Simulation of the Indian summer monsoon regional climate using advanced research WRF model

Authors


Abstract

In this study, the performance of the weather research and forecasting (WRF) ARW regional model was evaluated for simulating the regional scale precipitation during Indian summer monsoon (ISM) at 30 km resolution over seven different homogeneous rainfall zones falling under different climatic (perhumid, humid, dry/moist subhumid, dry/moist semiarid, arid) regions of India. Seasonal scale simulations were made for ten summers (JJAS months) over 2000-2009 using the boundary conditions derived from the National Centers for Environmental Prediction (NCEP) reanalysis data. Sensitivity experiments were conducted with three convection schemes (Kain-Fritsch, KF; Betts-Millor-Janjic, BMJ; Grell-Devenyi, GD). Simulated regional climate was evaluated by comparison of precipitation with 0.5° India Meteorological Department (IMD) gridded rainfall data over land, Tropical Rainfall Measuring Mission (TRMM) rainfall data over the ocean and atmospheric circulation fields with 1° NCEP global final analysis (FNL). Although all the simulations showed spatio-temporal rainfall patterns, BMJ had least bias towards dryness whereas KF had moist bias and GD had higher dry bias. BMJ could simulate low, moderate and high rainfall reasonably well with relatively higher correlations and threat scores, lower bias and mean absolute errors in most zones as compared to better simulation of heavy precipitation events with KF and low rainfall days alone with GD scheme. The better performance of BMJ scheme is evident owing to better simulation of surface pressure, temperature, and geopotential, lower and upper atmospheric flow fields. Simulations revealed a relatively less intensive heat, weaker low-level westerly winds, weaker north-south geopotential gradients, weaker subtropical easterlies in the El Niño years than in the La Niña years, which indicate the model is able to simulate the interannual variations in monsoon characteristics. Copyright © 2012 Royal Meteorological Society

1. Introduction

The Indian summer monsoon (ISM) is a major feature of atmospheric general circulation that prevails significantly over the eastern hemisphere (Rao, 1976). About 70% of rainfall over India is contributed by the ISM during June to September (JJAS). Discrepancy in monsoon seasonal rainfall affects the agriculture and economy of India. Precise forecasting of the monsoon seasonal rainfall is considered highly essential in view of its role in the sustenance of the agricultural systems and the economy of the country. Several global, regional circulations such as the El Niño-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD) and parameters such as sea-surface temperature (SST), Eurasian Ice covers etc. are shown to influence the Indian monsoon and cause interannual variations (Goswami, 1994; Bhaskaran et al., 1996; Annamalai et al., 1999; Kelkar, 2009) in onset, nature and distribution of rainfall over different parts of India. Forecasting of monsoon on a seasonal scale is very essential for agricultural planning and guidance in different climatic zones of India, and requires application of skillful regional models. The present global warming and future climate change scenario emphasizes this need.

Statistical and dynamical methods are commonly used in seasonal and long-term predictions of summer monsoon. The statistical methods based on empirical–statistical relationships of ISM rainfall with global atmosphere–ocean parameters (Walker, 1914, 1923; Thapliyal and Kulshrestha, 1982; Gowariker et al., 1991; Rajeevan et al., 2000; Fasullo and Webster, 2003) are qualitative and are of limited use at the subregional scale. Dynamical methods with atmospheric general circulation models (GCMs) are used to simulate the large-scale circulation of the monsoon system and seasonal rainfall quantitatively (e.g. Lau et al., 2009; Chen et al., 2010). However, the GCMs as a result of coarse resolution cannot represent the regional characteristics of monsoons (Gadgil and Sajini, 1998; Krishnamurti et al., 2000; Kang et al., 2002; Gadgil et al., 2005; Krishna Kumar et al., 2005; Wang et al., 2005; Satyaban et al., 2010). The features of orography, low pressure trough across India and the off-shore trough on the west coast associated with the monsoon need to be represented precisely to capture the regional circulation patterns and the rainfall. To address this, regional climate models (RCMs) of high spatial/temporal resolution are dynamically coupled to GCMs (e.g. Giorgi and Mearns, 1999; Kato et al., 2001; Nicolini et al., 2002; Wang et al., 2003). The RCMs by virtue of higher horizontal resolution are expected to better simulate the regional dynamics associated with intraseasonal oscillations on different time scales, break spells associated with variations in Intertropical Convergence Zone (ITCZ) and those associated with life cycles of monsoon depressions than the driving GCM (Bhaskaran et al., 1996). Secondly, the RCMs have the advantages of high resolution and improved physics (such as convection) to resolve the regional features associated with topography, local climate and the associated precipitation (Bhaskaran et al., 1996; Das et al., 2006). Several studies on ISM and its seasonal rainfall patterns have been reported using RCMs (Bhaskaran et al., 1996; Ji and Verneker, 1997; Bhaskar Rao et al., 2004; Fu et al., 2005; Das et al., 2006; Cui et al., 2007; Im et al., 2008; Ashfaq et al., 2009; Ratnam et al., 2009; Sato et al., 2009; Dobler and Ahrens, 2010; Polanski et al., 2010). These studies used regional models with resolutions of 50 km and less and reported successful simulation of the intraseasonal oscillations and considerable improvements of ISM rainfall over atmospheric GCMs. Dobler and Ahrens (2010) have studied the representation of the ISM system in a RCM COSMO-CLM driven by global reanalysis (ERA-40) and ECHMA5 global climate model outputs and showed that the application of COSMO-CLM to ECHAM5 improves the temporal correlations of the modeled ISM indices, and the spatial patterns are better simulated in COSMO-CLM with 0.44° horizontal grid spacing than in the large-scale ECHAM5 data. Polanski et al. (2010) used a regional model HIRHAM at 50 km resolution driven with ERA-40 reanalysis fields and showed successful simulation of regional monsoon patterns, and their long-term variability in terms of dry and wet years with observational comparison. Ashfaq et al. (2009) have showed with a nested regional model RegCM3 that enhanced greenhouse forcing results in overall suppression of summer precipitation, a delay in monsoon onset and an increase in the occurrence of monsoon break periods.

