There are two main aims of this investigation. The first is to compare the influence of a convective parameterized scheme used in a mesoscale model (MM5) on its predicted precipitation. The region under consideration is the complex topography of the Pakistani south Asian region from the northern Himalayas to the southern Indus valley plains that have not been studied in this context before. This study focussed on the time of the summer monsoon (July and August) of the year 1998 and 2001. The second objective is to compare the influence of different convective schemes on dynamic variables such as ambient temperature, potential temperature, relative humidity and moist static energy. This analysis would subsequently be helpful in the assessment of atmospheric stability conditions and the establishment of dispersion coefficients for dispersion studies in the area. As extent of precipitation plays an important role in scavenging the pollutants from the atmosphere, if the physical and chemical removal of pollutants is to be incorporated in the overall dispersion modelling, it is essential to simulate the trends of precipitation. Once a specific parameterized scheme of MM5 is finalized for the region and time interval, trans-boundary pollution dispersion studies may be carried out by coupling the mesoscale model with a specific Eulerian- or Lagrangian dispersion model such as CHIMERE or FLEXPART.
Mesoscale climate models were originally designed for climatic simulation of some specific convective environments in North America and Europe (Tadross et al., 2006). However, nowadays, these models, with some modifications, are being employed for the atmospheric simulation of various other regions of the world. Different empirical procedures, i.e. parameterization schemes, have been incorporated in these models in order to simulate diverse physical environmental processes. The appropriateness of a selected scheme for a region may be dependent on the climate and topography of the target region. Simulation studies conducted using different mesoscale models revealed the fact that some cumulus convective schemes have not satisfactorily simulated the environmental parameters in some specific regions of the world, including the Asian monsoon regions (Leung et al., 2003). One of the causes of failure of some convective schemes may be the scale interaction that is intensively complex due to the effects of the Tibetan plateau, ocean–continent contrast and air–sea interactions in these regions. It is, therefore, very important that a mesoscale model is validated against a region of interest giving due consideration to selection of the cumulus convective parameterized scheme (CPS), before using it for the purpose of simulating the atmospheric parameters. These convective parameterization schemes containing different closure assumptions help in computing the cumulus-scale convection (Prater and Evans, 2002). There is no perfect methodology to determine the convection, so the choice of a cumulus convective parameterization to be used is often furnished in numerical prediction models. These schemes are dependent on inadequate data having weak theoretical roots (Cohen, 2002) and describe the changes taking place in atmospheric parameters such as moisture and the temperature of the air column with the start of convection and interaction between the process of convection and grid-scale dynamics extracting grid-scale information from the numerical model (Kerkhoven et al., 2006).
The proficiency of simulation for many environmental parameters (for example wind, sea level pressure and temperature) has been improved remarkably due to the introduction of new schemes, but the quantitative precipitation simulation has always been a most challenging task for forecasters due to insufficient knowledge of precipitation processes and cloud microphysics (Mazarakis et al., 2009). Different studies have been conducted using the MM5 mesoscale model (ucar.edu/mesouser/MM5V3) to examine the sensitivity of quantitative precipitation and simulation of some other meteorological parameters to the choice of the convective parameterization schemes for different regions of the world. For example, a comparison study was performed by Wang and Seaman (1997) using the Anthes–Kuo, Betts–Miller, Grell and Kain–Fritsch schemes for the simulation of rainfall events for both cold and warm seasons over the continental USA. The findings of this study were that although there was no consistently outperformance from any convective scheme, the KF scheme provided better results than other schemes did. Another simulation study using the Kain–Fritch, Grell and Betts–Miller convective schemes was conducted by Mazarakis et al. (2009) for the numerical forecasting of quantitative precipitation during the warm period of 2005–2007 over Greece. Results obtained for the 8 km nested grid indicated a trend of overestimation for light to moderate precipitation and underestimation for high precipitation amounts with all three schemes. More consistent behaviour in simulating the quantitative precipitation in this study was showed by the Kain–Fritsch scheme. For evaluating the performance of five convective schemes (Anthes–Kuo, Betts–Miller, Fritsch–Chappell, Kain–Fritsch