Simulation of intense organized convective precipitation observed during the Arabian Sea Monsoon Experiment (ARMEX)



[1] We examine a deep precipitating system that formed over the west coast of India during 26–28 June 2002 producing heavy rainfall of 2–61 cm day−1. The system developed into a well-marked low pressure area due to interaction between an eastward moving westerly trough and a westward moving monsoon low. We used the PSU/NCAR Mesoscale Model (MM5) to make 10-day interactively nested simulations at 90, 30, and 10 km grid-resolutions. We used observations from a special data set collected during an Arabian Sea Monsoon Experiment (ARMEX) conducted June–August, 2002. Nudging the observations produced a balanced atmosphere which, in turn, matched the location of vortex with cloud clusters observed from satellite. The simulated rainfall corresponded well with observations. We then use the simulated fields as forcing and boundary conditions for MM5 run in cloud-system resolving mode at 2 km grid-resolution to detail cloud-clusters embedded within the monsoon disturbance. We examined the sensitivity to different physical parameterizations and also the effect of continuous nudging four dimensional data assimilation (FDDA) on the rainfall forecasts. While the simulation of the convective event improved with certain combinations of physical parameterizations, the rainfall was not forecasted at the correct location, no matter which parameterization was used, unless continuous FDDA was performed in all domains throughout the integrations. Finally, cloud-cluster properties of the cloud-system resolving simulations were compared with observations.

1. Introduction

[2] Deep precipitating convective systems often form over the west coast of India producing heavy rainfall of the order of 10–50 cm day−1 during their mature stage. Such events are generally associated with an offshore trough at the sea level, a cyclonic circulation extending from lower to the middle troposphere and a wind shear line extending from the Arabian Sea to the Bay of Bengal across the southern peninsular India or across the Indo-Gangetic plains during certain phases of the south Asian summer monsoon season. During the onset phase in June and also at times during the revival phase particularly in August, the shear line may be positioned over peninsular India. During the active and established phases in July it is positioned over the Gangetic plains. On occasions, an eastward-moving upper-level westerly trough (also known as western disturbances) interacts with the westward-moving monsoon lows at lower levels along the monsoon trough over the Indo-Gangetic plains. The two systems intensify interactively, forming a deep depression that produces heavy rainfall over the region. Such systems can significantly influence the large-scale monsoon circulation through thermodynamic interaction. While many studies have been carried on the synoptic features associated with the western disturbances and monsoon lows [Krishnamurti and Hawkins, 1970; Ramanathan and Shah, 1972; Rao, 1976, Hatwar et al., 2005], very few studies have focused on the dynamics of the system and the role of cloud microphysics parameterization.

[3] Intense rainfall events on the west coast of India are also associated with mid-tropospheric cyclones (MTC), and organized convection in the tropical convergence zone [Mak, 1975; Krishnamurti et al., 1981; Benson and Rao, 1987; Ogura and Yoshizaki, 1988; Carr, 1997; Gadgil and Francis, 2001; De and Dutta, 2006]. Studies have shown that about 10% of the intense rainfall events are associated with MTC, but the causes of the remaining 90% of the intense convective events have been little studied. Does intensification of convection over the Bay of Bengal and monsoon trough zone provide large-scale conditions favorable for the growth of systems off the west coast? In order to understand such issues, an Arabian Sea Monsoon Experiment (ARMEX) was conducted during June–August, 2002 to study the convection associated with intense rainfall events on the west coast of India including the genesis, intensification and propagation of convective systems over the eastern Arabian Sea leading to intense rainfall events along the west coast [Rao, 2005; Sikka, 2005; Rao and Sikka, 2005; Bhat et al., 2005; Das Gupta et al., 2006]. Intensive observation periods (IOP) were designed to collect special observations from surface-based facilities (land, ship and buoys), upper-air stations, radar, aircraft and satellites.

[4] We use ARMEX data in conjunction with mesoscale and cloud-system resolving simulations to examine cloud systems resulting from interaction between a middle-latitude westerly trough and a tropical depression. Numerous previous studies have examined mesoscale convective systems in midlatitudes [i.e., Zhang and Fritsch, 1988; Weisman and Davis, 1998; Zhang, 1992; Wang et al., 1993; Davis and Weisman, 1994; Trier et al., 2000, Xu and Randall, 2000] and the tropics [i.e., Xu and Randall, 1996; Moncrieff and Klinker, 1997; Moncrieff et al., 1997; Wu et al., 1998; Liu et al., 2001a, 2001b among others]. However, few have examined interaction between midlatitude and tropical weather systems. The generation mechanism is key to both. Zhang and Fritsch [1988] noted that interaction among a propagating mesoscale vorticity disturbance, pre-existing low-level frontal forcing and a convectively favorable environment ahead of the front generates an organized area of upward motion conducive to vortex development. In a study of a mesoscale vortex that caused heavy precipitation in the lee side of the Tibetan Plateau, Wang et al. [1993] found that midlevel vorticity stems from diabatic heating associated with the MCS.

[5] Many questions remain to be answered, for example: Which system (i.e., tropical or extra-tropical) plays the dominant role in the evolution of a monsoon disturbance? Is the system warm core or a cold core? What are the origin, strength and scale of these vortices? Do they depend on ambient vertical wind shear; scale of the convective segment, and life-cycle of the convective system? What are the cloud microphysical structures of these systems? Can the intense precipitation observed during the monsoon disturbances be predicted? What is the optimum resolution for their simulation?