With the advent of high performance computing, regional models are preferably used to simulate the climate at moderately high resolutions (e.g. Mukhopadhyay et al., 2010; Taraphdar et al., 2010; Ratnam et al., 2011). In recent studies, Mukhopadhyay et al. (2010) used the WRF ARW model at 15 km resolution to study the ISM rainfall climatology over 2001–2007 and compared with 1° IMD (India meteorological department) rainfall data. They reported that the model monsoon rainfall was sensitive to the convective parameterization. Hariprasad et al. (2011) studied the ISM extreme rainfall climatology with ARW at a grid resolution of 30 km and reported that the model rainfall has a dry bias over the west coast with the Grell–Devenyi convection scheme. Although RCMs are run at high resolutions, their evaluation requires high-resolution rainfall data to convincingly account for the model bias. Secondly, the cumulus physics schemes within RCMs need to be assessed at the scale of their application for their suitability to ISM rainfall. In this context, the recent high resolution (0.5°) rainfall data from IMD (Rajeevan and Bate, 2008) provides an opportunity to realistically assess the high resolution RCMs for ISM rainfall. Several studies on regional climate modeling described earlier (e.g. Cui et al., 2007; Sato et al., 2009; Dobler and Ahrens, 2010; Heikkila et al., 2010; Polanski et al., 2010) considered typical grid sizes of 30–50 km as sufficient to resolve the hydrology-related processes for seasonal scale simulations. The objectives of our study are to examine the ability of the WRF ARW model to simulate the ISM regional climate and the spatio–temporal rainfall distribution over seven climatic zones of India, to evaluate different convection parameterizations at a scale of 30-km resolution for ISM rainfall, and to study the model fidelity to simulate the characteristics of contrasting monsoon (El Niño/La Niña) climates. To achieve the above objectives, the ARW model is run in a controlled condition (i.e. unaffected by a driving GCM bias) with real initial and boundary conditions derived from the National Centers for Environmental Prediction (NCEP) reanalysis fields.

2. Methods

2.1. Experimental details

The Advanced Research Weather Research and Forecasting model (ARW) version 3.1 (Skamarock et al., 2008) developed by NCAR, USA, is used in the present study for seasonal scale ISM simulations. It is a limited area, primitive equation, non-hydrostatic and terrain-following sigma coordinate model and incorporates several parameterizations for different physical processes. The model is configured with a single domain of horizontal resolution of 30 km and 28 vertical levels with its top at 10 hPa. The model domain covers the Indian monsoon region (45°E–109°E and 8°S–40°N) with 230 grid points in the west–east direction and 190 grid points in the north–south directions (Figure 1(a)). The physics included the WSM3 explicit microphysics, Dudhia scheme (Dudhia, 1989) for shortwave radiation processes, Rapid Radiation Transfer Model (RRTM) for longwave radiation processes (Mlawer et al., 1997), the Yonsei University scheme for the Planetary Boundary Layer (PBL) turbulence (Hong et al., 2006; Noh et al., 2003) and five-layer soil thermal diffusion scheme for land surface processes. To study the sensitivity of the monsoon rainfall simulation to the convection parameterizations seasonal scale (JJAS) numerical experiments are conducted with three different schemes viz., Grell–Devenyi ensemble scheme (Grell and Dévényi, 2002), Kain–Fritisch (Kain and Fritsch, 1993; Kain, 2004) and Betts–Miller–Janjic (Betts and Miller, 1986; Janjic, 2000). These experiments are called GD, KF, BMJ, respectively. These three schemes are conceptually different in the representation of vertical fluxes and the calculation of the vertical heating and moistening profiles.

Figure 1.

Model domain with (a) topography and (b) different rainfall zones

The three-dimensional atmospheric fields at the initial time and the time varying boundary conditions are taken from the NCEP global reanalysis fields (Kalnay et al., 1996) available at 2.5° latitude/longitude resolution and at 6-h interval (http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.html). The SST in the model is updated at 6-h interval from the NCEP fields. The model topography (elevation, land cover and soil information) is provided at arc 5 min (∼9 km) resolution from the USGS global data set (Figure 1(b)) as available with the model. The model is integrated continuously for 5 months, starting from 1 May to 30 September for the 10-year period (MJJAS 2000 to MJJAS 2009). The first 20-d period (1 May to 20 May) of each simulation is discarded owing to model spin-up time for the topography and other high resolution fields.