and Grell), another study was conducted by Kerkhoven et al. (2006) in east China during the summer monsoon season. Simulating the light, moderate and heavy phases of precipitation events, it was found that the Grell scheme was the most robust, performing well at all rainfall intensities and spatial scales, while the Kain–Fritsch scheme showed good performance for moderate rainfall rates. Moreover, improvement in predicted results was seen using the schemes having downdrafts with those schemes that are without the inclusion of downdrafts. Ratnam and Cox (2006) used the Grell and Kain–Fritsch schemes of the MM5 model for the sensitivity study of the simulation of monsoon depressions to the cumulus parameterization schemes. They found that the Grell scheme overestimated the rainfall, while the Kain–Fritsch scheme seemed to be successful in capturing the distribution of rainfall comparable to observations, although the location of maximum rainfall was not exact. The dependency of rainfall on resolution of the model was also observed in the study. Cohen (2002) tried to compare the performance of the Anthes–Kuo, Betts–Miller, Kain–Fritsch and Grell schemes in the MM5 model for idealized sea breeze simulations. The most realistic results for this study were obtained using the Kain–Fritsch scheme. Ferretti et al. (2000) analysed several precipitation events during June 1990 in the Alpine region using different convective schemes and found that the fair estimation of the amount of precipitation was obtained using either the Grell or Kain–Fritsch scheme. On the other hand, the Anthes–Kuo scheme produced a strong overestimation of the precipitation. Colle et al. (2003) tested the convective schemes of MM5 over the northeastern United States for the cold and warm seasons of 1999–2001 and 2000, respectively. During the warm season it was observed that the Kain–Fritsch scheme produced overestimated results over the coastal area and underestimated the precipitation inland over the Appalachians. A representative case study showed that the coastal region received less precipitation with the use of the Betts–Miller and Grell schemes. A recent study by Mukhopadhyay et al. (2010) was conducted for the simulation of Indian summer monsoon precipitation climatology. The Weather Research Forecast (WRF) model was used in the study with two nested domains having horizontal resolutions of 45 and 15 km with three convective schemes (Grell, Betts–Miller and Kain–Fritsch) for the monsoon seasons of 2001–2007. It was found that the Kain–Fritsch scheme seemed to have a high moist bias for the coastal Indian region, while the Grell scheme showed the opposite results. On the other hand, the Betts–Miller scheme produced reasonable estimates. Further analysis indicated that the Grell scheme overestimated the rainfall for the lighter rain rate category and underestimated for the moderate rain rate, while the heavy rain category was overestimated by the Kain–Fritsch scheme. In the light of the above, it was found that choice of scheme is strongly dependent upon the locality and topographical features, intensity of rain and horizontal grid spatial resolution. Considering the fact that the south Asian region, specifically in the geographical boundaries of Pakistan, has complex topography as described in the next section and that such studies have not been performed there, an assessment of which parameterized scheme is suitable for such region is significant. The reason why the specific years have been selected for this study is that the beginning of the year of 1998 brought a severe famine in the region, which lasted up to the year 2000 and then the year of 2001 brought a wet summer monsoon season. This drought left hazardous impacts on the agricultural yield of the region and, hence, this time interval is worth considering.
1.1. Climatology of the region in focus
The region simulated in present study consists of three major geographical areas. The northern highlands have rough topography and the rigours of climate. These areas enclose parts of some high mountain ranges of the Hindu Kush, the Karakoram and the Himalayas. The northern area has some renowned mountainous peaks as K2 (8475 m) and Nanga Parbat (7980 m) (Shamshad, 1988). The southeastern part of the region has the vast plains irrigated by the river Indus with its tributaries, and the Balochistan Plateau (Figure 1).
The region is located in the moderate zone and has a dry climate with hot summers and cold winters. Extensive variations persist between extremes of temperature at given locations. The monsoon circulation of Asia is the major source of precipitation in the region, especially in the summer (July to September). More than 22% of the World's population depends on the south Asian summer monsoon precipitation for their agricultural and economical needs and the rainfall in this season makes approximately 75% of the overall yearly precipitation in major areas of the region (Moetasim et al., 2009). All areas of the region, except for some western regions, receive higher precipitation in the summer monsoon than in the winter monsoon. The area of highest precipitation in the summer monsoon mostly consists of mountains.
