The following is a description of the results from RCM simulations for the Indian subcontinent. The model output data are compiled season-wise and spatial distributions of seasonal averages along with corresponding CCSM4 simulations and ERA global reanalysis are presented and discussed. The regional climate simulations in this study are influenced by the global climate simulations as well as the dynamics and physics of the regional climate model. The model validation is confined to the comparison of the basic atmospheric variables only as noted in Sect. 2.
3.1 Season–1 (January-February)
Spatial distributions: The spatial distribution of the 2-m temperatures is presented in the Fig. 2a. During this season, which is the coldest, the continent is cooler than the surrounding oceans. The spatial distributions of RCM agree with those of ERA and CCSM4, with the southern, central and northern parts having temperatures of 295-3000K, 290-2950K and 285-2900K respectively. The simulated temperatures are cooler by of 1–30 over south and northwest parts and slightly warmer over the hilly west-coast region and east-central parts. This feature indicates a slightly higher temperature gradient between the oceans and the continent due to cooler (warmer) bias over the continent (surrounding ocean). It may be noted that the WRF does not predict the ocean surface temperatures but retain those from the climate model. To assess the temperature simulation, the standard deviation values are computed for both the ERA and WRF (not shown). It is noted that the WRF simulation had slightly higher variability over the northwest and central parts of India, whereas it is same over the rest of the subcontinent. The noted higher variability over north-central parts implies a stronger response of the surface processes over the desert soil of north-west India. The sea level pressure distribution is presented in the Fig. 2b. The spatial distributions of the WRF and CCSM4 agree with ERA, except that the WRF simulates slightly higher pressure values over peninsular India. The surface pressure gradients which are generally weak during this season were simulated in agreement with the analysis. But, the slightly higher pressure values denote homogeneity of the continent to the south masking the ocean effect. The spatial pattern of 500 hPa geopotential, simulated by WRF is same as of ERA and CCSM (not shown), but with higher magnitudes by ~ 100 gpm over central and southern parts.
The wind flow and magnitudes of the zonal wind at 850 hPa are presented in Fig. 3a. As high pressures exist over the continent during winter, the Indian land mass exhibits a well defined clockwise circulation around a higher pressure region. The WRF simulates the high pressure with anti-cyclonic circulation over the central latitudes (~ 20 N), agreeing with the analysis. A strong ridge pattern around 200N latitude extending west to east, with well-marked centers over India and North Africa region shows a remarkable agreement between the regional and global model simulations and analysis. However, the WRF model simulated the easterly flow, to the south of the ridge, slightly higher in magnitude than of the CCSM4 but closer to the ERA analysis. The simulation of the zonal flow during this season is significant because of their role within the global scale circulation. The winds at 100 hPa level denote a strong west to east oriented ridge along 10N with westerlies (easterlies) to the north (south) (not shown). The simulated pattern is in good agreement with CCSM4 and ERA analysis. The spatial distribution of the rainfall (Fig. 3b) during winter season by the WRF model is in agreement both of the pattern and magnitudes. As the Indian subcontinent experiences rainfall during this season mostly over the north and east-central regions due to transient western disturbances, the agreement of the spatial distribution validates the model performance in the simulation of the synoptic scale disturbances. The slightly higher precipitation over east-central parts and south peninsula indicate the adopted convection parameterization scheme to show a higher bias.
3.2 Season 2 (March-April-May)
Spatial distributions: The seasonal mean temperatures (Fig. 4a), show that the southern parts are warmer than the northern parts in March; nearly homogeneous in April and northern parts become warmer in May due to the northward progression of the Sun. The mean spatial distribution shows lower (higher) values over south and northwest (east-central) parts, which may be attributed to differential heating associated with the seasonal transition. The variability of the surface temperature is similar in both the simulations and observations, with the differences being limited to 0.5–10. The spatial distribution of the mean sea level pressure shows similar features as of CCSM4 and ERA (Fig. 4b). However, WRF simulated slightly higher pressures over northwest India consistent with the temperatures. The spatial distribution of geopotential at 500 hPa shows that the WRF model simulated values are higher by 100 gpm than ERA (not shown), which indicates slightly warmer troposphere. However, the meridional gradient remains the same both in the simulation and analysis.