2.2. Analysis

During the decade 2000–2009 there were four weak monsoons (2002, 2004, 2006, 2009) characterized with deficit rainfall associated with El Niño events and two strong monsoons (2000, 2007) characterized with excess rainfall associated with La Niña events. Thus, simulations for this decade enable us to understand the model performance for contrasting monsoon climates and the related interannual variability. The model output is retrieved at 1-day interval corresponding to 00 UTC using which mean monthly fields and seasonal (JJAS) fields are computed. The simulated fields for lower (850 hPa) and upper tropospheric (200 hPa) winds, and geopotential, lower atmospheric (925 hPa) temperature are compared with the NCEP final analysis (FNL) data at a horizontal resolution of 1°. The gridded rainfall data available at 0.5° resolution over the Indian region from the IMD (Rajeevan and Bate, 2008) is used for validation of the model-derived rainfall over the Indian land region and the Tropical Rainfall Measuring Mission (TRMM) daily precipitation (3B42V6; Huffman et al., 2007) over ocean. The model verification includes comparisons of model fields of winds, temperature, geopotential, rainfall with month-wise analysis fields as well as season (JJAS) as a whole for the 10-year composite (2000–2009), El Niño (2002, 2004, 2006, 2009) and La Niña (2000, 2007) composites individually, and time series of daily rainfall for JJAS from different zones and from the entire monsoon region. In order to assess the model monsoon rainfall in time and space, a set of seven zones of relatively homogeneous monsoon rainfall with different climatic types falling in different geographic regions are considered (Table I; Figure 1(b)) (Thornthwaite and Mather, 1955; Das, 1968; Rao, 1976; India Meteorological Department, 1978; Mandal et al., 1999). Six statistical indices are considered to assess model rainfall from the sensitivity experiments. These are the root mean square error (RMSE), mean absolute error (MAE) between the forecast and the observations, the Bias score, which measures the model tendency to systematically overestimate or underestimate a parameter, the correlation coefficient (R), standard deviation (STD) between forecast and observations and the threat score (T) for rainfall which measures the model ability to forecast rainfall, classified into categories (Anthes et al., 1989). The correlations for model rainfall are obtained at 95% confidence level.

Table 1. Details of different zones considered for model evaluation
Name of zoneRegionClimate typeMoisture index (Im)Mean monsoon rainfall (mm)
Zone 1NorthernDry semiarid− 33.4 to − 83.3700
Zone 2NorthwestArid< − 66.7300–500
Zone 3CentralDry subhumid− 33.3 to 0900
Zone 4EasternMoist subhumid0–201500–2000
Zone 5NortheastHumid+ 20 to + 1002000
Zone 6WestPerhumid> + 1002000–3000
Zone 7SoutheastMoist semiarid− 49.9 to − 33.4700

3. Results and discussion

3.1. Spatial and temporal rainfall pattern

During the ISM, the rainfall distribution is inhomogeneous over different parts of India owing to the differences in the topography, latitude, land cover and the associated physical processes in various zones. On the west coast, foot hill Himalayan regions in northern India and the northeastern parts of the country, orography plays a major role in forced convection and high monsoon precipitation. The formation of off-shore vortices during monsoon also causes heavy precipitation along west coast. These systems have horizontal extent of few tens of kilometers and time span of 1–2 days which require a model with sufficiently high resolution to predict their life cycle and associated rainfall. The transient low-pressure systems that develop in the Bay of Bengal along the monsoon trough produce precipitation in the central, eastern and northwestern parts by dynamical forcing. Thus a realistic simulation of the monsoon rainfall and its distribution over India requires all the above physical processes to be realistically represented in the model which depends largely on a proper choice of convection schemes.

Results for rainfall with the three convection schemes are presented in Figure 2. Simulated mean rainfall from all experiments indicated establishment of monsoon (rainfall > 2 mm d−1) over a major part of India, coverage of monsoon (rainfall 4–18 mm d−1) over most parts in July, and complete establishment of monsoon in August over India (not shown). These features agree reasonably with the IMD rainfall and TRMM rainfall patterns. However, the aerial distributions depict large differences in the three experiments. Simulated seasonal (JJAS) mean precipitation climatology shows that the GD scheme tends to produce considerably low rainfall in the eastern, northeastern, western parts and along the west coast (Figure 2(e)) relative to the IMD observed rainfall. The mean daily precipitation is underestimated by about 10 mm d−1 along west coast, northeast and by 5 mm d−1 in the central parts. Overall, GD produces a dry bias in rainfall along the west coast, northeast and central parts. The simulation with KF (Figure 2(c)) has produced a higher precipitation over the Bay of Bengal, Arabian Sea and adjoining coastal land portions of the country as noted from comparison with TRMM, IMD data sets. It considerably overestimated the rainfall in the west coast, southern peninsula and central India with simulation of 8–18 mm d−1 against observation of 2–8 mm d−1, thus indicating a moist bias. On the other hand, the simulations with BMJ scheme (Figure 2(d)) produce a rainfall scenario where the precipitation in the west coast, central India and northeast are reasonably well simulated although with a slight underestimation of rainfall along the west coast and northeastern parts. Of the three schemes, BMJ brings out a more representative spatial distribution of rainfall than KF and GD schemes as it could reproduce the two rainfall maxima (along west coast, northeast) and the minima (northwest) in intensity and spatial distribution reasonably well besides depicting the moderate rainfall pattern in the rest of the country. Comparisons with TRMM rainfall data show that all the three schemes produce higher rainfall in the ocean over east Bay of Bengal and in the east Arabian Sea along the west coast than the observations. The model produced less precipitation in El Niño years relative to La Niña years (Figure 3) as seen from higher rainfall differences in El Niño phase (10-year mean ENSO) relative to La Niña, the former marked with relatively lesser rainfall in northern India, central India, eastern and western parts in agreement with the corresponding IMD observed precipitation. The experiment BMJ brings out the differences in the two ENSO phases very clearly and indicates the model's capability to simulate the interannual variation in regional precipitation associated with the global phenomena through dynamical downscaling.