The word monsoon is used for the winds that have a seasonal reversal in direction. With the onset of winter a strong high pressure system, the Siberian high, is created over Asia and winds flow off the Asian continent towards the inter-tropical convergence zone, a low pressure system, near the equator. These dry continental winds cause little precipitation for southern and southeastern Asian regions. With the start of summer, a low pressure system is created over northern India and produces high rainfall due to the flow of warm and moist winds from the adjacent oceans of the Arabian Sea and Bay of Bengal towards Asia (Lutgens and Tarbuck, 1979).
The Fifth-Generation NCAR/Penn State Mesoscale Model MM5, version 3 (available from ftp://ftp.ucar. edu/mesouser/MM5V3), has been used in this study to simulate the total monthly rainfall intensity/patterns during the summer monsoon seasons of 1998 and 2001. MM5 is the most recent version of a model first developed in the 1970s by Anthes (Anthes and Warner, 1978). A detailed description of MM5 can be found in Grell et al. (1995) and Dudhia et al. (2005). It is a limited-area, non-hydrostatic and terrain-following sigma coordinate, meteorological modelling system designed to simulate the complex meteorological phenomena occurring at the mesoscale, and perform numerical forecasting. The model includes multiple nesting capabilities, non-hydrostatic dynamics, four dimensional data assimilation (x, y, z, and t), different parameterization schemes (clouds, planetary boundary layer, humidity, radiation, surface temperature) and portability between different computational platforms (Grell et al., 1995). The model uses Arakawa-B staggering of the velocity vectors with respect to the scalars defining velocity vectors at the dot points and the scalars such as temperature and relative humidity etc. at the cross points (Ratnam and Cox, 2006).
A model domain was selected from 20 to 40°N and 60 to 80°E with spatial resolution of 30 km to focus on the region of interest for simulation of precipitation. Data of USGS (United States Geological Survey) comprising 24 categories of land use and 10 min spatial resolution for global topography were used. The details regarding 24 categories of land use data are given in Table I. Figures 1 and 2 present the topographic map and land use map respectively for the selected domain. For simplicity, some categories which have almost similar features are grouped together and are shown with same colours as shown in Figure 2.
Table I. Description of 24 categories of land use data
Categories of land use
Urban Land, Dry Cropland and Pasture, Irrigated Cropland and Pasture, Mixed Dry and Irrigated Cropland/Pasture, Cropland/Grassland Mosaic
For initial and boundary conditions, the archive of analysed data produced by the US National Centre for Environmental Prediction (NCEP) at 2.5° of spatial and 12 h of temporal resolution (http://dss.ucar.edu/ datasets/ds083.0/data/) was used. The spatial resolution of 30 km was taken for the selected model domain. Because of the limited computing power availability the resolution is set to 30 km. For instance, with a grid of 2220 km2 domain with 30 km horizontal grid spacing, a single PC of standard specification takes about 6 h to give one job output; if resolution is increased it may take even days. In this study the time step of 90 s was used for the integration of the domain while the simulation was performed for the whole month with the output being saved once every 2 h. As mentioned previously, there are different parameterization options (cumulus, planetary boundary layer, humidity, radiation, surface temperature) available in MM5, in this study the main emphasis is on the cumulus schemes in the model that were tested to identify a suitable convective scheme to predict the summer monsoon precipitation for July and August of 1998 and 2001 over the region of interest. The months of July and August were specifically considered because rainfall in these months defines the start and peak of the summer monsoon season, respectively, and contributes approximately 75% of the total annual rainfall in the region. The next step is the selection of other physical options with a specific cumulus convective scheme. There may be several combinations available for selection. This open choice of selection of a model configuration makes it difficult to choose an appropriate combination of these schemes for simulating the local climate of target region. Even if such a cumbersome and lengthy strategy is followed, it cannot be compared with other studies done in the nearby regions available in the literature. Therefore, a unique set of other physical options (Table II) with a cumulus convective scheme was selected for the present study, that was commonly used in other studies conducted in the nearby regions (Kerkhoven et al., 2006; Ratnam and Cox, 2006).