The wind flow at 850 hPa denotes two high pressure systems over North Africa and Indo-China region with an embedded ‘Col’ region over southern parts of the Indian Subcontinent in the ERA (Fig. 5a). WRF model and CCSM4 simulated all these three features, which indicate the good performance of both the WRF and CCSM4 models. The winds at 100 hPa denote an east-west oriented ridge along 150N in the global analysis and both the WRF and global CCSM simulated this feature (not shown). However, the meridional width of the ridge in WRF simulation is slightly smaller than the CCSM4 and analysis, which indicate the simulation of stronger westerlies to the north of the ridge system. During this season, Indian subcontinent receives meager rainfall (Fig. 5b) over most of the subcontinent except over northeast India and southwest coast, where the rainfall is due to the local thunderstorms and the occurrence of high rainfall over northern parts of Jammu and Kashmir and Ladakh is due to the passage of western disturbances in the early part of the season. WRF model had simulated the magnitude and the spatial distributions agreeing with the analysis except over the east-central and southeast parts, where the rainfall is underestimated by WRF indicating model’s inability to simulate the local thunderstorm activity.
3.3 Season 3 (June-July-August-September)
Spatial distributions: For the Indian subcontinent, this season is the most important as it yields about 80% of the annual rainfall with large spatial variability. Considering the complexity of the ISM system, any of the simulations for the Indian subcontinent would be considered as good and acceptable only if the model could simulate the essential features of the ISM which are the heat low over northwest, monsoon trough extending NW-SE from about New Delhi to head Bay of Bengal, low-level westerlies over south peninsula, and easterly jet in the upper easterlies at ~ 10N.
WRF model simulated slightly higher temperatures over the central parts of India, moderately higher over north India and slightly lower along the west coast (Fig. 6a). The simulation of temperature over the ocean part being same, the meridional temperature gradient over the Indian subcontinent is more than of the analysis. The model could simulate lower temperatures along the west coast in agreement with the analysis, but the slight underestimation is attributed to the treatment of orography of the Western Ghats in the model. The variability of the temperatures is same as of the analysis except over central India, where it is slightly higher. The mean sea level pressure (Fig. 6b) distribution shows good correspondence between CCSM4 simulation and ERA analysis, whereas WRF simulated stronger pressure gradients over the Indian subcontinent due to lower mean seal level pressures over northwest, which is consistent with the simulated higher temperatures over this region. WRF model simulated 500 hPa geopotential higher by ~ 100 gpm as compared to ERA analysis and 50 gpm than of CCSM4 simulation (not shown). However, the spatial gradients do not differ significantly as the simulated values are higher consistently all over the region. A concentric region of lower magnitudes (~ 20 gpm) over southeast coast is simulated that agrees with ERA and CCSM4.
The simulated 850 hPa wind flow by CCSM4 and WRF model shows very good agreement with ERA analysis (Fig. 7a). WRF model had simulated the monsoon westerlies, monsoon trough oriented northwest to southeast extending up to head Bay of Bengal with anticlockwise circulation around the monsoon trough. The WRF model simulated slightly stronger monsoon westerly flow over the Arabian Sea and the westerlies become stronger meridionally over the Indian subcontinent, implying a stronger cyclonic flow around the monsoon trough region. The simulated axis of the monsoon trough is in good agreement with the CCSM4 and ERA analysis. The model simulated 100 hPa winds by WRF and CCSM4 are in very good agreement with the global analysis in terms of both the magnitude and distribution (not shown). The anticyclonic circulation around 30N associated with the subtropical ridge is well simulated. Another important feature is the simulation of tropical easterly jet over Indian latitudes around 15N latitude during this season, which corresponds very well with ERA and CCSM4 simulation. The spatial distribution of the rainfall (Fig. 7b) showed good agreement between WRF simulation and the IMD gridded data. The model had simulated higher rainfall over the west coast, east-central and northeast regions as of the observations. The good simulation of both the spatial distribution and magnitudes of the rainfall indicate the performance of WRF model.