Figure 2.

Seasonal (JJAS) mean precipitation (mm d−1) for 2000–2009 from (a) IMD (b) TRMM (c) ARW KF (d) ARW BMJ. Contours are drawn at 0.2, 1, 2, 4, 8, 12 and 18 mm d−1. This figure is available in colour online at http://wileyonlinelibrary.com/journal/joc

Figure 3.

Seasonal (JJAS) mean precipitation difference (mm d−1). (a) IMD 10 year mean—El Niño years, (b) TRMM 10 year mean—El Niño years, (c) ARW KF 10 year mean—El Niño years, (d) ARW BMJ 10 year mean—El Niño years, (e) ARW GD 10 year mean—El Niño years, (f) IMD 10 year mean—El Niño years, (g) TRMM 10 year mean—La Niña years, (h) ARW KF 10 year mean—La Niña years, (i) ARW BMJ 10 year mean—La Niña years and (j) ARW GD 10 year mean—La Niña years

3.2. Quantitative rainfall comparisons

The seasonal (JJAS) cycle of simulated mean daily rainfall averaged over the Indian land region as well as over different zones for 2000–2009 from 1 June to 30 September from experiments with the three convection schemes along with corresponding IMD rainfall is illustrated in Figure 4. The peak rainfall is found between 8 July and 23 August both in simulations and IMD data, indicating that the model is able to pickup the timing of monsoon rainfall. It is seen that simulated rainfall with KF scheme agrees well with IMD data up to 25 June and thereafter significantly overestimates, indicating high moist bias in the advancement and late phases of monsoon. Simulations with both BMJ and GD schemes show good comparisons of rainfall with IMD data till 12 June. Thereafter, both BMJ and GD underestimate rainfall until 15 August indicating dry bias in the early and advancement phases. This indicates that BMJ and GDE give underestimation and KF overestimation of rainfall during convective monsoon conditions in July and August. While KF has a strong moist bias (3.46 mm d−1), the GD has strong dry bias (–4.7 mm d1) for all India monsoon rainfall (Table II). Further, the KF scheme produces the highest rainfall errors (MAE = 3.69 mm d−1, RMSE = 4.37 mm d−1) over the GD (MAE = 2.62 mm d−1, RMSE = − 2.89 mm d−1) and BMJ (MAE = 1.81 mm d−1, RMSE = 2.16 mm d−1). Threat scores calculated for simulated seasonal rainfall under four categories i.e. low (>5 mm d−1), moderate (5–10 mm d−1), high (10–20 mm d−1) and very high (>20 mm d−1) (Table III) indicate that both GD and KF schemes produce low threat scores for low and moderate rainfall categories with GD giving slightly better scores than KF. BMJ is noted to produce a threat score of 0.50 for low rainfall and 0.76 for moderate rainfall indicating appreciable performance for low and moderate rainfall ranges which are the dominant monsoon precipitation categories as per IMD data (Table III). Although KF is able to simulate high precipitation category well, it exaggerates the range by 10–20 mm d−1 (Table III). Thus, considering both the seasonal cycle and quantitative rainfall, BMJ seems to produce more realistic monsoon rainfall simulations over India than KF and GD schemes.

Figure 4.

Time series of daily rainfall (mm d−1) over India and its different climatic zones from experiments with different convection schemes. This figure is available in colour online at http://wileyonlinelibrary.com/journal/joc