Table II. The physics options used in the MM5 model
In addition, the ‘Simple ice scheme’ (Schultz, 1995) was selected under the option of ‘Explicit moisture scheme’. This scheme deals with the rain water and ice cloud processes as well, without any requirement of additional PC memory. Moreover, this simple microphysical scheme does not include graupel or hail properties and may be suitable to reduce the complexity of the analysis. In the next step, under the heading of ‘Planetary Boundary Layer Scheme’, the ‘Medium Range Forecast Model’ was selected which was implemented in the National Centre for Environmental Prediction (NCEP) medium range forecast model by Hong and Pan (1996). This scheme was selected due to its computational efficiency in order to incorporate the atmospheric vertical diffusion processes due to turbulence in an unstable environment. Subsequently, under the option of ‘Radiation Scheme’, the ‘Cloud Radiation Scheme’ (Dudhia, 1989) was selected which is based on the interaction of both short and long wave radiation with environmental air and cloud, by considering absorption and scattering processes in the atmosphere. In addition, this scheme also considers the surface radiation fluxes. Finally the ‘Five layer soil model’ was selected under the option of ‘Surface Scheme’ (Dudhia, 1996). For the prediction of ground temperature, soil temperature and soil moisture etc., this multilayered surface scheme takes into account five soil layers of approximate thicknesses at 1, 2, 4, 8 and 16 cm with a fixed substrate located below.
In order to investigate the influence of different convective parameterization schemes on some other meteorological parameters such as ambient temperature, potential temperature, relative humidity and moist static energy, some simulations were performed with spatial resolution of 90 km for the model domain. The reason for selecting the coarse spatial resolution of 90 km in this case was only the scarcity of the data of fine resolution as 30 km for the validation purpose. The Reanalysis data of US National Centre for Environmental Prediction (NCEP) (http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis2.pressure.html) has been used to validate the model-predicted results for the vertical profiles of these parameters.
To validate the ‘model predicted monthly accumulated rainfall’ the results of MM5 were compared with TRMM satellite derived rainfall data (3B43RT) for the corresponding months while the Reanalysis data of US National Centre for Environmental Prediction (NCEP) for a specific location of ‘Islamabad’ within the domain studied was employed for the validation of model predicted results for the vertical profiles of above mentioned dynamic variables. The acquired results of entire simulations covering the four parameterized schemes of Grell, Kain–Fritsch, Anthes–Kuo and Betts–Miller for the months of July and August of the years 1998 and 2001 are discussed in the following sections.
3. Results and discussion
3.1. Precipitation during the months of July 1998 and July 2001
Figures 3 and 4 present the satellite derived (3B43RT) data of total monthly precipitation for July of the years 1998 and 2001, respectively. The climatological facts of the precipitation pattern over this area are obvious from these figures. The first thing in this regard is the indication of less precipitation in July 1998 and comparatively higher rainfall in the same month in 2001, which indicates that the years of 1998 and 2001 experienced dry and wet summer monsoon seasons, respectively. The second thing is the higher amount of precipitation over northern mountainous areas compared to the southern plain areas of the simulated region. Such precipitation patterns were expected due to the reason that winds coming from the adjacent oceans to the south Asian region in the summer monsoon experience extensive uplift when approaching the high mountain ranges of the northern areas, and produce heavy rainfall over there.
With regard to the performance of the parameterized schemes in the model, it can be observed that for July 1998 the satellite data (Figure 3) show rainfall with intensity of 100 mm over a broad northern area located at (32–34°N, 70–74°E) and peak rainfall with intensity of 200 mm over a very small area located at 33°N and 74°E. Comparing the simulated results of the Grell scheme (Figure 5) for the same month of the same year with the satellite data, it was observed that the Grell scheme captured the precipitation well over the northern areas though with a slight underestimation missing the small peak of rainfall of 200 mm and showing the precipitation with intensity of 100 mm over a comparatively small area, but at the same location as displayed in the satellite data. This scheme showed better performance for the southern plain areas by capturing the precipitation contours of 25 and 50 mm of satellite data. Moreover, the Grell scheme clearly demonstrated a greater amount of rainfall over northern mountainous areas of the region and comparatively less rainfall over the southeastern plain areas that is in good qualitative and quantitative agreement with the satellite derived data. On the other hand, the Kain–Fritsch scheme (Figure 6) overestimated results for both northern and southern areas of the simulated region, showing peak rainfall intensity of 400 mm though at the same location as in satellite data. It may be noted that, as with the Grell scheme, the Kain–Fritsch scheme, irrespective of rainfall intensity, seemed to be successful in capturing the location of peak rainfall as well as following the trend of higher precipitation over northern areas and less precipitation in plain areas as displayed in the satellite data. Colle et al. (2003) also showed the overestimated results by the Kain–Fritsch scheme and lesser precipitation by the Grell scheme, though Colle et al. (2003) investigated patterns of the total monthly precipitation over the coastal region of the United States.