As mentioned earlier, simulation of the monsoon features is noted to be difficult due to the complex interactions of different scales of the atmospheric motions and the seasonal reversal of winds from northeasterly to southwesterly arising from differential heating due to land-ocean contrast that is typical for this region as Indian subcontinent alone experiences monsoon activity spanning over four months yielding 80% of the annual rainfall. Simulation of the atmospheric circulation features and the rainfall associated with the Indian southwest monsoon are very important for its impact on the management of agriculture and water resources over this region. The WRF regional climate simulation denotes very good reproduction of all the characteristic features of the Indian southwest monsoon, which emphasizes the good performance of WRF model.
Evaluation of regional rainfall: As a part of the model evaluation, statistical metrics (i.e.) mean, standard deviation (SD), root mean square error (RMSE) and correlation coefficient (CC) are computed. Indian subcontinent is divided in to 6 zones based on earlier studies (Niu et al. 2015; Srinivas et al. 2013) which are shown in Fig. 8. Statistical metrics for the 6 identified zones and for “All India” are presented in Table 3. North East (NE) and Western Ghats (WG) zones receive the highest rainfall due to orography; East Peninsula (EP) and North West (NW) zones receive lesser precipitation while Core Monsoon (CM) and West Central (WC) zones reflect average monsoon season rainfall. The values of the mean and standard deviation over different zones are in good agreement, except for the standard deviation over the WG and NE zones. The spatial CCs are higher than 0.65 except over NE zone where the CC is 0.53. The CC (RMSE) is the highest (lowest) over NW, which receives the lowest rainfall (314 mm) as it denotes desert and dry region where the rainfall is scanty. Similarly the CC is high and RMSE is low over EP (512 mm) where the rainfall is lower than all other zones except NW. The CM and WC zones show higher CCs (0.67 and 0.74) with lower RMSEs (248 and 136 mm) where with the rainfall is substantial (927 and 976 mm). The statistical metrics indicate a very good simulation of the monsoon rainfall considering the magnitudes and the variability over different zones. The model underestimated the rainfall over the NE and WG zones by about 200–300 mm where the model variability is about 50 % of the observations. Higher RMSE over NE and WG zones with lower SD values are attributed to the model’s inability to simulate high precipitation events. Higher CC (0.9) over WG denotes the model’s ability to predict the spatial distribution accurately whereas lesser CC (0.53) over NE zone is indicative of model deficiency to simulate precipitation over the hilly regions embedded within the plains of NE India. These statistical metrics for the different zones indicate a very good performance of the WRF model in simulating the rainfall which is an important component of the ISM.
Table 3
Statistical metrics of regional scale evaluation of rainfall of the monsoon season. (Root Mean Square Error (RMSE) and Correlation Coefficient (CC) are between WRF and IMD rainfall)
Zone
|
Mean
|
Standard Deviation
|
RMSE
|
CC
|
IMD
|
WRF
|
IMD
|
WRF
|
North East (NE)
|
1562.7
|
1216.2
|
465.4
|
261.1
|
398.5
|
0.53
|
North West (NW)
|
314.4
|
241.0
|
103.0
|
99.6
|
28.8
|
0.96
|
West Central (WC)
|
976.2
|
923.5
|
189.2
|
186.8
|
136.5
|
0.74
|
East Peninsula (EP)
|
512.4
|
598.7
|
206.1
|
190.3
|
103.4
|
0.87
|
Western Ghats (WG)
|
1051.3
|
819.9
|
970.4
|
425.0
|
618.8
|
0.89
|
Core Monsoon (CM)
|
927.7
|
795.0
|
310.2
|
298.9
|
248
|
0.67
|
All India (AI)
|
837.8
|
751.5
|
549.7
|
581.4
|
470.6
|
0.65
|
Meridional temperature: The onset of the southwest monsoon over India is characterized by the reversal of the meridional temperature gradient in the upper troposphere (Xavier et al. 2007). To assess its simulation, the mean daily upper tropospheric (600 − 200 hPa) temperature gradient between the southern region (15S-5N) and northern region (5N-35N) latitudes averaged between the 40-100E and for 600–200 hPa are analyzed and presented (Fig. 9). The model simulation shows remarkable agreement with the corresponding ERA values indicating good simulation of the temperature distribution over the Indian subcontinent throughout the year showing the reversal of the temperature gradients associated with the onset and withdrawal phases of the ISM.
