality modeling. Atmos. Chem. Phys., 14, 5233–5250, https://doi.org/10.5194/acp-14-5233-2014.
Han, J.-Y., S.-Y. Hong, K.-S. S. Lim, and J. Han, 2016: Sensitivity of a cumulus parameterization scheme to precipitation production and its impact on a heavy rain event over Korea. Mon. Wea. Rev., 144, 2125–2135, https://doi.org/10.1175/MWR-D-15-0255.1.
Hasanean, H., and M. Almazroui, 2015: Rainfall: Features and variations over Saudi Arabia, a review. Climate, 3, 578–626, https://doi.org/10.3390/cli3030578.
Herwehe, J. A., K. Alapaty, T. L. Spero, and C. G. Nolte, 2014: Increasing the credibility of regional climate simulations by introducing subgrid-scale cloud-radiation interactions. J. Geophys. Res. Atmos., 119, 5317–5330, https://doi.org/10.1002/2014JD021504.
Hoell, A., C. Funk, and M. Barlow, 2015: The forcing of southwestern Asia teleconnections by low-frequency sea surface temperature variability during boreal winter. J. Climate, 28, 1511–1526, https://doi.org/10.1175/JCLI-D-14-00344.1.
Hong, S.-Y., and J.-O. J. Lim, 2006: The WRF Single-Moment 6-Class Microphysics scheme (WSM6). J. Korean Meteor. Soc., 42, 129–151.
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 38–55, https://doi.org/10.1175/JHM560.1.
Huffman, G. J., R. F. Adler, D. T. Bolvin, and E. J. Nelkin, 2010: The TRMM Multi-Satellite Precipitation Analysis (TMPA). Satellite Rainfall Applications for Surface Hydrology, F. Hossain and M. Gebremichael, Eds., Springer-Verlag, 3–22.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, https://doi.org/10.1029/2008JD009944.
Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122, 927–945, https://doi.org/10.1175/1520-0493(1994)122<0927:TSMECM>2.0.CO;2.
Jin, F., A. Kitoh, and P. Alpert, 2011: Climatological relationships among the moisture budget components and rainfall amounts over the Mediterranean based on a super-high-resolution climate model. J. Geophys. Res., 116, D09102, https://doi.org/10.1029/2010JD014021.