For July 2001, the period of highest rainfall intensity in the present study, the comparison of model simulated data and satellite derived data illustrates that the Grell scheme (Figure 7), whilst capturing the peak rainfall with intensity of 100 mm over the southern plain areas centred at 24–25°N and 68–71°E well, which is somewhat in quantitative agreement with satellite data (Figure 4), the scheme largely underestimated the peak of rainfall with intensity of 400 mm located at (32–34°N, 70–74°E) as displayed in the satellite data showing precipitation with intensity of only 50 mm. With slightly overestimated results, on the other hand, the Kain–Fritsch scheme (Figure 8) showed a comparable trend with the satellite data showing maximum precipitation of 400 mm over the northern areas located at 32–34°N and 70–74°E, which is in good agreement with the satellite data. For the plain areas, this scheme again provided overestimated results. It may be noted here that the Kain–Fritsch scheme showed overestimated results over the northern region for the month of July 1998 which was a dry month and comparatively little amount of precipitation was observed over the mountain region. This strange behaviour of the KF scheme may be due to the performance dependence of convective schemes on precipitation intensity. This dependence has been studied by Mukhopadhyay et al. (2010) who investigated the overestimated and underestimated behaviour of the same convective scheme for different intensity levels of the precipitation.
3.2. Precipitation during the months of August 1998 and August 2001
For August 1998, comparing Figure 9 with Figure 10, it may be noticed that the Grell scheme (Figure 10) captured very well the intensities and locations of the two precipitation contours over the southern plain areas centred at 26–29°N and 66–68°E and 24–26°N and 68–71°E of intensities 25 and 50 mm respectively, showing an excellent qualitative and quantitative agreement. With slight underestimation, this scheme showed also reasonable results for northern mountainous areas showing peak rainfall with intensity of 100 mm over the same place (32–34°N, 72–74°E) as displayed in Figure 9. On the other hand, the Kain–Fritsch scheme (Figure 11) again presented overestimated results for both mountainous and plain areas, indicating peak rainfall with intensity of 300 and 100 mm centred at 32–34°N and 72–74°E and 26–29°N and 66–68°E, respectively, when compared with the satellite data (Figure 9). It may be noted that the Grell scheme seemed to be successful in capturing precipitation patterns over the plain areas. This is in line with the findings of Kerkhoven et al. (2006) who also found that the Grell scheme is the most robust scheme for the region of the east China plains for summer monsoon precipitation.
The satellite data (Figure 12) of August 2001 reveals that most of the areas of simulated region experienced very small amount of precipitation throughout the month. A single peak of rainfall with intensity of 100 mm for the total month, was again observed over the north at 32–34°N, 72–74°E. Comparing the satellite derived precipitation in Figure 12 with that predicted by the Grell scheme in Figure 13 for August 200l, it was noted that the Grell scheme captured the peak of rainfall with intensity of 100 mm, though over a smaller area compared to that displayed in satellite data, but at the same place in the recorded data. On the other hand, this scheme provided satisfactory results for the southern plain areas with slight overestimation indicating peaks of rainfall of intensity of 50 mm at scattered locations when compared with satellite data. Here it may be noted that the Grell scheme indicated slightly overestimated results for the plain areas for the first time in this study. The reason for this behaviour may be attributed to the fact that in this month very little amount of precipitation was observed over the plain areas. This is consistent with the findings of Mukhopadhyay (2010) who concluded that the Grell scheme gives the overestimated results for lighter precipitation. On the other hand, the Kain–Fritsch scheme again showed slightly overestimated results for both northern mountainous and southern plain areas indicating a single peak of rainfall with intensity of 200 mm (Figure 14) at exact location of maximum rainfall with intensity of 100 mm as displayed in satellite data (Figure 12).