[6] We address some of these questions, in particular, those involving the dynamics of the intense convective systems and cloud microphysics over the Arabian Sea adjoining west coast of India. Using the MM5 mesoscale model with the nested option, we simulate heavy precipitating event associated with a marked low pressure area that resulted from interaction between a monsoon low and an extra-tropical westerly trough: a convective system over the west coast of India during 26–28 June 2002. During this period the ARMEX field-campaign measurements are used for initial/boundary conditions and to nudge the model toward more accurate rainfall forecasts. These reanalyzed fields are then used to force the MM5 run in cloud-system-resolving mode in order to quantify the role of cloud-microphysics and meso-convective dynamics. Section 2 summarizes the observed structure of the convective system. Results of the numerical experiments based on different cloud-physical parameterizations and nested domains are discussed in section 3. The elementary four-dimensional data assimilation (nudging) using ARMEX observations is presented in section 4. Section 5 presents the results of cloud-system resolving simulations. The paper concludes with a summary in section 6.

2. Observed Structure of the Convective System

[7] Several episodes of heavy rainfall occurred over the west coast of India during the 2002 summer monsoon. We examine a typical case that involved interaction between a westerly trough traveling eastward and a monsoon low traveling westward during 18–28 June 2002. Figures 1 and 2illustrate the streamline patterns at 850 and 500 hPa for 00 UTC of 19, 23, 25, and 28 June 2002. A monsoon low originated over the head of the Bay of Bengal on 18–19 June [(Figure 1a)] and moved westward across the Indo-Gangetic plains. During the same period, an upper-level westerly trough approached India from Iran/Afghanistan [(Figure 2)]. The heavy line on these diagrams demarks the position of the westerly trough at 500 hPa. The two systems merged to produce a distinct low-pressure area and rainfall ranging from 2–61 cm day−1 over the west coast, particularly over the Gujarat region during 27–28 June [(Figure 1)]. Miller and Keshavamurthy [1968] suggested that the west coast rainfall is associated with offshore trough and midtropospheric cyclones (MTC) embedded with mesoscale bands. They suggested that the formation of an MTC is linked to a disturbance moving into the region from the Bay of Bengal.

Figure 1.

Streamline analysis at 850 hPa for 00 UTC of (a) 19th (b) 23rd (c) 25th (d) 28th June 2002.

Figure 2.

As in 1 but for 500 hPa.

[8] The system produced extensive cloud bands spanning several hundreds of kilometers with embedded mesoscale structures as evident in the METEOSAT cloud imagery [(Figure 3a)]. The sharp edge of the cloud is due to strong wind-shear caused by merging westerly and easterly systems. The advection of cyclonic vorticity at upper and lower levels formed a strong vortex. The mean sea level pressure dropped to about 994 hPa [(Figure 3b)] forming the depression over Gujarat region. Figure 3b also shows an offshore trough. The water vapor channel [Mohanty et al., 2002] showed intense moist and dry air regions side-by-side over Gujarat.

Figure 3.

(a) METEOSAT cloud imagery and (b) mean sea level pressure (+1000 hPa) on 00 UTC 28 June 2002. Strong pressure gradient in the north is due to the Himalayas.

[9] Figure 4 shows the observed widespread rainfall distribution over the west coast during 26–28 June 2002. The highest rainfall (61 cm) was recorded in Gujarat at Pardi, 54 cm at Valsad, and 42 cm at Bhavnagar over a 24 h period. The observed distribution of rainfall was obtained by merging rain gauge and TRMM observations [Mitra et al., 1997]. Deep, strong low-level moisture convergence associated with cyclonic vortices triggers deep (parameterized) cumulonimbus convection whose stratiform regions cause grid-scale precipitation. The result is a convective complex and intense rainfall. The lifecycle of the convective complex may span for 1–2 days. Figure 3a also indicates cloud bands spiraling over a large area. Simulation of the dynamical and cloud-microphysical structure of such a system is a numerical modeling challenge.

Figure 4.

The 24 h observed rainfall in cm (obtained from merged TRMM and rain Gauges analyses) at 03 UTC on 26, 27, and 28 June 2002. The box shown in the fourth diagram is used for averaging the rainfall in Figure 9.

[10] In order to evaluate the spatial-temporal evolution of convective activity, the daily averaged outgoing long-wave radiation (OLR) measured from NOAA-AVHRR satellite [Gruber and Krueger, 1984] is shown in Figure 5 for the period 26–28 June 2002. The lowest OLR (deepest cloud) of about 120 Wm−2 occurred over northeast Arabian Sea and adjoining Gujarat and Maharashtra coast on 26 June. On the same day the anomalous convective activity increased as evident in the negative OLR anomaly (OLRA) of about 120 W m−2. Deep convection prevailed on 27 and 28 June with negative OLRA around 90 watts m−2.

Figure 5.

Outgoing long-wave radiation (OLR) and OLR anomaly (OLRA) in watts/m2 during 26–28 June 2002. The negative anomalies are shaded.

[11] The precipitable water (PWAT) and its anomaly (PWATA) obtained from the NCEP/NCAR [Kalney et al., 1996] reanalysis data set (T62 spectral resolution) (not shown) indicates that on 26 June the moist layer deepened with a positive anomaly exceeding 10 kg m−2 over a large region of central India and subsequently extended northeastward during 27–28 June. The deep moist layer extending from Gujarat and Maharastra to the states of Harayana, Punjab and west UP are manifestations of wind circulation as seen in Figure 2d. The presence of the deep moist layer resulted in increased convective activity over Gujarat and Maharashtra region, causing widespread rainfall during the entire 26–28 June period.