3.4 Season 4 (October-November-December)
This season is characterized by the onset of the winds from northeast, gradual reduction of temperatures over the subcontinent due to southward transition of Sun and the passage of easterly waves over southern latitudes leading to their strengthening as tropical cyclones under favorable conditions. During this season, the land part loses heat gradually attaining winter circulation by December. As mentioned earlier, the occurrence of the tropical cyclones moving from east to west is maximum during November. Along with the atmosphere attaining more stability, the pressure gradients would slowly reduce leading to lighter winds. Rainfall during this season is mostly confined to south peninsula resulting from the flow of northeasterlies gaining moisture over Bay of Bengal that would provide rainfall to southeast peninsula region.
Spatial distributions: The model simulated spatial distribution of 2-m level temperatures agreeing with ERA analysis, except slight overestimation over the northern parts (Fig. 10a). A closer assessment shows that the WRF model had simulated temperatures lower (higher) over the southern (northern) parts. This may be due to slightly higher temperatures simulated by the model over the Arabian Sea and west Bay of Bengal. The variability of the simulated temperature is higher over Central India by 1–20 as compared to the analysis, which implies that the southern parts that had higher temperatures also had higher variability. The mean sea level pressure distribution shows higher pressure by 2 hPa over the northern parts and decreasing to south which is consistent with the temperature distribution (Fig. 10b). Broadly the spatial pattern shows good agreement with analysis. The simulated 500 hPa geopotential pattern is similar to the analysis but both the WRF and CCSM models simulated higher geopotential values by 50 and 100 gpm respectively (not shown). This implies the lower troposphere to be slightly warmer in the simulations.
The low level wind flow shows two centers of cyclonic circulation over south Bay of Bengal and the Middle East (Fig. 11a). Both the WRF and CCSM4 models had simulated the stream flow and the cyclonic circulations in agreement with the ERA analysis. The models had simulated high pressures over north India consistent with the simulated temperature distribution. The simulation of the cyclonic circulation over south Bay of Bengal is important for the development of transient easterly waves that transform as cyclonic systems. The winds at 100 hPa denote a ridge around 150N, which is well simulated by WRF and CCSM4 models (not shown). However, the strength of the ridge was simulated stronger by the WRF model as compared to CCSM4 and ERA analysis. The stronger simulation by the WRF model than CCSM4 could be due to the internal dynamics of the regional model. The simulated spatial distribution of rainfall shows higher rainfall over the south peninsula and the southeast coast and lower rainfall over central and northwestern parts in agreement with the analysis (Fig. 11b). However, rainfall over the northern most Jammu-Kashmir-Ladakh regions is overestimated, which may be due to the simulation of stronger transient western disturbances. The simulation of higher rainfall over southern parts during this season indicates good performance of the model.
