A framework for the evaluation of the South Asian summer monsoon in a regional climate model applied to REMO



Considering the complex spatial and vertical structure of the South Asian summer monsoon (SASM), the validation of a regional climate model (RCM) dealing with only a few surface variables is considered insufficient. Therefore, we have proposed an evaluation framework for the better assessment of the capability of an RCM in capturing the fundamental structure of SASM. This framework has been applied to the regional climate model REMO using ERA40 lateral boundary conditions for the period 1961–2000.

The application of framework yielded satisfactory performance of REMO in capturing the lower, middle, and upper component of the SASM circulation. REMO has higher correlation between different SASM indices as compared to ERA40, showing its ability in capturing the dynamical link between these indices better than ERA40. We have employed different criteria for the assessment of the monsoon onset, and the movement of the Intertropical Convergence Zone (ITCZ) during the boreal summer and REMO has captured these phenomena reasonably well. The model has also shown the association of the meridional temperature gradient with the easterly shear of zonal winds. These results lead us to the conclusion that REMO is well suited for long-term climate change simulations to examine projected future changes in the complex SASM system. Copyright © 2011 Royal Meteorological Society

1. Introduction

With the dependence of a large percentage of world's population on its annual rainfall, the South Asian summer monsoon (SASM) has been the subject of numerous scientific studies since long. For a couple of centuries, scientists have been trying to discover the factors that influence the SASM system. Initial work on SASM by Edmond Halley in 1686, H. F. Blanford in 1877 and Sir Gilbert Walker in 1904 (Mohanty et al.2007; Kripalani et al.2007; Katz RW (2002)) set the tone for further research in the coming centuries. Because of its ‘notoriously difficult to predict nature’ (Jayaraman, 2005), SASM poses challenges to scientific community working over the region even at present times.

Today, coupled atmosphere ocean general circulation models (AOGCMs) are considered to be the most advanced numerical tools to carry out global climate simulations. The general circulation of the atmosphere, which is driven by large-scale climatic forcing, can be effectively simulated by AOGCMs. The typical horizontal resolution of AOGCMs is about 200–300 km, and many regional-scale climatic processes go beyond the scope of AOGCMs. For this reason, the nested regional climate modelling technique was developed to downscale the AOGCM results to the regional scale, and thus these models are called Regional Climate Models (RCMs).

The past decade has witnessed a rapid growth in the development and application of RCMs. By simulating the climate at a limited model domain, the higher-resolution simulations have become computationally less expensive than AOGCMs. RCMs now run on a grid size between 50 and 10 km and on time scales up to 150 years (Jacob et al.2007, Jacob et al.2008). Today, this technique is becoming popular among the developing countries as well (Pal et al.2007). A brief review of regional climate modelling, from its ensuing stages in the late 1980s to the recent past can be found in Giorgi (2006).

The simulation of the South Asian summer monsoon, which extends from Tibet to the southern Indian Ocean, and from East Africa to Malaysia, has been a focus in meteorological research due to its complex mechanism. In the past, there have been attempts to capture monsoon features with RCMs on a shorter time scale of up to a few years (Bhaskaran et al.1996; Ji and Vernekar 1997; Dash et al.2006; Patra et al.2000; Bhaskar Rao et al.2004; Singh et al.2007). More recently, Dobler and Ahrens (2010) conducted a comprehensive study on the analysis of the regional model ‘COSMO-CLM’ forced by reanalysis data and a GCM. Regional models have also been used to study the long-term effects of climate change due to enhanced greenhouse gases over South Asia (Kumar et al.2006, Ashfaq et al.2009). Most of the above mentioned studies are either restricted to very short time periods of a couple of seasons or lacking in the rigorous evaluation of the model when forced with reanalysis data over a longer time scale of 30 or more years which is a typical time scale for climate studies.