[12] The Tropical Rainfall Measuring Mission (TRMM) satellite provides excellent data sets related to rainfall rates and vertical profiles of cloud hydrometeors on a daily basis: cloud liquid water (CLW), rainwater (RNW), snow, ice and latent-heating rates. The TRMM flies precipitation radar (PR) and a microwave imager (TMI) that measures the outgoing microwave radiation at 10.4, 19.4, 21.3, 37, and 85.5 GHz frequencies [TRMM, 1996]. The PR has a data swath of 215 km and TMI 780 km. Together the PR and TMI data provide a unique measure of synoptic-scale rainfall. Figure 6 illustrates the instantaneous rainfall, vertical cross sections of the cloud hydrometeors and latent heating rates during 26–28 June 2002. The domain averaged (69–73°E and 19–24°N) vertical profiles of the cloud microphysical properties observed by TRMM for each of the three days are shown in Figure 7. The time averaged profiles of the hydrometeors are also shown. Intense rainfall rates up to 32 mm h−1 are observed at different locations over the coast of Gujarat depending upon the passage of the satellite. Pockets of intense convection are seen. Vertical cross sections of the cloud microphysical properties retrieved along east–west direction at the latitude of the maximum rainfall on each day are illustrated in Figure 6. The characteristics of the convective cells depend upon their intensity. The CLW is confined below 10 km and extends almost down to surface on 26 June during intense rainfall. The maximum CLW is about 80 mg m−3; rainwater is confined below 8 km with a maximum of 1400 mg m−3; snow occurs above 2 km with a maximum of about 1400 mg m−3; and ice occurs above 6 km with a maximum of 170 mg m−3. The maximum latent heating is about 15 K h−1 and elevated at about the 9 km level. The magnitudes of the hydrometeors seem high, likely because these cross-sections are for the maximum precipitation observed by the TRMM.

Figure 6.

TRMM observations for rainfall (mm hr−1), rainwater (RNW), cloud liquid water (CLW), snow and ice in units of mg m−3 and latent heating rate (LH) in K hr−1. Upper, middle and lower panels are for 26, 27, and 28 June respectively. The vertical cross sections of RNW, CLW, snow, ice and LH are drawn along the line of maximum rainfall as shown in the diagrams.

Figure 7.

Time mean vertical profiles of (a) Rainwater, RNW (b) cloud liquid water, CLW (c) SNOW, (d) ICE, and (e) latent heating rate averaged over the domain (69–73°E and 19–24°N) obtained from TRMM during 26–28 June 2002.

3. Simulations Employing Different Physical Processes and Nested Domains

[13] We use MM5, a non-hydrostatic, terrain-following sigma coordinate model [NCAR, 2003], which was adapted for real-time mesoscale weather forecasts at the National Centre for Medium Range Weather Forecasting (NCMRWF) in India [Das, 2002]. We evaluate its usefulness for advance warning of heavy convective rainfall events. Table 1 summarizes the initial model setup and parameter values used. We apply triple-nested domains at (90, 30, and 10 km grid-resolutions) as shown in Figure 8. We conducted more than 20 experiments (summarized in Table 2) using different combinations of physical parameterization schemes, nested domains and four dimensional data assimilation (FDDA). Initial and lateral boundary conditions are provided by the operational global T80 model of NCMRWF.

Figure 8.

Domains at 90, 30, 10, and 2 km resolutions over India and neighborhood.

Table 1. Summary of the MM5 Model as Implemented at NCMRWF
  Horizontal (triple nested)
   Outer domain-1 (39 E – 121 E, 11 S – 51 N) at 90 km resolution
   Middle domain-2 (64 E – 100 E, 4 N – 45 N) at 30 km resolution
   Inner domain-3 (73 E – 83 E, 32 N – 40 N) at 10 km resolution
   Staggered Arakawa B-Grid
  Vertical: 23 Levels (sigma-hybrid), being increased to 42
  Time steps: Domain-1: 270 S, Domain-2: 90 S, Domain-3: 30 S
  Topography: USGS (interpolated depending on resolution)
  Vegetation/land use: 25 Categories (USGS)
  Two-way nesting
  Prognostic equations: U, V, T, q, Ps, (optional: TKE, cloud water, rain, ice, snow, graupel/hail)
  Time integration: semi-implicit
  Boundary conditions
   Upper radiative condition
  Horizontal diffusion
    Fourth order for inner domains
    Second order for the coarser domain
  Cumulus parameterization: Grell (simplified Arakawa-Schubert)
  Explicit moisture schemes: simple ice (Dudhia)
  Radiation scheme: simple cooling
  Land surface processes: five layer soil model
Table 2. Summary of Experiments
S. No.Name of ExperimentsDescription
1CUP-GRLCumulus parameterization (Grell), Simple ice Microphysics (Dudhia), Planetary Boundary Layer (MRF), Cloud-Radiation scheme, 5-layer soil model
2CUP-BMIAs in CUP-GRL, but Betts-Miller convection scheme
3CUP-KF2As in CUP-GRL, but Kain-Fritsch scheme
Cloud Microphysics
4CLM-SICEAs in CUP-GRL with Simple ice scheme
5CLM-GSFCAs in CUP-GRL, but GSFC cloud Microphysics
6CLM-RSNRAs in CUP-GRL, but Reisner2 cloud Microphysics
7CLM-SULZAs in CUP-GRL, but Schultz cloud Microphysics
Planetary Boundary Layer
8PBL-MRFAs in CLM-GSFC with MRF PBL scheme
9PBL-MYAAs in CLM-GSFC, but Mellor-Yamada PBL scheme
10PBL-BLAAs in CLM-GSFC, but Blackadar PBL scheme
11RAD-CRINAs in PBL-MYA with Cloud-Radiation scheme
12RAD-CCM2As in PBL-MYA, but CCM2 Radiation scheme
13RAD-RRTMAs in PBL-MYA, but RRTM Radiation scheme
Land Surface Processes
12LSP-MSMAs in RAD-CCM2 with 5-Layer Soil Model scheme
13LSP-OSUAs in RAD-CCM2, but with OSU Land Surface scheme
14EXP-DOM1Single Domain at 90 km horizontal resolution with Physics as in CUP-GRL
15EXP-DOM22 Nested domains at 90 and 30 km horizontal resolutions
16EXP-DOM33 Nested domains at 90, 30 and 10 km resolutions
17FDDA-DAY1Four Dimensional Data Assimilation (FDDA) with cyclic nudging and CUP-GRL for 1 day
18FDDA-DAY4FDDA with cyclic nudging for 4 day
19FDDA-DAY7FDDA with cyclic nudging for 7 day
20FDDA-DAY10FDDA with cyclic nudging for 10 day