3.5 Tropospheric wind variations: To complement the evaluation of season-wise regional climate simulations, the lower troposphere annual temperature variations along 78E and meridional tropospheric wind flow variations between winter and summer seasons have been analyzed and compared with corresponding ERA. The time-latitude sections of the monthly mean temperatures at 850 hPa level as simulated by the WRF model along with ERA are presented in Fig. 12. The model simulation shows gradual transition of heating from lower latitudes in March to higher latitudes till June, with attainment of highest values during May-August between 25-35N that persist till September, synchronous with Sun movement. Although the spatial distribution agrees well with ERA analysis, the model simulated values are stronger during summer which indicates a warmer bias. As noted earlier, warmer simulation had reflected as higher geopotential of the troposphere. The latitude-height variations of the zonal wind along 78E are plotted for January and July in the (Fig. 13). WRF model had simulated the differences between the winter and summer regimes that agree with ERA analysis. Strong westerlies in the upper troposphere at 25-30N agree with ERA both in magnitude and depth. However, WRF model simulated slightly stronger easterlies at lower latitudes during winter. The simulation of monsoon westerlies in the lower troposphere and stronger easterlies at higher levels of 200 − 100 hPa above agrees with ERA, but, with slightly stronger magnitude and depth of the lower level westerlies. Simulated strong westerlies at upper levels over middle latitudes are consistent with ERA. Good simulation of the contrasting wind flow in the troposphere between the winter and summer seasons affirm the model’s performance.
3.6 Evaluation of surface temperature and rainfall: The above description of the results, indicate that the WRF model had simulated the characteristic features of all the 4 seasons over the Indian subcontinent. It is known that any climate model should primarily simulate the annual cycle of the temperatures. The WRF and CCSM simulated surface temperatures (2-m level) are compared with ERA and are presented in Taylor plots (Fig. 14) for each of the 4 seasons. The results indicate outstanding correspondence between the model and analysis with CCs of 0.99 for CCSM and 0.95 for WRF (0.9 for season 3). The model simulated variability is slightly higher and with RMSE of 0.4–0.5 for CCSM and 1.5–2.0 for WRF. Considering the temperature to be the prime variable that is to be simulated accurately by any model in climate mode integration, the present WRF simulation is considered to be good. This also implies that the physical processes of radiation, surface physics, planetary boundary layer and convection are well represented. WRF model, in the present study, simulated the seasonal transition of the temperatures and their spatial distributions specific to the Indian subcontinent surrounded by oceans over East, West and South; high mountains to the North depicting distinct features that are unique to the study region. The present study provides a unique regional climate data set for the Indian subcontinent at 25-km resolution that spans for 30-years of current climate.
3.7 Sub-regional scale evaluation: Evaluation of RCM simulation is made on sub-regional scale, of the 2-m level temperature and precipitation for 30 subdivisions of the Indian subcontinent by comparison with IMD gridded daily temperature and rainfall data. The segregation of the subdivisions following IMD are shown in Fig. 15.
3.7.1 Winter (J-F): The simulated mean temperatures are cooler over northern parts as compared to south. WRF had simulated the mean better than CCSM4 for 20 subdivisions, with slightly negative (positive) bias over southern (northern) parts (not shown). WRF had simulated the standard deviation better than CCSM4 for 12 subdivisions. The variability of the temperature is overestimated indicating larger inter-annual variability. This seems to be arising because of negative (cooler) bias in the WRF simulation over north India. The winter rainfall yields meager rainfall with less than 0.84 mm/day for any subdivision. Considering the subdivisions with the mean value higher than 0.4 mm/day are to be only 13, WRF model had simulated better than CCSM4 for 9 subdivisions. CCSM4 had underestimated the rainfall consistently for all the subdivisions, whereas WRF model is overestimated at a few of the subdivisions. This implies that the WRF was able to simulate the low rainfall magnitudes much better than the CCSM4. Considering the standard deviation, WRF had simulated better than CCSM4 over 7 subdivisions in terms of percentage error (not shown). This inference implies that the WRF simulation is consistently better than CCSM4 in terms of both mean and standard deviation.