On similar lines, the intensity and pattern of precipitation for July and August for 1998 and 2001 were also investigated for the Anthes–Kuo and Betts–Miller schemes. When compared with the satellite data, both the schemes were unable to provide appropriate results over the region of interest, showing either under-predicted precipitation or over-estimated contours of rainfall patterns, while the locations of rainfall were also found to be substantially out of phase. For this reason, only typical figures are being referred to here instead of including all of them. For instance, Figures 15 and 16 show the accumulated monthly precipitation in millimetres for August 1998 as predicted by the Anthes–Kuo and Betts–Miller schemes, respectively. Both the schemes predicted 200 mm precipitation contours at locations extending from 68 to 72°E and from 28 to 32°N that are quantitatively as well as qualitatively in disagreement with the satellite data (Figure 9). The previous studies of Wang and Seaman (1997) and Ferretti et al. (2000) also showed the poor performance of the Anthes–Kuo and Betts–Miller schemes in capturing the precipitation patterns over the Alpine region of Italy and continental region of USA. In both studies the Kain–Fritsch scheme showed better performance in capturing both the intensity and location of maximum precipitation
To find out the exact reasons responsible for the superiority of one convective scheme over others is not the objective of present study. However, the plus point that seems to favour the better performance of Grell scheme may be its close assumption that mixing in deep convection occurs at the cloud top rather than through its sides in plume-like fashion, hence interaction of cloudy air with the environmental air occurs only at the top of the cloud (Reuter, 1986). On the other hand, the feature which may support the Kain–Fritsch scheme is probably the inclusion of the entrainment and detrainment processes in the formation of cloud. The other reason for better performance of the Kain–Fritsch scheme, especially over high mountainous regions, may be the triggering factor of the scheme related to the uplift, which in a region of complex topography may play a major role. The failure of the Betts–Miller scheme may be due to the use of a single stable moisture profile in scheme which is based on observations of tropical storms that prevents its application to other environments. The Anthes–Kuo scheme is also unable to show better performance, probably due to its unrealistic assumptions regarding the nature of atmospheric convection. In view of the above, the Anthes–Kuo and Betts–Miller schemes may not be suitable for the complex terrain of the south Asian region.
3.3. Simulated results for vertical profiles of ambient temperature, potential temperature, relative humidity and moist static energy
In order to evaluate the variation in above mentioned parameters with increasing height, monthly averaged vertical profiles of these parameters were plotted for August 1998 and compared with the Reanalysis data of the US National Centre for Environmental Prediction (NCEP) for the corresponding month. Simulations were conducted taking the model domain with spatial resolution of 90 km. The specific location (33.6°N, 73.1°E) of the city of Islamabad was selected for the purpose of validation. This is because the selected area was always observed to be receiving the peak of rainfall throughout the analysis.
Figure 17 illustrates a comparison of monthly averaged ambient temperatures predicted by different convective schemes (lines) at different vertical heights with that of NCEP data (dot) for August 1998. This figure depicts an overall decreasing trend in temperature with increasing height. A close agreement can be seen between the observed and model predicted temperatures. The other noticeable thing is that the results of the Grell and Betts–Miller schemes are closer to the observed ones as compared to those by other schemes, though it is difficult from Figure 17 to distinguish the curves of different schemes because the plot covers a wide range of data presented for the whole vertical domain. Figure 18 shows the vertical variation in monthly averaged potential temperatures predicted by the different schemes (lines) with the increase in height for August 1998. As the potential temperature is a dynamically important quantity more than the actual temperature, the model simulated results for this parameter are also presented and compared with the NCEP data (dot). Potential temperature is a useful measure of the static stability of atmosphere. For a stable atmosphere, potential temperature increases with height, while it decreases with height for an unstable, convective, atmosphere. One thing that is very clear from Figure 18 is the presence of a steep potential temperature gradient in the upper atmosphere and a very small temperature gradient, almost zero, at the lower heights. This trend indicates that the upper atmosphere is more stable than the lower. The almost constant value of potential temperature at the lower heights is indicative of weak convection, as expected, due to the fact that the year 1998 had experienced little convective precipitation throughout the summer monsoon season. The comparison of model-predicted results with the observed data of NCEP shows again that the results of the Grell and Betts–Miller schemes are closer to the observed data than those of the other schemes.