RCM of the Max Planck Institute for Meteorology, the REgional MOdel (REMO; Jacob 2001) has been used widely for various studies in many parts of the world (e.g. Europe: Jacob et al., 2007; South America: Silvestri et al., 2009; South Africa: Haensler et al., 2010), however, it has not been extensively tested over the South Asian sub-continent monsoon region. Although, Saeed et al. (2009) (Sa09 hereafter) have recently presented REMO results over the SASM domain with the prime focus on identifying the effects of irrigation on SASM circulations and associated rainfall. Sa09 has evaluated the performance of REMO over SASM region rather briefly and it was restricted to surface variables such as temperature, mean sea level pressure, precipitation and low-level winds. Considering the complexities and deep vertical structure associated with the SASM system, there is a need to evaluate these results in much more detail. Usually, the analyses of RCM applications focus on surface parameters, thereby often neglecting the origin of biases in these parameters. However, if it can be shown that a RCM is able to capture the fundamental structure of SASM, then biases in surface variables can be attributed to deficiencies in model parameterizations or missing processes.

In the present study, an evaluation framework for the representation of SASM circulation and associated rainfall is presented. The performance of REMO in simulating SASM features is rigorously evaluated over a 40-year time scale using the recent REMO version, which we think was missing in earlier studies done by different RCMs over this region. Such a validation is necessary for gaining the confidence in RCM, in this case REMO, to carry out in-depth studies such as climate change and its probable impacts over the region. A brief description of REMO and experiment design is given in Section 2, and in Section 3 the evaluation framework for validation of model is introduced. Section 4 deals with the simulation results and final considerations are presented in Section 5.

2. Model and experimental setup

The regional climate model REMO is based on the Europamodell/Deutschlandmodell system (Majewski and Schrodin 1994). It uses a modified version of the physical parameterization package from the general circulation model ECHAM4 (Roeckner et al.1996). A linear fourth-order horizontal diffusion scheme is applied to momentum, temperature, and humidity. The finite difference equations are solved on an Arakawa-C grid. At the lateral boundaries of the model domain, a relaxation scheme according to Davies (1976) is applied, which adjusts the prognostic variables in a boundary zone of eight grid boxes calculated separately for the different compartments within a model gridbox (Semmler et al.2004). The vegetation-dependent land-surface parameters are taken from the LSP2 dataset (Hagemann S. 2002), and monthly variations are implemented by Rechid and Jacob (2006; leaf area index, vegetation ratio) and Rechid et al. (2007; surface background albedo). In REMO, the improved Arno scheme (Hagemann and Dumenil Gates 2003) is implemented to represent the separation of rainfall and snowmelt into surface runoff and infiltration.

REMO has been applied at a resolution of 0.5° (∼55 km) over the South Asian domain which encompasses 12°S to 42°N and 35°E to 110°E with 109 grid points along the latitudinal and 151 grid points along the longitudinal direction. A continuous simulation is made for the years 1961–2000 with 3 years of spin-up from 1958–1960. The model is forced with so-called perfect lateral boundary conditions at a resolution of 1.125°, obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) 40-year reanalysis (ERA40; Uppala et al.2005).

3. Evaluation framework

As mentioned earlier, Sa09 briefly described an initial evaluation of REMO over SASM, which was limited to only a few surface and near-surface variables. Therefore, readers are advised to see the performance of variables such as 2 m temperature, mean sea level pressure, etc. in Sa09. But these variables may not be the true representatives of the complex SASM system, because some of these variables (such as precipitation) are modelled very late in the process chain of climate models. There are also other processes, such as irrigation, as mentioned in Sa09, which are not represented in these models. Therefore, conclusions based on these results may be misleading unless special consideration is given to the performance of the model in capturing the fundamental structure of SASM system. Here we will focus more on the SASM circulation, its associated indices and other features that are better representatives of the complex spatial and deep vertical structure of the SASM system.

O'Hare (1997) in his explanation of the SASM highlighted the importance of deep vertical structure of SASM circulation. According to him, the differential heating between land and ocean creates a pressure gradient from ocean to land at the surface level even in the pre-monsoon months of April in May. However, rising air over hot land surface creates high pressure aloft in the middle troposphere only up to the height of around 6–7 km, resulting in another pressure gradient from land to ocean. Since the upper tropospheric component remains from ocean to land, the complete monsoon circulation does not become obvious in pre-monsoon months of April and May. In the active monsoon months of June to September, in addition to the lower and middle tropospheric circulation, the high land of Tibet having average height of 4000 m produces strong sensible and latent heat fluxes resulting in the development of a high-level anticyclone. This high-level anticyclone produces upper atmospheric easterlies, which causes an upper tropospheric circulation from land to ocean that was absent in earlier pre-monsoon months. Therefore, in monsoon months, an ocean-to-land component in the lower troposphere and land-to-ocean component in the middle and upper troposphere are observed. The schematic diagram of O'Hare's concept is presented in Figure 1.