Table 2. Statistical skill scores (Bias, MAE, RMSE, SD, Correlation) for model mean daily rainfall (mm d−1) during the monsoon season (JJAS) over India and its different climatic zones for 2000–2009 for simulations with different convection schemes
  Statistical index
ExperimentZonesBias (mm d−1)MAE (mm d−1)RMSE (mm d−1)STD (mm d−1)R (mm d−1)
GDZone 1− 0.392.132.72.410.38
 Zone 2− 3.33.785.293.650.35
 Zone 3− 1.72.533.443.220.59
 Zone 4− 3.013.544.663.250.24
 Zone 5− 5.295.35.893.630.46
 Zone 6− 6.796.838.325.160.45
 Zone 70.172.2332.450.28
 INDIA− 4.702.622.892.090.79
KFZone 11.242.593.553.190.47
 Zone 224.626.525.880.47
 Zone 34.695.497.235.90.57
 Zone 43.945.336.935.130.32
 Zone 51.994.095.544.160.21
 Zone 615.716.0518.611.120.24
 Zone 75.65.96.84.340.3
 INDIA3.463.694.373.420.69
BMJZone 11.332.352.832.420.41
 Zone 2− 1.813.134.73.40.21
 Zone 3− 1.252.623.563.030.44
 Zone 4− 2.083.194.313.250.25
 Zone 5− 4.594.675.313.340.41
 Zone 6− 1.13.774.944.570.43
 Zone 7− 1.452.873.282.360.22
 INDIA− 1.711.812.161.860.80
Table 3. Threat scores for model rainfall over different categories in different zones for experiments with different convection schemes
  No. of observations/model valuesNo. of hitsThreat score
ZonesRainfall (mm d−1)IMDBMJGDKFBMJGDKFBMJGDKF
INDIA> 51629678151660.500.2390.333
 5–10959254368146250.7640.4470.236
 10–20100077008000.101
 > 200000000000
Zone 1> 5684529452551250.280.630.28
 5–10486790603424320.420.370.42
 10–20481151000.0900
 > 200000000000
Zone 2> 55267104422553250.260.510.36
 5–1047531637252250.330.030.35
 10–202000330012000.29
 > 201008000000
Zone 3> 5342937222024120.4650.510.27
 5–10558676274841120.510.450.17
 10–20315755012000.0270.30
 > 2000016000000
Zone 4> 512273332200.050.040
 5–1064767513312980.280.260.09
 10–204417127584350.150.070.4
 > 2000012000000
Zone 5> 5019341000000.
 5–10368473401414180.130.140.31
 10–208417137052520.050.020.51
 > 200009000000
Zone 6> 502310000000
 5–1043398221212100.4140.200
 10–20637971442020.420.00.027
 > 20140085001200.00.138
Zone 7> 551743712261700.260.2390
 5–10574377311952160.2350.6340.22
 10–20113675009000.117
 > 201002000000

The time series indicates that GD produces an underestimation of rainfall and fails to capture many of the observed peaks in Zones 2, 4, 5 and 6 which receive relatively high monsoon rainfall (>1000 mm). The GD underestimates the monsoon season rainfall by 33–53% in Zones 2, 4, 5 and 6 and reasonably simulated the low rainfall (300–900 mm) categories in Zones 1, 3 and 7. The time series of simulated rainfall with KF clearly shows a significant overestimation throughout the monsoon season in all the zones. It overestimated the rainfall amounts in different zones by 17% (in Zone 5) to 123% (in Zone 6) giving considerable moist bias for monsoon precipitation. The BMJ gives an intermediate rainfall trend between GD and KF and depicts well the peaks and troughs in most zones with modest underestimation. The BMJ has overestimated the rainfall by 28% in Zone 1 and underestimated the rainfall by about 9–40% in Zones 2–7. The daily rainfall time series from simulation with BMJ closely follows the IMD rainfall in Zones 1, 3, 4 and 6. An insight of this simulation shows that it could reproduce most of the peak rainfall events in all zones except Zones 2, 5 and 7 where it missed some of the high rainfall episodes in July and August. However, the time series with BMJ indicates lower moist or dry bias relative to KF and GD in most zones.

Quantitative comparison of simulated mean daily rainfall in different zones with IMD rainfall (Table II) shows that KF gives high moist bias, MAE and RMSE indicating overestimation of rainfall in all zones. Especially, it produces high moist bias ( ≥ 4 mm d−1), high MAE (>5 mm d−1) and RMSE (>6 mm d−1) in high rainfall Zones 3, 4, 6 and 7. The KF scheme has given relatively higher threat scores than GD, BMJ in Zones 4 and 5, especially for moderate and high rainfall categories. The GD scheme is noted to give dry bias, moderate to large MAE and RMSE in most zones. It has produced high dry bias (<− 3.0 mm d−1), moderate to large MAE (2–6 mm d−1) and RMSE (3–8 mm d−1) in Zones 2, 4, 5 and 6 suggesting significant underprediction of rainfall in these high rainfall zones. However, the GD has lower bias and lower MAE and RMSE in the low to moderate rainfall Zones 1, 3 and 7 indicating it could simulate the low rainfall reasonably well. The GD scheme also has given relatively higher threat scores in Zones 1, 3 and 7 for low, moderate rainfall categories (Table III). Unlike both KF and GD schemes, the BMJ scheme has given moderate bias (1 to–4 mm d−1), moderate MAE (2–4 mm d−1) and moderate RMSE (2–5 mm d−1) in all zones suggesting that it could simulate low as well as high rainfall. This scheme showed very low error statistics for most zones except the high rainfall Zone 5. The threat scores obtained from BMJ are comparable with those from GD for low rainfall in Zones 3 and 7; relatively higher than those from GD and KF for moderate rainfall in Zones 1, 2, 3, 4, 5 and 6 and higher than that from KF for high rainfall in Zone 6. Observed rainfall climatology over the period 2000–2009 (Table III) shows that excepting Zones 5 and 6 low and moderate rainfalls are the major rainfall category over India for which BMJ scheme performs better than GD and KF. The high rainfall category (10–20 mm d−1) is noted to occur in the Zones 5 and 6 where both KF and BMJ have shown comparable threat scores. Thus KF scheme performs better for high to very high rainfall categories alone, and BMJ for both low and moderate rainfall categories besides simulating the high rainfall also. Overall, the time series indicates that the BMJ scheme produces better trends in temporal rainfall and its distribution in different climatic zones than KF and GD. KF has shown considerable moist bias while the GD shows significant dry bias for rainfall.