3.7.2 Pre-monsoon (M-A-M): WRF model had simulated the mean temperatures much better than CCSM4 over 27 subdivisions and the standard deviations better over 23 subdivisions (Table 4). Both the CCSM4 and WRF had slightly negative (positive) bias over northern (southern) parts. Simulation of temperature during this season is very important as it is the hottest and the meridional temperature gradients would be reversing during the latter part of May. Better simulation of the mean and standard deviation of this season imply a comparatively better prediction of the sub-regional scale temperature distribution by the WRF model. During this season, the rainfall is meager with 60% of the subdivisions recording mean seasonal rainfall less than 0.8 mm/day (not shown). Considering only the subdivisions with mean > 0.8 mm/day, WRF model had simulated the mean better for 10 and standard deviations better in 11 subdivisions out of 13 than CCSM4. It is also observed that CCSM4 had consistently underestimated the mean and standard deviation.
3.7.3 Monsoon (J-J-A-S): WRF had simulated the mean values better than CCSM4 for 21 subdivisions. Both the CCSM4 and WRF have slightly positive bias and WRF model had higher bias over central parts of India. WRF model had shown the standard deviations larger than CCSM4 over majority of the subdivisions (not shown). However, it is to be noted that the magnitudes of the standard deviation hover between 0.3 and 0.7 for the observations and the standard deviation values simulated by the WRF are higher by only 0.5. Considering the magnitude of the standard deviation < 1 where the mean values range from 25–30, the simulations by both the CCSM4 and WRF are considered as reasonably good
Table 4
Statistical metrics of surface temperature (C) for the Pre-Monsoon season
S.No
|
Name of the subdivision
|
Mean (C)
|
|
Standard Deviation (C)
|
IMD
|
WRF
|
CCSM4
|
IMD
|
WRF
|
CCSM4
|
1
|
Assam & Meghalaya (3)
|
23.74
|
27.08
|
23.13
|
0.66
|
0.94
|
0.75
|
2
|
N.M.M.T (4)
|
22.08
|
23.04
|
20.62
|
0.73
|
0.79
|
0.65
|
3
|
West Bengal & Sikkim (5)
|
29.23
|
30.61
|
27.31
|
0.71
|
0.88
|
0.87
|
4
|
Gangetic West Bengal (6)
|
29.40
|
30.83
|
27.40
|
0.68
|
0.85
|
0.78
|
5
|
Orissa (7)
|
29.93
|
31.74
|
27.28
|
0.61
|
0.74
|
0.94
|
6
|
Jharkhand (8)
|
28.63
|
29.67
|
24.51
|
0.71
|
1.04
|
0.93
|
7
|
Bihar (9)
|
28.10
|
31.37
|
25.57
|
0.75
|
0.93
|
1.02
|
8
|
East U.P(10)
|
26.95
|
29.29
|
23.69
|
0.90
|
0.98
|
1.09
|
9
|
West U.P (11)
|
28.56
|
29.57
|
23.95
|
0.93
|
0.93
|
1.02
|
10
|
Haryana (13)
|
27.11
|
28.28
|
23.42
|
1.09
|
0.90
|
0.86
|
11
|
Punjab (14)
|
25.13
|
27.07
|
21.76
|
1.13
|
1.06
|
0.85
|
12
|
West Rajasthan (17)
|
28.98
|
28.52
|
25.10
|
1.04
|
0.78
|
0.78
|
13
|
East Rajasthan (18)
|
29.08
|
29.61
|
25.19
|
1.04
|
0.80
|
0.80
|
14
|
West M.P (19)
|
29.83
|
29.30
|
25.94
|
0.65
|
0.79
|
1.00
|
15
|
East M.P (20)
|
28.61