3.1. Experiments With Different Physical Parameterization Schemes

[14] We first simulated rainfall observed during 27–28 June using different combinations of physical parameterization scheme. Table 2 presents the list of experiments conducted using different parameterization schemes of physical processes: deep convection, cloud-microphysics, boundary layer, radiation and, land surface processes. The first set of experiments involving cumulus parameterization (CUP) consisted of the Grell et al. [1994], Betts and Miller [1993] and the Kain and Fritsch [1993] schemes. They are named CUP-GRL, CUP-BMI and CUP-KF2 respectively in the Table 2. The MM5 model was integrated for 72 h based on the initial and boundary conditions of 00 UTC, 25 June 2002. Figure 9 presents the sensitivity of simulated rainfall to different parameterizations at 30 km horizontal resolution. The simulated and observed (24 h accumulated) rainfall have been averaged over the region of intense precipitation (67–77°E, 15–27°N) shown by the box in Figure 4d. Results indicate that all the schemes underestimated the rainfall on 26, 27 and 28 June. While the BMI closely matched the observed rainfall on 28th, the GRL produced relatively better rainfall on 26 and 27th June. The observed trend of rainfall during the three days period was better represented by the CUP-GRL. Rao and Prasad [2006] [hereafter referred to as RP], Mandal and Mohanty [2006] [hereafter referred to as MM], Trivedi et al. [2006] [hereafter referred to as TMV] studied the sensitivity of different physical processes on the simulation of track and intensity of the Orissa super cyclone over the east coast of India using the MM5 model. The studies of RP and MM indicated that, while the intensity of the tropical cyclone (lowest sea level pressure and maximum wind) was better simulated by both GRL and KF2 schemes, it was better simulated by the KF2 and Anthes-Kuo schemes in the study of TMV. The study of RP also noted that while the track of the cyclone was better predicted by the KF2, the intensity was better simulated by both KF2 and GRL schemes. We present the spatial distribution of simulated rainfall by the CUP-GRL scheme for 28th June in Figure 10a. The diagram shows that the rainfall values are largely underestimated compared to observations [(Figure 4)], and the area of maximum rainfall is confined to the west of the actual maximum.

Figure 9.

Sensitivity of different physical processes on the simulation of rainfall. See Table 2 for details of the experiments. Simulations are based on the initial conditions of 00 UTC, 25 June 2002 and carried out for 72 h.

Figure 10.

Simulated rainfall (cm) valid for 00UTC, 28 June 2002 based on different combinations of physical parameterization schemes (see text for details). Shaded area indicates rainfall between 1 to 10 mm.

[15] The second group of experiments evaluated the sensitivity of simulated rainfall to different parameterizations of cloud microphysics (CLM) namely the simple ice scheme of Dudhia [1989], the GSFC [Tao et al., 1989, 1993], the Reisner et al. [1998] and the Schultz [1995] schemes. They are named CLM-SICE, CLM-GSFC, CLM-RSNR and CLM-SULZ respectively in the Table 2. We used the GRL for convection in these sets of experiments as it was found to produce relatively better precipitation in the previous set of experiments. The simulated precipitations obtained from these experiments are shown in Figure 9b. All the schemes underestimated the rainfall during the three days. Results indicate that while the SICE and GSFC schemes were relatively closer to the observed mean, the trend of rainfall was relatively better simulated by using the GSFC. The studies of RP showed that the mixed phase (MP) scheme produced relatively less track errors in the case of the Orissa super cyclone, but the intensity of the cyclone was best predicted by both SICE and MP schemes. The spatial distribution of simulated precipitation by this scheme is presented in Figure 10b for 28 June. The diagram shows that the rainfall values are largely underestimated compared to observations [(Figure 4)], and the area of maximum rainfall is confined along the west coast.

