Temperature and Precipitation Climatology Assessment over South Asia using the Regional Climate Model (RegCM4.3): An Evaluation of the Model Performance

Figure 2a-f shows the seasonal surface air temperature of winter season (DJF) for observations (CRU and UDEL), driving data (ERA40 and ECHAM5) and two RegCM4.3 simulations forced by ERA40 and ECHAM5 for the period of 1971-2000. Temperature biases with respect to CRU and temperature differences for winter season between two observations, two driving data and CRU, ECHAM5 and ERA40 and between two simulations are displayed in Figure 3a-h.

Both observational data sets show the similar spatial pattern for 2 m air temperature over the most part of the domain. However UDEL remained warmer over the NPI as compared to CRU and warm bias up to 4°C is recorded over the northern area of Pakistan. The two boundary forcing data sets also exhibited similar spatial characteristics of surface temperature. ERA40 reanalysis recorded cooler comparative to ECHAM5 GCM over NPI and cold bias up to -4°C observed particularly over CI and WG of the domain. The behavior of RegCM4.3-ERA40 and RegCM4.3-ECHAM5 in simulating the surface temperature is very similar. The biases are considerably reduced after down scaling two global data set with RegCM4.3. In winter season RegCM4.3-ERA40 output with respect to CRU show warm temperature between 2 to 4°C over the Central India. On the other hand RegCM4.3-ECHAM5 exhibit colder conditions over the HKH and NPI regions. The maximum seasonal mean temperature for winter season (DJF) is recorded around 25°C over WG and -20°C minimum over NPI. Winter temperatures for the rest of the regions (SP, BB and CI) remained between -10 to 20°C.

The 2 m air temperature and respective biases between the observed as well as derived and simulated data set are shown in Figure 4 and 5. There are significant difference between the CRU and UDEL data set in the mean temperature (Figure 4a and 4b). CRU is warmer over the central Pakistan and Tibetan Plateau.

Large differences between ERA40 reanalysis and ECHAM5 GCM are seen over some part of the domain. ECHAM5 produced 4 to 8°C warmer climatology in the CI and NPI. A cold bias of -2 to -4°C is also seen in ECHAM5 over Southern part of the domain as compared to CRU (Figure 5d). However this bias is reduced to -1°C after downscaling ECHAM5 with RegCM4.3 and the differences between RegCM4.3-ECHAM5 and CRU are very small (Figure 5g). ECHAM5-forced RegCM4.3 and ERA40-forced RegCM4.3 output of near surface air temperature are similar to that of observations for summer season except over some part of the domain where temperature are overestimated. Cold bias of -2 to -5°C and warm bias of 2 to 5°C between RegCM4.3-ERA40 and CRU observed along the foothills of HKH and Central India and Pakistan regions respectively (Figure 5f). The warm bias is considerably small in case of RegCM4.3-ECHAM5 (Figure 5g). Maximum summer temperature recorded up to 35°C in the Central Indian sub-continent and minimum temperature around 20°C occurred in CIP region.

In general RegCM4.3 had a cold bias in winter over northern part of the domain when downscaled with both ERA40 and ECHAM5 boundary forcing data sets. This cold bias is strong in summer and reached up to -8°C. Warm winter bias of RegCM4.3-ERA40 with respect to CRU also exists in the central part of the domain. On the other hand RegCM4.3 has a stronger warm summer bias over the Central India and Pakistan as compared to the warm winter bias.

These biases seem to come from the driving data set as well as from the Physics of the model. RegCM has some systematic biases over this region as reported by Gao et al. [4,31].

Winter precipitation over South Asia is mostly associated with the eastward propagating westerly’s which are originating from the Mediterranean. In this section the spatial pattern of the winter precipitation (DJF) are examined. In Figure 6a-f mean seasonal precipitation climatology is presented with CRU, UDEL, ERA40, ECHAM5, RegCM4.3 ERA40 and RegCM4.3-ECHAM5 data set. Whereas the comparison of seasonal mean precipitation for winter season of RegCM4.3 which is forced with ECHAM5 and ERA40 with respect to observational data set of CRU, UDEL and deriving global data set for the period of 1971-2000 are given in Figure 7a-h.