|
29.93
|
25.40
|
0.75
|
0.82
|
0.98
|
16
|
Gujarat (21)
|
29.73
|
30.28
|
27.56
|
0.68
|
0.67
|
0.76
|
17
|
Saurastra & Kutch (22)
|
28.50
|
26.62
|
26.86
|
0.61
|
0.42
|
0.58
|
18
|
Konkan & Goa (23)
|
26.04
|
26.91
|
27.73
|
0.37
|
0.36
|
0.50
|
19
|
Madhya Maharashtra (24)
|
29.59
|
30.30
|
26.13
|
0.74
|
0.82
|
1.02
|
20
|
Marathwada (25)
|
30.09
|
29.96
|
27.84
|
0.54
|
0.75
|
0.80
|
21
|
Vidarbha (26)
|
30.81
|
32.79
|
28.47
|
0.72
|
0.76
|
0.87
|
22
|
Chhattisgarh (27)
|
30.27
|
32.30
|
27.96
|
0.66
|
0.72
|
0.98
|
23
|
Coastal A.P (28)
|
31.59
|
31.23
|
29.17
|
0.53
|
0.60
|
0.56
|
24
|
Telangana (29)
|
31.71
|
31.91
|
28.94
|
0.60
|
0.68
|
0.69
|
25
|
Rayalaseema (30)
|
28.57
|
26.49
|
26.51
|
0.53
|
0.73
|
0.58
|
26
|
Tamil Nadu (31)
|
30.84
|
30.01
|
28.24
|
0.47
|
0.68
|
0.60
|
27
|
Coastal Karnataka (32)
|
26.83
|
27.66
|
28.42
|
0.42
|
0.38
|
0.40
|
28
|
North Interior Karnataka (33)
|
27.27
|
26.81
|
26.72
|
0.44
|
0.54
|
0.54
|
29
|
South Interior Karnataka (34)
|
28.61
|
28.11
|
27.45
|
0.44
|
0.50
|
0.62
|
30
|
Kerala (35)
|
26.16
|
28.15
|
28.55
|
0.41
|
0.47
|
0.62
|
The season yields the highest amounts of rainfall over major parts of the Indian subcontinent. It is observed that 29 subdivisions received rainfall higher than 2.5 mm/day and only one subdivision (Tamilnadu) located on the southeast peninsula received the lowest amount of 2.15 mm/day, Konkan & Goa and NMMT (Nagaland, Manipur, Mizoram and Tripura) subdivisions received the highest rainfall of 29.5 and 13.5 mm/day respectively (Table 5). It is observed that both the CCSM and WRF models underestimated (overestimated) over high (low) rainfall regions. Comparatively, WRF model had produced better simulation of the mean rainfall over 19 and the standard deviation better in 17 subdivisions than CCSM4. It is of interest to note that CCSM4 had higher standard deviation for NMMT and lower for Konkan & Goa as compared to lesser standard deviation values for both of these subdivisions by WRF in better agreement with observations. It indicates a better consistency of the bias in the WRF model than CCSM4.
Table 5
Statistical metrics of rainfall (mm) for Southwest Monsoon season
S.No
|
Name of the subdivision
|
Mean
(mm/day)
|
|
Standard Deviation
(mm/day)
|
IMD
|
WRF
|
CCSM4
|
IMD
|
WRF
|
CCSM4
|
1
|
Assam & Meghalaya (3)
|
8.10
|
11.34
|
12.19
|
1.76
|
2.58
|
4.39
|
2
|
N.M.M.T (4)
|
13.51
|
15.63
|
16.32
|
2.94
|
2.43
|
4.28
|
3
|
West Bengal & Sikkim (5)
|
9.03
|
8.26
|
9.00
|
1.87
|
1.72
|
2.39
|
4
|
Gangetic West Bengal (6)
|
9.58
|
8.79
|
9.32
|
1.94
|
2.06
|
2.24
|
5
|
Orissa (7)
|
9.06
|
8.85
|
8.44
|
1.66
|
1.45
|
2.04
|
6
|
Jharkhand (8)
|
8.65
|
8.98
|
8.53
|
1.63
|
1.98
|
2.45
|
7
|
Bihar (9)
|
7.27
|
6.88
|
7.01
|
1.90
|
2.00
|
1.94
|
8
|
East U.P(10)
|
5.85
|
6.58
|
6.62
|
1.83
|
2.29
|
2.09
|
9
|
West U.P (11)
|
4.84
|
3.95
|
5.37
|
1.61
|
2.29
|
1.52