[16] The third set of experiments evaluated the sensitivity of simulated rainfall to different parameterizations of the planetary boundary layer (PBL) namely the non-local closure scheme of Hong and Pan [1996], the turbulent kinetic energy based closure scheme of Mellor and Yamada [Janjic, 1994] and the Blackadar [1979] schemes. They are named PBL-MRF, PBL-MYA and PBL-BLA respectively in the Table 2. Here we used the combination of GRL and GSFC schemes which had produced best results in the previous experiments. The simulated rainfalls obtained from these experiments are shown in Figure 9c. All the schemes underestimated the rainfall on 26, 27, and 28 June as in the earlier experiments. Results indicate that while the MRF and MYA schemes are relatively close to the observed mean, the trend of rainfall was relatively better simulated by the MYA. The studies of RP had suggested that the MYA scheme simulated better intensity of the tropical cyclone, but the track errors were less than MRF in their studies. The studies of MM and TMV indicated that the MRF scheme was better. Clearly, no combination of physical processes produces best results universally. We present the spatial distribution of simulated rainfall by the PBL-MYA scheme for 28 June in Figure 10c. The diagram shows that the rainfall values are largely underestimated compared to observations [(Figure 4)], and the area of maximum rainfall is confined over the sea along the west coast.

[17] The fourth set of experiments evaluated the sensitivity of simulated rainfall to different parameterizations of radiation (RAD) namely the cloud-radiation interaction scheme involving explicit cloud and clear air, the CCM2 and the RRTM radiation schemes. They are named RAD-CRIN, RAD-CCM2 and RAD-RRTM respectively in the Table 2. These schemes were used in combination with the GRL, GSFC and MYA, which produced the best simulation of rainfall thus far. The simulated rainfalls obtained from these experiments are shown in Figure 9d. All the schemes underestimated the rainfall during 26–28 June as in the earlier experiments. Results indicate that both CRIN and the CCM2 produce rainfall close to the observation, particularly on 27 and 28 June. The spatial distribution of simulated rainfall by the RAD-CCM2 scheme is shown in Figure 10d for 28 June. The diagram shows that while the simulated rainfall distribution improved over the west coast, the values were still highly underestimated.

[18] The final group of physics experiments evaluated the sensitivity of simulated rainfall to different land surface processes (LSP) schemes namely the multi layer (5 layers) soil model and the Noah land surface model. They are named LSP-MSM and LSP-OSU respectively in the Table 2. These schemes were used in combination with the GRL + GSFC + MYA + CCM2, which had produced the best simulation of rainfall thus far. The simulated rainfalls obtained from these experiments are shown in Figure 9e. As in the previous experiments, all the schemes underestimated the rainfall during 26–28 June. The best simulation of rainfall was produced by the MSM.

[19] Note that these results do not necessarily represent the best combination of physical processes in the statistical sense. However, the simulated mean rainfall in all these experiments do not differ from each other significantly. The mean difference from observation is usually largest during the first two days.

3.2. Experiments With Different Nested Domains

[20] In order to examine the impact of different nested domains on the simulation of rainfall, we conducted three sets of experiments listed in the Table 2. The first experiment (EXP-DOM1) had single domain at 90 km resolution, the 2nd experiment (EXP-DOM2) had two nested domains at 90 and 30 km resolutions with two-way feed back and, the third experiment (EXP-DOM3) had three nested domains at 90, 30, and 10 km resolutions with two-way feed back. The simulated rainfall shown in Figure 9f was not much improved by the increased resolution. While higher resolution produced a finer structure of the rainfall distribution [(Figure 10f)], the values were still severely underestimated.

[21] In summary, while all combinations of parameterization schemes and the nested domains produce rainfall distributions over the west coast, none resembles the observational measurements: all underestimate the rainfall. The area of maximum precipitation is generally pushed north–eastward.

4. Four-Dimensional Data Assimilation

[22] In an attempt to improve convective rainfall forecasts, experiments were carried out with four dimensional data assimilation (FDDA) by nudging observations at different resolution as now described. FDDA is a technique by which observations are incorporated in the model running with full moist physics [Stauffer and Seaman, 1994; Davis et al., 2001; Liu et al., 2002a, 2002b; Hsu and Liu, 2002]. Observational measurements keep the model close to the true state and the model atmosphere tends to dynamical consistency. This alleviates errors in the initial analysis as well as deficiencies in the model physics (i.e., convection, boundary layer and micro-physics parameterizations). There are two methods: (a) analysis or grid-nudging and (b) observational or station nudging. We examined the analysis nudging.

[23] In the analysis nudging, Newtonian relaxation terms are added to the prognostic equations for wind, temperature, and water vapor using the equation

equation image

where α is a prognostic variable (u, v, t, q), equation image the domain-averaged quantity, the subscript ‘obs’ indicate observation and τ is the relaxation timescale. This procedure rapidly relaxes the simulated domain-averaged variable toward observations when τ is small. If the domain-averaged quantity is instantaneously adjusted to the observed value (i.e., without nudging), the simulation can become unstable within a few hours. Herein the nudging coefficients used for wind and temperature were 2.5 × 1.0−4 in the outer domain and 1.0 × 1.0−4 in the inner domain. For moisture nudging a uniform value of 1.0 × 1.0−5 was used in all the domains. For details see NCAR [2003].

[24] The period 26–28 June 2002 corresponds to one of the Intensive Observation periods (IOP) of ARMEX when very heavy precipitation occurred over the west coast of India. As mentioned earlier, special observations from surface (land, ship, and buoys), upper air stations, radar, aircraft and satellites were collected during the ARMEX period. Figure 11 illustrates the observation network over the west coast and the adjoining Arabian Sea. All observations (SYNOP, SHIP, BUOY, TEMP, PILOT, AIREP, SATEM, SATOB, ATOVS, SSMI) were passed through quality control checks and reanalysis was carried out at the resolution of the MM5 model.

Figure 11.

Observations network during the Arabian Sea Monsoon Experiment (ARMEX), June–August, 2002.