Figure 1.

North to south vertical transect of the Indian monsoon circulation concept after O'Hare (1999). The westerly jet (Jw) moves ‘out of the paper’ while the easterly jet stream (JE) moves ‘into the paper’. Two monsoon phases are shown (a) April to May (pre monsoon); (b) June to September (active summer monsoon). The vertical arrow indicates the mean position of the sun

Over the past years, several monsoon indices have been developed in order to understand, measure, and predict the strength of SASM, as well as for the evaluation of climate models. A brief review of these indices can be found in Wang and Fan (1999). Following Parthasarathy et al. (1992) and Goswami et al. (1999), we have defined the Indian Monsoon Rainfall (IMR) and Extended Indian Monsoon Rainfall (EIMR) indices as the departure of rainfall from the climatological mean, averaged over 70°–90°E; 5°–25°N (land points only) and 70°–106°E; 10°–30°N, respectively. Basing on Webster and Yang (1992), an index (WY) has been defined as U*850–U*200, where U*850 and U*200 are zonal wind anomalies at 850 hPa and 200 hPa, respectively, averaged over the region 40°–106°E; 0°–20°N. Similarly, basing on Goswami et al. (1999), we have defined the local Monsoon Hadley (MH) index as V*850–V*200, where V*850 and V*200 are meridional wind anomalies at 850 hPa and 200 hPa respectively, averaged over the region 70°–106°E; 10°–30°N.

The shift of the Intertropical Convergence Zone (ITCZ) or monsoon trough (MT) from the equator to the SASM region is also studied at each grid point on a monthly basis. Two criterions are defined based on precipitation and vertical velocity. For precipitation, the threshold of six mm/day is considered after Gadgil & Sajani (1998) (GSI), which means that if this threshold is attained for a particular grid point in a particular month, we assume the appearance of the ITCZ in the respective grid point and month. We have also considered vertical velocity (VVI) to be a proxy for the ITCZ, since an upward motion is associated with the ITCZ. Therefore, we have defined the criterion that if on all pressure levels (taken to be 925, 850, 775, 700, 600, 500, 400, 300, 200, 100 hPa), a positive (upward) vertical velocity is occurring for a particular grid point in a particular month, we assume the appearance of ITCZ in the respective grid point and month.

The onset of the monsoon has been defined after Wang and LinHo (2002) and He and Sui (2003). Following Wang and LinHo (WLO), the climatological monsoon onset at each grid point is defined as the first pentad (five-day mean) in which the climatological mean precipitation exceeds five mm/day and the grid-point January mean. Following He and Sui (HSO), the onset can be defined according to the following relationship

equation image


equation image

Here, U850 and U200 are mean zonal wind at 850 and 200 hPa, V850 the regional mean meridional wind at 850 hPa and |V850| represents the absolue value of V850. Iu is calculated over the region 40°–110°E; 0°–20°N, whereas V850 is taken over the region 50°–85°E; 5°N–20°N and Hs–Hn is the 850 hPa geopotential height difference between two selected points at (70°E, 5°N) and (50°E, 20°N).

The meridional tropospheric temperature gradient (MTG) has been defined after Webster et al. (1998), as the difference of the five-day climatological mean temperature between the upper tropospheric layers (200 hPa –500 hPa) at 30°N and 5°N, averaged over the zonal belt between 50°E and 85°E. Following Li and Yanai (1996), we calculated the easterly vertical shear of zonal winds (ESZW) as the difference of the five-day climatological mean of zonal winds between 200 hPa and 850 hPa averaged over the region 50°E–90°E; 0°–15°N.