To examine how well the model precipitation for a given duration compares with its long-term precipitation of the same duration from different zones, the standardized precipitation index (SPI) (McKee et al., 1993; Guttman, 1999) is estimated for the 10-year period (2000–2009) over monthly time scale. The SPI classifies the simulated and observed rainfall as a standardized departure with respect to rainfall distribution function for each zone. Comparison of SPI derived from observed and simulated rainfall indicates that the model reproduces contrasting positive and negative SPI trends characterizing wet and dry periods over India as well as its different rainfall zones (Figure 5). In general SPI from simulation is slightly higher positive in 15–25 June, 30 July–10 Aug in Zone 1 (northern India), in 5–15 July, 22–28 Aug in Zone 2 (northwest), 2130 July, 10–20 Aug in Zone 6 (west coast) and slightly higher negative SPI in 15–30 July in Zone 2, in 11–20 July in Zone 3 (central India) relative to the IMD SPI. There are also a few reverse trends in model versus observed SPI for Zones 2, 4 and 6, such as some of the dry periods simulated as wet periods in these zones. These extremities are pronounced more in the SPI trends with KF, GD than with BMJ. The model SPI with KF scheme compares well with IMD SPI in Zones 3, 4 and 5 alone and it produces spurious and excessively wet spells in other zones, which is owing to overprediction of rainfall using KF. The SPI derived from simulation using GD agrees with IMD SPI in Zones 1, 3, 4 and 7 in moderate wet and dry spells alone. It fails to represent many of the observed wet and dry spells in Zones 2, 5 and 6 and also produces unrealistic wet spells in Zones 2 and 6. The SPI with BMJ closely followed the IMD SPI in Zones 2–6 but has produced a few spurious wet and dry spells in Zones 1 and 7. Thus, the SPI time series suggests that by and large simulation with BMJ convection scheme is able to bring out most of the wet and dry periods in different zones.

Figure 5.

Time series of the standardized precipitation index (SPI) over India and its different climatic zones from simulations with different convection schemes along with SPI from IMD rainfall. This figure is available in colour online at http://wileyonlinelibrary.com/journal/joc

3.3. Simulated monsoon flow characteristics

Rest of the analysis are only shown for BMJ scheme over the region 0–36°N, 50–105°E for the JJAS mean patterns of sea level pressure (SLP), 925 hPa temperature, flow field and geopotential at 850 hPa and 200 hPa along with corresponding values from the NCEP final analysis fields available at 1° resolution.

The SLP patterns averaged over monsoon season (JJAS) derived from NCEP and ARW BMJ outputs along with their difference (NCEP–ARW) are shown in Figure 6. Simulated pressure distribution indicates that the model could reproduce the heat low over north India with its location centered in northwest India and Pakistan agreeing with NCEP data but with a slight underprediction (1–2 hPa) of its strength. Except that the SLP is slightly underestimated (1–2 hPa) over Arabian Sea and slightly overestimated (1–2 hPa) in Indo-Gangetic plains, the features of ridge line on the western parts of southern peninsular India and trough line off the west coast which causes heavy precipitation along west coast are well simulated. Much of the rainfall during the summer monsoon is due to the convection generated in the region of monsoon trough and its adjoining low pressure areas. The simulated monsoon low over northwest India is noted to be relatively stronger in July than in June, August and September months in agreement with the reanalysis data. The model predicts slightly weaker pressure gradients than that in NCEP data in the monsoon season over central and northern India. The pressure difference between the 10-year mean and the ENSO years indicates that the heat low is less intensive (Δp = − 0.2 hPa) during the El Niño years than in La Niña years (Δp = 1.0 hPa) which is an interesting feature indicating pressure anomaly over contrasting monsoons. The model produced heat low for La Niña years also showed a wider extent over the northwest, north India, Pakistan and adjoining regions. Thus, ARW could reasonably simulate the monsoon pressure pattern during its different phases (onset, progression and weakening) and the associated features like monsoon low over northwest India, off-shore trough in the west coast during June and the semi-permanent monsoon trough across India.

Figure 6.

Mean sea level pressure (hPa) averaged for JJAS. Top panels (a, b and c) are average for 10 years (2000–2009), middle panels (d and e) represent differences of the 10-year average, El Niño years, (f) difference of NCEP, ARW for El Niño years and bottom panels (g, h and i) give the same as in middle panel for La Niña years

Intense heating of the land region and the lower atmosphere during the summer season causes the development of low pressure over the Indian subcontinent which sets on the monsoon by drawing the southeasterly trade winds to the northern latitudes. Simulated temperature distribution at 925 hPa (Figure 7) indicates slight overestimation of temperature (by 0.5° to − 1 °C) over central India. A warm region extending from northwest India to the Bay of Bengal is seen in agreement with the NCEP reanalysis. Simulated temperature difference between the 10-year mean and the ENSO years indicates a higher heating (Δt = 0.3 °C) during the La Niña years than in El Niño years (Δt = 0 °C), which is slightly an underestimation to the corresponding differences (Δt = 1 °C) from NCEP reanalysis. In all these cases, although the meridional temperature gradient is maintained, the temperatures are slightly underestimated in northern India, central India and the northern Bay of Bengal regions. Overall, the heating is well simulated in southern and central peninsular India but relatively less strong in the simulation than in the analysis over the northwest and northern parts of India.