[25] Figure 12 illustrates the 850 hPa wind field analysis obtained by using the FDDA with additional observations collected during the IOP and the control analysis without the ARMEX observations. The control analysis contained only the routine surface and upper-air observations including AIREP, SATEM, SATOB, ATOVS, and SSMI. The wind field shown in the Figure 12a is a product of 12-hourly cyclic nudging using ARMEX observations starting from 26 June and ending on 28 June using forecast fields as the first guess for the next cycle. The location of the vortex is improved in the reanalysis as compared to the control [(Figure 12b)]. The vortex is also better organized with stronger intensity and wind shear in the southwest sector. The position of the cyclonic circulation in Figure 12a is closer to the observed cloud clusters as shown in Figure 3a. This clearly indicates that the initial conditions of the model were improved by including the ARMEX observations.

Figure 12.

Wind at 850 hPa on 00 UTC 28 June 2002 (a) with FDDA and (b) without FDDA. Terrain height above 850 hPa are shaded.

[26] In one of the experiments, MM5 was nudged for 24 h from 00 UTC of 26 June to 27 June and simulations were carried out for the next 24 h valid for 00 UTC of 28 June. The 24-h rainfall forecast based upon the improved analysis is shown in Figure 13a. The location of the maximum rainfall was improved by nudging, but the amounts are still much less than observed. The improved location of the maximum rainfall is due to the improved position of the vortex as shown earlier. Improved locations of maximum rainfall forecasts by the analysis-nudging method were also shown by Rao and Prasad [2005] but, they too, could not accurately simulate rainfall intensity.

Figure 13.

Simulated rainfall on 00UTC 28 June 2002 using FDDA starting from 00 UTC of (a) 27th, (b) 24th, (c) 21st and (d) 18th June 2002. Rainfall between 1 to 5 mm are shaded.

[27] In a ‘cold-start’, the model fields are not in a suitably balanced state, leading to an initial shock which usually degrades a forecast [Daley, 1991; Fast, 1995]. The nudging procedure resembles intermittent assimilation where data is inserted at specified times (usually six hourly). The subsequent analysis normally shows imbalances between the mass and momentum fields, which can lead to the generation of spurious high-frequency gravity waves in the early stages of the model integration.

[28] In tropical regions it is particularly important to operate the nudging cycle long enough to achieve realistic circulations since the large-scale balance is weaker than in the midlatitudes: data sparseness, weaker large-scale balances, etc., increase the model spin-up time. Experience of real-time operational forecasting in the tropics indicate that for a cold start a model usually takes about two weeks to achieve balance. We carried out six hourly cyclic nudging, where the first guess was obtained through a previous forecast from the mesoscale model followed by a reanalysis with observations. Investigations were carried out to examine whether nudging should be carried out for 24, 48, 72 h, etc. prior to the event being studied, and whether the nudging should be carried in one computational domain or in all three domains. In order to improve the prediction of the rainfall intensity, we performed experiments where nudging began 10 days prior to the heavy rainfall event (i.e., 18th June). The best results were obtained when all the three domains were nudged for a period of about 10 days prior to the event. The diagram clearly indicates that both intensity and the location of the maximum rainfall are improved, because the model atmosphere is guided by observations.

[29] Figure 14 shows the time series of vertical cross section of vertical velocity and rainfall during the period 00 UTC of 18–28 June 2002 with continuous nudging. The values are averaged over the area 68°–75°E and 16°–25°N, where maximum cloudiness and intense rainfall were observed. Episodes of strong and weak upward motions occurred, particularly in the upper levels [(Figure 14a)]. The lower levels had weak downward motion except during 20–21 June and 26–28 June when ascent occurred throughout the troposphere. Observations show that both periods featured very heavy precipitation at several locations over the west coast. Higher rainfall amounts were reported during the later period. Figure 14b shows the time series of simulated rainfall averaged over the same domain. The rainfall time series shows that MM5 forecasted with reasonable skill the periods of suppressed convection (less rainfall) during 18–20 June and 22–25 June, as well as periods of intense convection (heavy rainfall) during 21–22 and 26–28 June. However, there was a time lag of about 12 h between the observed and simulated peaks.

Figure 14.

Time series of (a) vertical profile of vertical velocity (cm sec−1), (b) rainfall rate (mm day−1) averaged over the domain (68–75° longitude and 16–25° latitude) during 18–28 June 2002 obtained using FDDA. Negative values are shaded in (a).

5. Cloud-System-Resolving Simulations

[30] In the previous section an improved analysis resulted from nudging the model solution toward ARMEX observations over a 10-day period. An accurate initial condition with dynamically balanced atmospheric state is crucial for cloud-system-resolving simulations. The cloud characteristics in Figure 3a and the observed rainfall distribution [(Figure 4)] indicate that during 27–28 June, precipitation was mainly from grid-scale convection (i.e., under-resolved explicit dynamics) rather than sub-grid-scale convective parameterization. Under-resolved explicit dynamics and the relationship of under-resolved dynamics to convective parameterizations is an emerging issue in high-resolution prediction models in convective conditions [Moncrieff et al., 2005; Moncrieff and Liu, 2006]. The relationship between grid-scale precipitation and parameterized convective precipitation is not straight-forward even at grid-resolutions of a few kilometers. Such classification may be based on the intensity of rainfall, the fraction of grid box covered by clouds, intensity of radar echo, etc. Short et al. [1997] classified the rainfall as convective when the intensity was greater than 22 mm h−1 (40 dBZ) based on a simple convective/stratiform texture algorithm using radar echoes in TOGA-COARE. The observed rainfall distribution by TRMM [(Figure 6)] and rain gauges [(Figure 4)] in the study herein indicate pockets of very high rainfall intensity (exceeding 22 mm h−1). However, a large portion of the area shows rainfall intensity less than this value, indicating the presence of explicit grid-scale circulations.