|
10
|
Haryana (13)
|
2.74
|
2.39
|
3.50
|
1.26
|
1.69
|
1.59
|
11
|
Punjab (14)
|
3.57
|
3.37
|
4.11
|
1.52
|
1.45
|
1.71
|
12
|
West Rajasthan (17)
|
2.87
|
2.52
|
3.34
|
1.15
|
1.90
|
1.48
|
13
|
East Rajasthan (18)
|
4.13
|
3.63
|
5.19
|
1.54
|
2.42
|
1.80
|
14
|
West M.P (19)
|
7.06
|
6.59
|
6.67
|
1.69
|
4.02
|
2.44
|
15
|
East M.P (20)
|
8.75
|
9.27
|
7.68
|
1.67
|
1.58
|
2.62
|
16
|
Gujarat (21)
|
6.22
|
6.23
|
5.82
|
2.39
|
2.44
|
2.54
|
17
|
Saurastra & Kutch (22)
|
3.91
|
3.43
|
3.71
|
2.17
|
1.71
|
1.63
|
18
|
Konkan & Goa (23)
|
29.58
|
24.87
|
14.78
|
5.87
|
5.54
|
3.72
|
19
|
Madhya Maharashtra (24)
|
9.02
|
8.56
|
8.55
|
2.17
|
2.13
|
3.03
|
20
|
Marathwada (25)
|
5.02
|
6.06
|
4.79
|
1.28
|
4.50
|
1.54
|
21
|
Vidarbha (26)
|
7.89
|
7.69
|
7.99
|
1.64
|
3.92
|
2.24
|
22
|
Chhattisgarh (27)
|
7.51
|
8.72
|
8.82
|
1.88
|
3.81
|
2.45
|
23
|
Coastal A.P (28)
|
3.77
|
4.50
|
3.77
|
1.30
|
1.98
|
0.95
|
24
|
Telangana (29)
|
5.77
|
6.69
|
6.76
|
1.87
|
1.62
|
1.85
|
25
|
Rayalaseema (30)
|
3.13
|
4.85
|
3.13
|
1.06
|
1.14
|
1.22
|
26
|
Tamil Nadu (31)
|
2.15
|
2.79
|
3.16
|
0.75
|
0.82
|
1.40
|
27
|
Coastal Karnataka (32)
|
27.79
|
19.25
|
16.58
|
5.95
|
4.81
|
3.32
|
28
|
North Interior Karnataka (33)
|
2.65
|
2.81
|
3.63
|
0.77
|
0.93
|
1.01
|
29
|
South Interior Karnataka (34)
|
3.61
|
4.05
|
4.50
|
3.12
|
2.00
|
1.28
|
30
|
Kerala (35)
|
11.78
|
13.71
|
12.73
|
3.62
|
4.28
|
3.47
|
3.7.4 Post-monsoon (O-N-D): WRF model had simulated the mean values better than CCSM4 for 20 subdivisions, specifically better over northern and central parts which are cooler as compared to south (not shown). WRF showed slightly negative bias as compared to positive bias in the CCSM4 model. As the standard deviation of the observations is less than 1°C in magnitude, WRF simulated higher standard deviations over majority of the subdivisions. While CCSM4 had negative bias, WRF model shows positive bias meaning that the inter-annual variability is slightly higher in the WRF simulation. The seasonal rainfall is very low, as 50% of the subdivisions receive less than 1 mm/day and most of the subdivisions with rainfall higher than 1.5 mm/day belong to southern parts of the Indian subcontinent, which receives rainfall in association with the northeast monsoon. Considering the subdivisions with > 0.8 mm/day, WRF had simulated the mean and standard deviation better than CCSM4 over 9 subdivisions out of 19, and the rainfall is overestimated over the subdivisions with lesser rainfall (not shown). WRF had simulated higher values of standard deviations than CCSM4 over the subdivisions with lesser rainfall in accordance with the simulation of mean.
The above description clearly depicts the regional climate simulation by WRF is better than the parent CCSM4 global simulation implying the advantages of dynamical downscaling to derive finer resolution climate data.