Regional climate model RegCM4.3 forced with global data set overestimated the spatial pattern of winter precipitation over the most part of the domain. CRU and UDEL show almost similar values and spatial pattern of DJF seasonal mean precipitation (Figure 6a and 6b).

The difference between CRU and UDEL is only 1 mm/day in the northern Pakistan and India. ERA40 is drier and poorly simulated the winter precipitation over the South Asia domain. A dry winter bias of 2 mm/day between ERA40 and CRU is observed over the Northern Pakistan (Figure 7c). ECHAM5 global model is wetter over the NPI regions and the differences between ECHAM5 and UDEL are very small. The maximum precipitation occurred over the NPI region up to 5 mm/day and minimum winter mean precipitation is observed below 2 mm/day. Wet bias between 2-4 mm/day is seen over the north east India (Figure 7d). Averaged winter precipitation from 1971-2000 for RegCM4.3 simulated with ERA40 and ECHAM5 is above 5 mm/day over northern Pakistan and adjoining areas of south India peninsular. RegCM4.3-ERA40 and RegCM4.3-ECHAM5 generates the maximum intensity of winter precipitation over the mountain areas of HKH and the adjoining areas of WG. The spatial distribution of wet winter bias is similar in both simulations.

Although RegCM RCM reproduce the regional precipitation pattern well but also tend to show the wet bias over some parts of the domain. Figure 7f and 7g depict the winter wet bias up to 4 mm/day over the HKH as simulated by RegCM4.3 with ERA40 and ECHAM5 forcing dataset. The cumulus convective scheme used in the present study is MIT Emannuel [24] which trigged when buoyancy is higher than the cloud base level. Emannuel scheme tend to produce excessive precipitation over land as compare to other schemes (Grell and Kuo). As soon as the MIT scheme activated particularly in the presence of an individual precipitation event it become very difficult to decelerate it [14].

We note that systematic bias depends on the choice of parameterization schemes and different schemes have different performance over different regions. For example Ozturk et al. [32], Dash et al. [6] and Syed et al. [4] used Grell scheme while simulating the major characteristics of South Asia precipitation using RegCM RCM. Their results showed that Grell scheme tends to produce the weak seasonal mean precipitation.

June, July and August is the South Asian summer monsoon season during which surface wind is northeasterly direction over the major parts of Arabian Sea and BB. Figure 8a-f present the summer monsoon precipitation climatology for observations (CRU and UDEL), driving data (ERA40 and ECHAM5) and two RegCM4.3 simulations for the period of 1971-2000. Figure 9a and 9h) show the respective biases in summer monsoon season with respective to CRU, along with the difference between the two boundary forcing data set and between the two simulations with RegCM4.3. The observational data set of CRU exhibit excessive precipitation over the complex terrain along the adjoining areas of Central India, Nepal, Bangladesh and Bhutan. The precipitation intensity decreases as monsoon penetrated into the NPI along the foothills of HKH.

Maximum precipitation intensity as observed from CRU is up to 12 mm/day. The mean precipitation over NPI is below 6 mm/day. There is no considerable differences noted between CRU and UDEL and both observations exhibit similar spatial characteristics of South Asian summer monsoon precipitation. ERA40 is drier as compared to ECHAM5 and again poorly simulated the seasonal mean precipitation over the most part of the domain. ECHAM5 GCM is wetter and overestimated the seasonal mean precipitation as compared to CRU and the maximum precipitation is above 12 mm/day (Figure 8d). A dry bias reached up to -8 mm/day in the case of ERA40 minus CRU (Figure 9c). RegCM4.3 forced with ERA40 and ECHAM5 global data sets reproduced the summer monsoon precipitation reasonably well as compared to respective driving field. RegCM4.3-ERA40 and RegCM4.3-ECHAM5 seasonal mean climatology appeared to be very close to the CRU and UDEL observations. Both simulations captured the maximum precipitation over Nepal, Bhutan and Bangladesh and along the adjoining areas of WG and BB. The mean precipitation reached up to 12 mm/day over Nepal, Bhutan and Bangladesh. The South Asian summer monsoon is not well penetrated into Northern Pakistan and most of NPI regions remain drier. Extensive differences can be seen between the forcing field and simulation (Figure 9c-g). ECHAM5 has a wet bias over most part of the domain and the maximum bias of 6 mm/day seen over Nepal (Figure 9d). Although in general both simulations have same behavior, the amount of precipitation in RegCM4.3-ECHAM5 is more as compared to RegCM4.3-ERA40 particularly over the BB and WG.