In Sa09, simulated surface variables were mainly compared to observed datasets. Since the present study is focusing more on the vertical structure of SASM, we are validating the model results against reanalysis datasets. Although they have some caveats due to modelling limitations, reanalysis datasets are representing the current climate better than any other model due to their assimilation of observations. In this study, we focused mainly on the validation of model against ERA40 reanalysis, however, for the computation of index correlation we have also considered the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis (Kalnay et al., 1996).

4. Results

In this section the results of our analysis of the different SASM indices are shown and discussed.

4.1. Winds

First the simulated wind flow at 850 hPa and 200 hPa, along with ERA40 data is shown in Figure 2. The regional characteristic features of the lower troposphere such as cyclonic flow pattern over the Bay of Bengal, cross-equatorial flow near the African coast forming the low-level jet stream called Somali jet that is an important mechanism in transferring heat and moisture to the subcontinent (Halpern et al.1999), and westerly flow over northern Arabian Sea and India are captured reasonably well by REMO (Figure 2(a)) compared to ERA40 (Figure 2(b)). Apparently, REMO realistically simulates the general structure of the lower tropospheric circulation, however, some discrepancies are also present. It can be noticed that although the location of the core of the Somali jet is well simulated, its strength is rather underestimated compared to ERA40. In the same way, the wind speeds over the northern Arabian Sea entering into India and Pakistan are overestimated. Sa09 suggested that this behaviour is caused by deficits in the simulation of 2 m temperature and mean sea level pressure due to the absence of an irrigation scheme in REMO. The erroneous intensification of the heat low causes an intense pressure gradient in the south north direction over the Arabian sea, overpowering the coriolis force, and hence, drawing more winds with higher wind speeds into the land areas as westerlies. However, it should also be noted that a significant difference between the reanalysis products exists for this region. For example, Annamalai et al. (1999) found differences in the strength of the Somali jet between ECMWF and NCEP/NCAR reanalysis of up to 5 m/s, as well as differences in the pattern of flow over the Indian peninsula and over the Indian Ocean. Considering the flow field at 200 hPa in Figure 2 (c) and (d), the major discrepancy between REMO and ERA40 is the overestimation of westerly winds in the northern part of the domain. Other than that, the model has ably captured the Tibet anticyclone as well as the strength and the direction of the upper tropospheric easterlies.

Figure 2.

Mean summer (JJAS) average wind (m/s) at 850 hPa. (a) REMO (b) ERA40; and at 200 hPa. (c) REMO (d) ERA40

Figure 3 is used to investigate whether O Hare's concept of three components SASM circulation can be reproduced by REMO. It can be seen that in the pre-monsoon period of April and May, the lower troposphere ocean-to-land, the middle troposphere land-to-ocean, and upper troposphere ocean-to-land component is simulated quite well by REMO (Figure 3(a)) in comparison with ERA40 (Figure 3(b)), and the structure is inline with the description given in Figure 1(a). However, in the active monsoon period of June to September, the lower and middle tropospheric components are slightly overestimated by REMO (Figure 3(c)) in comparison to ERA40 (Figure 3(d)). It should be noted that only the meridional wind component is shown, thus, the strong upper tropospheric component from land to ocean is not obvious here. The reason for this is that winds associated with the Tibetan anticyclone have a strong zonal component, which overpowers the meridional component in the northern part of the domain. However, in the southern part, a noticeable upper tropospheric land-to-ocean component exists in both REMO (Figure 3(c)) and ERA40 (Figure 3(d)), representing easterlies associated with the Tibetan anticyclone.

Figure 3.

North to south vertical transect of the Indian monsoon circulation over the region 60°–105°E, 0°–32°N reflected by the meridional (v) wind component (m/s) for April-May (pre monsoon) (a) REMO (b) ERA40; and for June to September (active summer monsoon) (c) REMO (d) ERA40

4.2. Correlation between precipitation and circulation indices

Goswami et al. (1999) defined a new index for precipitation (EIMR) on a broader region including Bay of Bengal and land masses of Myanmar, Thailand and parts of China, as compared to previously recognized IMR index, which was defined only over the land mass of India. They argued that the large precipitation over the Bay of Bengal having significant interannual variability cannot be ignored in the definition of SASM, and hence, EIMR is a better indicator of convective heating fluctuations associated with the SASM than IMR.