[31] In order to study the macro-dynamics and cloud-microphysical structures of the convective system, MM5 was run at a 2 km grid-resolution over a domain of 600 km × 600 km. This is the fourth computational domain shown in Figure 8, approximately 68°–74°E and 19°–24°N, which spans the Gujarat region where heavy rainfall events were observed during ARMEX. The boundary conditions were generated from the reanalyzed fields produced by nudging observations in the coarser domain at 10 km resolution by using the NESTDOWN procedure of MM5. Integrations were carried out for 48 h starting from 00 UTC, 26 June 2002. As the explicit cloud-microphysics is highly interactive with dynamics at 2 km grid resolution, we evaluate the cloud-microphysical properties simulated by different schemes. The four cloud-microphysics schemes used in these simulations are now summarized:

[32] Bulk warm/cold rain with simple ice: This parameterization is described in Dudhia [1989]. Precipitation processes are based on two prognostic equations for cloud water (qc) and rainwater (qr) following Hsie et al. [1984]. Snow occurs for a temperature below freezing (T < 0°C). Super-cooled water is not permitted and snow is assumed to melt instantaneously below the freezing level.

[33] GSFC: This scheme is based on NASA/Goddard Space Flight Centre (GSFC) cloud microphysics scheme detailed in [Tao et al., 1989, 1993]. It has prognostic equations for two-categories of liquid water (cloud water and rain) and three-categories of ice (cloud ice, snow, and hail/graupel) based on Lin et al. [1983] and Rutledge and Hobbs [1984].

[34] Reisner: This parameterization includes equations for cloud water, rainwater, snow (qs), ice (qi) and graupel (qg) mixing ratios and predicts number concentrations for ice, snow and graupel. It includes mixed phases (i.e., cloud ice, super-cooled water, rain, and snow) and melting snow, evaporation of melting snow and the heterogeneous freezing of cloud water. The explicit prediction of mixed-phase and super-cooled liquid water is important for accurate delineation of the regions of icing potential for aviation community. As for GSFC this scheme is suitable for cloud system-resolving models. The scheme is detailed in Reisner et al. [1998].

[35] Schultz: As for Reisner, this parameterization predicts cloud water, rainwater, pristine ice crystals, snow and graupel (sleet or hail) but aggregation of snow crystals is not included. The virtue of this scheme is computational efficiency, important for NWP applications and four-dimensional data assimilation, because of the few parameters and the minimal number of floating point operations. Details can be found in the work of Schultz [1995].

[36] Figure 15 shows the simulated vertical profiles of cloud liquid water (CLW), rainwater (RNW), ice, snow, graupel (GRAP) and radiative heating tendency (RadT) using SICE, GSFC, RSNR and SULZ cloud-microphysics parameterizations. These fields are averaged over the area of the fourth domain. There is a maximum of CLW in the lower troposphere at about 2 km above the ground and a secondary maximum in the middle troposphere about 8 km. The third maximum at 14 km for the SICE scheme is due to the presence of ice — this scheme treats CLW and ice-mixing ratios by the same equation and differentiates the two species based only on the temperature. The maximum value of the CLW of about 10 mg m−3 is in the lower troposphere.

Figure 15.

Time mean vertical profiles of (a) cloud liquid water, CLW (b) rainwater, RNW (c) ICE (d) SNOW (e) graupel, GRAP and (f) radiative heating tendency averaged over the domain (68°–75°E and 16°–25°N) at 2 km resolution obtained using different cloud microphysics parameterization schemes.

[37] The observed maximum values of CLW measured by TRMM shown in Figure 7 are higher, about 5–10 mg m−3. The TRMM measurements are instantaneous and depend on the time and location of the satellite over-flight and may miss the pockets of observed intense rainfall shown in the Figure 4. Moreover, the TRMM amounts are retrieved using radiation algorithms, and should be considered approximate and dependent on the retrieval assumptions.

[38] The simulated hydrometeor profiles shown in Figure 15 match very well with the domain averaged time mean observed profiles shown in Figure 7. Stano et al. [2002] described the hydrometeor structure of a composite monsoon depression as seen by the TRMM PR radar but did not show the magnitudes of different hydrometeor species.

[39] The simulated profiles are similar to those found in TOGA-COARE and GATE studies by Das et al. [1998, 1999] and Grabowski et al. [1996] using single-column models and cloud-system resolving models. Their profiles had only one maximum value in the middle troposphere, unlike herein where the mid-tropospheric maxima may contain super-cooled water in a mixed phase. All four schemes produce CLW of similar magnitude except the GSFC scheme has the highest value.

[40] The time mean, domain averaged vertical profiles of RNW obtained from different schemes are shown in Figure 15b. Except for the SICE parameterization, the profiles have similar magnitude (about 2–4 mg m−3 from 1–8 km above the ground). The SICE scheme has a maximum value of about 8 mg m−3 at about 12-km altitude. The RNW is the snowfield at T < 0°C for SICE and has no graupel. The snow falls unrealistically slowly and that may be reason for large rain fields at those levels since it is snow, not rain. The profiles of ice mixing ratios obtained from different schemes are shown in Figure 15c. The simple ice scheme has only two species of hydrometeors so these values are not shown. The ice content is maximum (about 6 mg m−3) in the upper-troposphere around 14 km followed by snow having a maximum (about 2.5 mg m−3) at a slightly lower level around 12 km below which graupel is present having a maximum value (about 5 mg m−3) around 9 km altitude. The RNW and snow observed by TRMM appear to be higher in amount than the simulated values by an order of magnitude. Similarly, the observed ice is higher by about 5–10 mg m−3 compared to the simulated values. Plausible reasons for these differences were described earlier. Comparison of different hydrometeors shows that the GSFC scheme produces the highest values of these parameters, compared to the RSNR and SULZ schemes.