Figure 19 illustrates the comparison of monthly averaged relative humidity predicted by different schemes (lines) with that of the observed data of NCEP (dot) for August 1998. Comparing the simulated results with NCEP data, it was observed that both the Grell and Betts–Miller schemes were again successful in capturing the overall trend of vertical variation in relative humidity with increasing height, although neither scheme was able to generate the values of humidity in close agreement to those of the NCEP data. The slight mismatching of model-predicted and NCEP data is most likely due to the difference between the spatial resolutions of the model domain and the NCEP dataset. Figure 20 depicts the comparison of the model-predicted vertical profiles of monthly averaged moist static energy for different convective parameterization schemes (lines) with that of the data of NCEP (dot) for August 1998. It is again clear from Figure 20 that both the Grell and Betts–Miller schemes seemed to be successful in capturing well the overall trend of vertical variation in moist static energy with increasing height compared to the other convective schemes. In the light of the above discussion on the results of precipitation, ambient temperature, potential temperature, relative humidity and moist static energy, one thing is very clear that the Grell scheme performed well for all the meteorological parameters over both resolutions for the specific month of August 1998. On the other hand, the Betts–Miller scheme, which produced overestimated results with 30 km resolution for the same period in the case of precipitation, seemed to be successful in generating reasonable results with 90 km model resolution for the dynamic variables of temperature, relative humidity and moist static energy. This behaviour of the scheme clearly indicates the dependence of a convective scheme on spatial resolution of the model, as reported also in previous studies (Kerkhoven et al., 2006; Ratnam and Cox, 2006).
4. Conclusions and recommendations
In the present work, the parameters of summer monsoon precipitation, ambient temperature, potential temperature, relative humidity and moist static energy for July and August 1998 and 2001 were simulated using the MM5 model. The model predictions have been compared with the TRMM satellite-derived product of tropical rainfall (3B43RT) data and the reanalysis data of US National Centre for Environmental Prediction (NCEP) for the city of Islamabad. It is observed that the ability of the MM5 physical model to predict reasonable rainfall patterns, intensities and the other meteorological parameters (ambient temperature, potential temperature, relative humidity and moist static energy) is highly dependent on the cumulus parameterization schemes used in it.
Among the schemes, the Grell scheme generated better results in the case of rainfall over the plain areas of the region which receive comparatively less precipitation throughout the year than the northern mountainous areas of the simulated region. This scheme seemed to be successful as well in northern mountainous areas in cases of lesser precipitation intensity, giving comparable results with satellite data, but it largely underestimated the precipitation over these areas in cases of heavy rainfall approaching an intensity of 400 mm. The better performance of the Grell scheme may be due to the reason that assumptions considered in this scheme are independent of any specific environments. On the other hand, the Kain–Fritsch scheme captured the rainfall patterns and intensity reasonably well over the northern mountainous region in cases of heavy rainfall, while slightly over-predicting the precipitation for the southern plain areas. The better performance to capture the rainfall intensity as well as its location may be due to the entrainment/detrainment processes included in this scheme. The other reason the Kain–Fritsch scheme has a better performance over high mountain ranges may be the triggering factor of this scheme related to the uplift, which in a region of complex topography may play a major role. The Anthes–Kuo and Betts–Miller schemes remained unable to produce realistic results for rainfall in this study, indicating their unsuitability in this region. One possible reason for this incapability may be that the Anthes–Kuo and Betts–Miller schemes do not consider the downdrafts which may play a role in enhancement of accuracy of the simulation and another reason may be the comparatively fine resolution of 30 km selected for the model domain which might be unsuitable for the performance of these schemes.
Comparing model-predicted results of different convective schemes for the ambient temperature, potential temperature, relative humidity and moist static energy, it was found that the Grell scheme again showed better performance even with the model spatial resolution of 90 km. The Betts–Miller scheme which remained unable to capture the precipitation well with model resolution of 30 km, seemed to give reasonable results for these meteorological parameters with model resolution of 90 km. It was a clear indication of the dependence of different convective schemes on the spatial resolution of selected model domain.
Since the present study is limited to the region of Pakistan, further research for similar regions is recommended to substantiate the conclusions drawn. It is expected that the set of physical options that is valid for precipitation, temperature, relative humidity and moist static energy in a target region may further be investigated for some other environmental parameters such as wind speed and wind direction. The effect of precipitation intensity and spatial resolution of the model domain with recommended schemes in this region may also be worth consideration. Simulations performed in this study were computationally intensive, hence each run takes almost 6 h on a single PC: this would be even longer if finer resolutions are considered. With the availability of computer clusters and high performance machines this time may be reduced substantially and finer resolution for nested domains with the different combination of convective schemes may be studied.
The authors highly appreciate the financial support provided by the Higher Education Commission (HEC), Islamabad. Authors are also thankful to Grid Computing Group, PIEAS for providing technical support.