Figure 7.

Temperature (in °C) at 925 hPa averaged for JJAS. Top panels (a, b and c) are average for 10 years (2000–2009), middle panels (d and e) represent differences of the 10-year average, El Niño years, (f) difference of NCEP, ARW for El Niño years and bottom panels (g, h and i) give the same as in middle panel for La Niña years

Simulated flow field and geopotential (gpm) at 850 hPa (Figure 8) shows that the stream lines follow the geopotential distribution and are in good agreement with NCEP data. The incidence of southwesterly flow (15–20 m s−1) over the west Arabian Sea is simulated in June in agreement with the NCEP field. Unlike the analysis, model flow along the west coast has a cyclonic wind shear in the lower atmosphere which is due to the long chain of mountains called the Western Ghats. The model at 30 km resolution could resolve the topography effect to simulate the cyclonic wind shear along the west coast. The simulation reproduces strong southwesterly currents (Somali Jet) of 15 m s−1 extending up to 65°E and the progression of strong winds of 10 m s−1 up to the Andamans very realistically as in the analysis. Simulated mean monsoonal (JJAS) wind flow is slightly stronger in southern latitudes (5–15°N) and slightly weaker in northern latitudes (20–27°N). The wind differences in ENSO phases from 10-year mean indicates relatively weak westerlies (ΔU = 0.5 m s−1) in El Niño years and strong easterlies in La Niña years (ΔU = 0.5–1.5 m s−1) in agreement with NCEP data. The southerly flow over middle-east Asia is also well simulated by the model. These results show that the monsoon low-level flow is well captured by the model although with slight underestimation of strength of winds over the west Arabian Sea and over central India. The model geopotential is low over northwest India in agreement with the analysis but with relatively lower intensity and lower north–south gradients. Both simulation and analysis show that the monsoon trough is characterized with lowest geopotential extending from northwest India to the head Bay of Bengal. The model geopotential is slightly higher in the El Niño years with relatively less north–south gradient than in the La Niña years as noted in the reanalysis. The region of monsoon trough is also found to be associated with relatively lower geopotential gradient in El Niño years thereby indicating a weaker monsoon in El Niño years than in the La Niña years.

Figure 8.

Flow field (in m s−1) and geopotential (m) at 850 hPa averaged for JJAS. Top panels (a and b) show NCEP FNL, ARW fields, middle panels (c and d) for differences of the 10-year mean and El Niño years from NCEP field and ARW field, the bottom panels (e and f) give the same as in bottom panels

Simulated flow field at 200 hPa (Figure 9) indicates the upper level easterly winds, and the Tibetan anticyclones are captured well by the model with realistic anticyclonic wind pattern above the low-level monsoon trough although with slight underestimation of easterlies west of 85°E. The upper air wind differences in ENSO phases from 10-year mean indicate weak easterlies (ΔU = 0.5–2 m s−1) in El Niño in the lower latitudes and strong westerlies at higher latitudes. During the La Niña years the difference winds indicate weak westerlies at higher latitudes and strong easterlies in the lower latitudes as in NCEP analysis (Figure 9(c)–(f)). A reversed horizontal (north–south) gradient in geopotential at 200 hPa relative to that at 850 hPa is simulated by the model in agreement with the NCEP analysis. The highest geopotential is distributed between 23° and 30°N latitudes extending zonally eastwards concentrating in the northeastern parts over the Tibetan high, indicating good representation of subtropical ridge in the model. Also unlike the analysis field, the model geopotential over the land region has complex pattern signifying the regional topography. Relatively stronger north–south geopotential gradient is simulated in La Niña years (Δgpm = 10 m) relative to the El Niño years signifying stronger monsoon in La Niña years.

Figure 9.

Flow field and geopotential (m) at 200 hPa averaged for JJAS. Top panels (a and b) show NCEP FNL, ARW fields, middle panels (c and d) for difference of 10 year mean and El Niño years from NCEP field and ARW field, the bottom panels (e and f) give the same as in bottom panels