In order to reflect the variation of monsoon heating, Webster and Yang (1992) defined a broad-scale circulation index WY, however Goswami et al. (1999) argued that WY index has little correlation with IMR and EIMR. Therefore, they have also identified a circulation index (MH) describing it as better representative of interannual variability of SASM as compared to the previous WY index. They established this by showing higher temporal correlations between MH index and precipitation indices as compared to WY index. The areas over which the IMR, EIMR, MH, and WY are defined are given in Section 3.

Table I shows temporal correlations among the above-mentioned indices for REMO, ERA40, and NCEP datasets. REMO shows higher correlations among most of the indices than the reanalysis datasets. Reanalysis being an assimilation product may have generated fields that may represent reality better than REMO due to the inclusion of observations, however, higher correlations among indices confirm the ability of REMO in capturing the dynamical link between them. The main conclusion of Goswami et al. (1999) that “… the MH index is significantly correlated with the EIMR”, is very well indicated in Table I for the case of REMO. It can be seen that MH and EIMR has highest correlation for the case of REMO as compared to other indices, which is not found for both reanalysis datasets. Since REMO is forced by ERA40 data on its boundaries, therefore, this behaviour may be regarded as added value of the dynamical downscaling with REMO.

Table I. Cross-correlation among different indices for: (a) REMO; (b) ERA40; (c) NCEP

4.3. ITCZ indices

The monthly shift of the ITCZ, or sometime called as TCZ (Tropical Convergence Zone) is shown in Figure 4. The gradual northward shift of the ITCZ from the equator to northern latitudes during SASM is simulated reasonably well by REMO both for VVI (Figure 4(a) and (b) and GSI (Figure 4(c) and (d).

Figure 4.

Monthly shift of ITCZ from April to August based on precipitation (GSI): (a) REMO (b) ERA40; and based on vertical velocity (VVI): (c) REMO (d) ERA40

Another feature of the SASM, the generation of the second precipitation zone over the equatorial Indian Ocean on the intra-seasonal time scale has also been documented in earlier literature (Sikka and Gadgil, 1980). Following the criterion for ITCZ indices defined in Section 3, REMO has reproduced the development of a secondary ITCZ along the equator, generally one month earlier as compared to ERA40 for both VVI and GSI. This is probably due to the reason that REMO simulates a higher amount of precipitation over the ocean along the equator (not shown) and less precipitation over northwest India and Pakistan as discussed in Sa09.

4.4. Onset indices

The comparison of the simulated WLO index as compared to ERA40 data (Figure 5) indicates that REMO has successfully simulated the onset over the Arabian Sea and western India. However, the model has simulated the onset one pentad too late over eastern and central India, whereas over the foothills of the Himalayas the onset is too early. Again, this behaviour may be attributed to the overestimation of the heat low, as described in Sa09, retarding the movement of westward-propagating cyclones from the Bay of Bengal and bringing them towards the foothills of the Himalayas.

Figure 5.

WLO Monsoon onset after Wang and LinHo (2002) (a) REMO (b) ERA40

Similarly, Figure 6 shows the HSO index, where positive values of the curve represent the active monsoon. Although, the behaviour of the HSO index over the year is simulated very well by REMO, it has simulated the onset and withdrawal of the monsoon one pentad later and earlier, respectively, which means the simulated monsoon period is two pentads shorter than that of ERA40 if the HSO is considered. The results indicate that based on both the spatial (WLO) and temporal (HSO) indices, REMO has performed reasonably to capture the onset of the SASM.

Figure 6.