RegCM4.3-ERA40 underestimated the mean precipitation over the central India and dry bias up to -8 mm/day can be seen that appeared to come from the driving field (Figure 9f). However this dry bias is less as compared to the ERA40 reanalysis data set. The wet bias of 2 to 6 mm/day observed over Myanmar and Bangladesh in the case of RegCM4.3 simulation with ERA40 reanalysis. RegCM4.3-ECHAM5 is wetter as compared to RegCM4.3-ERA40 and dry bias less than -5 mm/day seen over Central India (Figure 9g and 9h). It is to be noted that the regional biases are smaller in both simulations compared to two driving fields. Previous studies have shown that the simulation of South Asian summer monsoon precipitation by RegCM is sensitive to the physics parameterization used. Kar et al. [33] and Acharya et al. [34] also found that RCM has some systematic biases while simulating the summer monsoon precipitation over this region. As described in section 2.1.3, RegCM4.3 simulated excessive precipitation with the MIT convective scheme. These biases again seem to come from the physics of the model as well as inherited from the driving field. However in both simulation RegCM4.3 reduces the temperature and precipitation biases and improve the simulation result with respect to the driving field. Gao et al. [35], Giorgi and Mearns [2], Brakovic and Gregory [36] and Rauscher et al. [37] also reported improvement in RCM over their respective regions.

The mean annual cycle of RegCM4.3 simulated temperature (°C) and precipitation (mm/day) is calculated in this section over NPI, BB and WG sub region of South Asia. Figure 10 display the area averaged values of annual cycle of temperature over each sub region for CRU, UDEL, ERA40 reanalysis, ECHAM5 GCM and two simulations. Three peaks of temperature are observed from May to July with the maximum in the month of June over the NPI region. ECHAM5 captured the phase of annual cycle slightly higher than the observation from March to June with a good agreement in the month of July but the values are under predicted in January, February and from September to December over NPI. However magnitude of temperature is quite similar to observation over BB except for the overestimation in the months of June and April (Figure 10b). ERA40 reanalysis tends to underestimate the magnitude of annual cycle of temperature throughout the year over NPI.

Temperature peaks of two simulations are very close to the observations over WG and BB particularly in May, June and September. Annual cycle peaks in June, July and August for ERA40, RegCM4.3-ERA40 and RegCM4.3-ECHAM5 reproduced well with slight differences over the NPI region. RegCM4.3-ECHAM5 simulated the annual cycle of temperature more realistically as compared to driving GCM over the tree sub regions of the domain. The driving field and two simulations over predicted the area averaged temperature in the month of January-February and November-December as compared to the two observations over BB (Figure 10b). Over the WG the simulated monthly temperatures are lower than the observation from March to May and higher in all other months but closer to the observation as compared to the driving field (Figure 10c). Therefore regional climate model RegCM4.3 improved the simulation by providing the values close to the observation for area averaged annual cycle of temperature over three sub regions of South Asia.

The simulated annual mean precipitation (mm/day) over the three sub regions NPI, BB and WG are given in Figure 11. Over NPI the two simulations showed maximum precipitation in all months of the year and relatively minimum in July and August compared to the observation (Figure 11a). RegCM4.3 overestimated the winter precipitation and underestimated the summer precipitation over the three sub region as discussed in the previous section. The two driving fields tend to underestimate the annual cycle of precipitation over all selected regions except in June-July over WG and in May and November over BB with ECHAM5 GCM. RegCM4.3-ECHAM5 reproduced the monsoon precipitation well over WG but underestimated in April-October (Figure 11b). RegCM4.3-ERA40 leaned towards observation over WG and shows best agreement in the monsoon season (Figure 11b). The annual cycle of precipitation in summer season is underestimated and not reproduced well over Western Ghats with ERA40 reanalysis. Over BB generally good agreement is found between the observed and simulated data sets (Figure 11c), however CRU and UDEL monthly average precipitation peaks are higher especially in summer season. Large differences between the driving field and two simulations indicate that annual cycle of precipitation could be mostly driven by the internal dynamics and physics of RegCM4.3.