The vertical section of zonal winds averaged over the monsoon region (50°E–100°E) for the season JJA (June, July and August) derived from the model and the NCEP analysis for the El Niño and La Niña years are presented in Figure 10 from the simulations using BMJ convection scheme. The NCEP analysis for both the cases indicates the presence of low-level westerly winds below 600 hPa up to 28°N, upper level tropical easterly jet stream above 500 hPa (shaded negative values) up to 25°N, upper level subtropical westerlies from 25°N associated with the Tibetan anticyclone and low-level feeble easterlies north of 32°N. These features are very well simulated by the ARW model. The upper-level easterly jet stream and the upper-level subtropical westerlies north of 28°N are stronger by 2–5 m s−1 in the simulation. This also indicates that the model produces a relatively stronger Tibetan anticyclone. The intensity of the simulated low-level monsoon flow over southern peninsular India and adjoining seas is higher in the simulation than that in NCEP analysis for both the cases. The model simulates a higher downward penetration of the tropical easterly jet around 20°N, as well as a larger zonal extent between 800 and 500 hPa levels as compared to the NCEP field for both the cases. Thus the stronger low-level westerlies and the additional downward penetration of the subtropical easterlies, as noted in the simulation suggest a deeper monsoon trough formation in the ARW simulation than that found in the analysis. The zonal averaged low-level westerlies over the monsoon region in the simulation are stronger during the La Niña years. The model simulates stronger subtropical easterlies in the La Niña years as compared to El Niño years in agreement with the NCEP analysis. The simulation shows an enhanced downward penetration of the easterlies to the south of the monsoon trough for the La Niña years, which indicates an intensification of the trough relative to El Niño years. These results clearly show that the ARW is able to capture the interannual variation in the zonal winds and hence the monsoon circulation characteristics between El Niño and La Niña years. The simulation brings out the features of stronger upper level easterly winds, stronger low-level westerlies over the monsoon region and an intense monsoon trough during La Niña years.

Figure 10.

Vertical cross-section of zonal wind averaged between 50 and 100 E in El Niño and La Niña years. Left panel shows NCEP FNL analysis and right panel shows the ARW field with BMJ scheme. This figure is available in colour online at http://wileyonlinelibrary.com/journal/joc

The sensitivity experiments indicate that BMJ produces better results for monsoon rainfall as compared to the GD and KF, which may be attributed to differences in heating and moistening rates of the atmosphere leading to different rainfall scenarios. It is also noted that pressures are underestimated with KF by 3–5 hPa and overestimated by 2–3 hPa with GD (not shown). Similarly, simulated winds are found stronger (∼8 m s−1) with KF and weaker (∼5 m s−1) with GD relative to the analysis over southern latitudes 5–15°N. The geopotential is underestimated by 10–20 m with KF and overestimated by about 10 m with GD. The BMJ scheme simulated a higher vertical ascent at lower levels than GD and KF schemes. The present results of dry bias with GD scheme and moist bias with KF scheme corroborate with the earlier findings from Mukhopadhyay et al. (2010) for simulation of monsoon rainfall climatology with their 15 km resolution. The reasonable performance with BMJ may be attributed to its handling of both shallow and deep convection processes responsible for monsoon rainfall.

4. Conclusions

The ARW regional model with 30 km resolution driven by NCEP/NCAR global reanalysis is used to simulate the summer monsoon regional climate and its associated rainfall over India for the 10-year period 2000–2009 marked with different monsoon scenarios. The model sensitivity for convection parameterization is studied using three cumulus schemes namely Kain–Fritsch, Betts-Millor-Janjic and Grell-Devenyi. Model performance is evaluated for seven different zones representing dry semiarid north, arid northwest, dry subhumid central, moist subhumid eastern, humid northeast, perhumid west coast, and moist semiarid southeastern peninsular India. The salient results are summarized as follows:

  • The model with 30 km resolution simulated regional scale features of monsoon circulation, such as heat low over northwest India, monsoon trough and off-shore west coast trough.
  • All the simulations show interannual variations, with weaker low-level monsoon southwesterlies and weaker monsoon trough during El Niño years as compared to La Niña years.
  • BMJ cumulus convection scheme produced the best simulation of surface pressure, 850 hPa geopotential, lower troposphere temperatures and winds. KF scheme overestimated the temperatures and winds and underestimated the geopotential. GD scheme overestimated surface pressure and underestimated winds.
  • BMJ convection scheme produced best simulation of rainfall as compared to moist bias with KF and dry bias with GD schemes.
  • BMJ scheme simulated the interannual variability of monsoon rainfall, with lesser rainfall during El Niño years and relatively higher rainfall during La Niño years as of observations.
  • BMJ scheme simulated a higher vertical ascent at lower levels than GD and KF schemes.
  • KF scheme consistently overestimated rainfall over the model domain, simulating exaggerated amounts of rainfall over semiarid southeast peninsular India.

The model is able to simulate main features associated with ISM and its variability in ENSO phases, although it exhibits some bias in the distribution of precipitation and temperature, as well as in the overlying atmospheric circulations. The precipitation distribution in the model is strongly effected by the choice of convective schemes and BMJ scheme provided reasonably good rainfall. This study reports the best simulation of all the prominent features of summer monsoon circulation over the Indian subcontinent with BMJ convection scheme at the 30 km resolution. The reproduction of regional scale features, such as off-shore trough, heat low and spatial variations of rainfall indicate that 30 km resolution is sufficient to produce regional scale climate features for long climate integrations.

Acknowledgements

The authors thank the reviewers for their insightful comments and suggestions for the improvement of the manuscript. NCEP Reanalysis data is provided by the NOAA/OAR/ESRL PSD, Boulder, CO, USA, from their website at http://www.esrl.noaa.gov/psd/. The authors acknowledge the NCEP, USA, for use of the NNRP/FNL analysis data and the India Meteorological Department for the gridded rainfall observations used in the study. The TRMM data is obtained from NASA, USA (http://mirador.gsfc.nasa.gov/collections/TRMM_3B42_daily-06.htm).

Ancillary