HSO monsoon onset index defined after He and Sui (2003). The positive values of the curve represent the active monsoon period

4.5. MTG and ESZW indices

Finally, MTG and ESZW are considered in Figure 7. For MTG, the positive (negative) difference indicates that the upper tropospheric air is warmer (cooler) at 30°N than at 5°N, representing the reversal of MTG in the upper troposphere. Li and Yanai (1996) indicated that the positive reversal of the MTG is concurrent with the monsoon onset and produces zonal winds with easterly vertical shear as depicted by ESZW here. It can be seen that the simulated behaviour of MTG matches very well with that of ERA40, with concurrent positive phase from 32nd pentad to 54th pentad. For ESZW at some pentads, the difference between REMO and ERA40 has reached values as high as 4 m/sec. However, for most of the time in the positive phase of MTG, REMO simulated ESZW remains higher than the critical value of 20 m/sec as found by Goswami and Xavier (2005), indicating a satisfactory job done by REMO in simulating this behaviour.

Figure 7.

Easterly vertical shear of zonal winds (ESZW) and monsoon tropospheric gradient (MTG) following Li and Yanai (1996) and Webster et al. (1998), respectively. Horizontal black line represent 0 °C temperature

5. Summary and conclusion

The South Asian region is considered to be one of the most challenging regions for the climate-modelling community due to its geography, orientation, population, and complexities of processes involved in the evolution of its climate. Within the present study, a regional climate simulation has been analysed and validated for the period 1961–2000. The simulation has been performed with the regional climate model REMO at a resolution of 0.5° driven by the ERA40 reanalysis at its lateral boundaries.

Precipitation (and other near surface variables) are modelled very late in the process chain in climate models and influenced by different processes, which are not represented in these models. Especially, the Indian subcontinent, which contains 22% of world's population, has a strong feedback to the atmosphere through processes like irrigation. Therefore, comparison of just a few surface variables against observation is insufficient for the performance evaluation of a climate model especially over the SASM domain. In the present study, in order to gain the confidence in physics and dynamics of REMO, we made an effort to evaluate how well the model is reproducing the basic structure of the monsoon circulation as explained in the theory. We have analysed different physically based indices, which have been used in the earlier literature to study the behaviour of SASM.

The model has successfully captured the patterns as well as the strength of the winds at different pressure levels. The location of the core of Somali jet is well simulated by the model with underestimated strength. On the other hand, the upper troposheric circulations are also reasonably well simulated. Different components of SASM circulation, as defined by O'Hare (1999), of SASM circulation are also captured well by REMO. Seasonal shift of ITCZ, based on two criteria of precipitation and vertical velocity, is also simulated quite well by the model also showing the characteristic secondary ITCZ in the Indian Ocean. REMO shows high correlations between precipitation (EIMR) and circulation (MH) indices as identified by Goswami et al. (1999), adequately representing the dynamical link between them, which was missing for the case of reanalysis datasets showing low correlations between those indices. The onset of monsoon on a cohesive spatial-temporal scale shows a better agreement over the western part of the Indian peninsular, whereas, over the eastern part the onset is somewhat delayed. The model has done a satisfactory job in simulating the reversal of the meridional temperature gradient and easterly vertical wind shear, although the strength of the later is slightly underestimated throughout the year. The behaviour of the curve for He and Sui (2003) onset index is also captured well by the model, however the length of the SASM season is squeezed by two pentads as compared to ERA40.

The application of the evaluation framework in the current study is the step forward to an earlier study done by Sa09, in which REMO was validated against the observations for near-surface variables, and an improvement of the results were shown due to the inclusion of irrigation representation. In this study, we have shown that REMO, even in its standard mode without irrigation representation, has the ability to reproduce the deep vertical structure and complexities involved in the SASM circulations. So far, to our knowledge, there have not been many studies available in which the rigorous evaluation of the regional climate model with reanalysis data over longer time scale is presented for the SASM region. We have proposed a comprehensive evaluation framework for the representation of SASM system based on different indices, which can prove to be very useful in gaining the confidence in the physics and dynamics of a regional climate model. Basing on the promising results of Sa09 and the present study, we conclude that REMO is well suited to conduct a long-term high-resolution climate change projection for the SASM region. In future, this framework shall also be applied to other RCMs over the SASM region within the EU projects WATCH and HIGHNOON.


We would like to thank the reviewers for very constructive and detailed comments. We acknowledge the Higher Education Comission (HEC), Pakistan, for funding this study under the project ‘Overseas Scholarship for PhD in Seclected fields’. This work is partially supported by the European Union FP-6 project WATCH (Contract No. 036946).