[41] Figure 15f shows the averaged profile of radiative tendencies obtained from the different parameterizations. All four produce cooling of about 2–3 K day−1 throughout the troposphere with a maximum at about 850 hPa. Warming is produced in the upper-troposphere with a maximum of about 2 K day−1 around 200 hPa. Significant differences in radiative cooling/warming exist above 12 km between the four schemes. Lower-level cooling and upper-level warming by radiative heating is similar to those obtained in GATE convection by Xu and Randall [1996]. Comparisons of the profiles associated with different micro-physical parameterizations do not feature significant differences.

[42] Simulations of precipitation values discussed earlier indicated that none of the combinations of convection and cloud-microphysics parameterizations produced the observed distribution of rainfall. Accurate values of rainfall were not simulated even by increasing the resolution from 30 to 10 km, unless nudging was applied. All four cloud schemes severely underestimated the rainfall by about 3–4 times compared to observations. Since higher moisture content inside the clouds are conducive for larger rainfall production, it is desirable that the physical parameterizations, particularly the cloud–microphysics, enable better hydrometeor profiles to improve rainfall distribution.

[43] The hydrometeor profiles discussed earlier indicated that the GSFC scheme produced maximum values with better vertical distribution compared to the others. The GSFC scheme has prognostic equations for all the cloud hydrometeors and has been used in diverse environments. Results agree well with remote (radar and passive microwave) and in situ measurements from aircraft [Simpson et al., 1988, 1996; Adler et al., 1991; Prasad et al., 1995]. Therefore henceforth we describe convective system characteristics of simulations using the GSFC scheme.

5.1. Analysis of Simulations Using the GSFC Micro-Physics Parameterization

[44] Figure 16 shows maximum condensate in the upper troposphere (about 100 mg kg−1). The maximum values occurred in the early hours of 27 June. The condensate had a secondary maximum in the lower troposphere in the early hours of 28 June when maximum precipitation was reported at many locations over the region. The maximum value of the condensate in the upper troposphere implies dominance of ice, snow and graupel, which indicates deep clouds with anvils during the depression. Higher values of the condensate in the upper troposphere are associated with strong upward motion.

Figure 16.

Time series of the vertical profiles of the total condensate (CLW + RNW + ICE + SNOW + GRAP) and vertical velocity averaged over the domain at 2 km resolution obtained using the GSFC cloud scheme. Shaded values are the total condensate (mg kg−1) and contours are the vertical velocity (cm s−1).

[45] We analyzed the simulated reflectivity (REF), outgoing long wave radiation (OLR), integrated cloud liquid water (CLW) and precipitable water (PWAT) at 00UTC of 28 June. Some of the key results are presented in Figure 17. Intense reflectivity occurred east–west between 19°N–21°N, corresponding to the sharp cloud boundary seen in the southern sector from the METEOSAT cloud image in Figure 3a. Though the cloud image shows the entire domain filled by clouds, the distributions of reflectivity and cloud liquid water showed gaps at cloud-scale in the simulation.

Figure 17.

Simulated (a) composite reflectivity (b) outgoing long-wave radiation (c) cloud liquid water path and (d) precipitable water at 00 UTC, 28 June 2002 using GSFC cloud scheme at 2 km horizontal resolution.

[46] The OLR is a measure of the intensity of convection and depends upon temperature, cloud amount and height. The simulated OLR distribution showed minimum values (∼150 w m−2) over the region of intense echoes. This agrees reasonably with the observed OLR values over the region [(Figure 5)]. Higher values of OLR (∼250 w m−2) are simulated over the cloud free regions. Similarly, the distribution of precipitable water and the cloud liquid water show that higher values are simulated by the model when intense echoes are present. The order of magnitude of these quantities compares fairly well with the observed values from Multichannel Scanning Microwave Radiometer (MSMR) aboard the Indian Remote Sensing – Programme 4 (IRS-P4) Satellite, obtained by Das et al. [2004] over the Indian region during an active monsoon season. The simulated precipitable water also compares reasonably well with the observed values during this period.

5.2. Dynamics of the Convective System

[47] We have discussed so far the microphysical characteristics of the convective system. The macro-dynamics of such systems depend upon whether the system is a cold core or a warm core, wind shear distribution, latent heating, CAPE, etc. Miller and Keshavamurthy [1968] suggested that the west coast rainfall is associated with offshore trough and MTC embedded with mesoscale bands, which are linked to the movement of a disturbance from the Bay of Bengal similar to the present case [(Figure 1)]. Ramage [1966] suggested that the heat low over northwest India plays an important role in the formation of MTC by transporting cyclonic vorticity into the region. Numerical simulations by Krishnamurti and Hawkins [1970] and Carr [1977] suggested that latent heating plays an important role in the intensification and maintenance of the MTC. Benson and Rao [1987] showed that several convective bands were embedded in the synoptic scale cloud cluster over the Arabian Sea in a heavy rainfall case of 20 June 1979 over the west coast. The bands form and decay as a result of complex interactions between the upper-level westerly flow, and the low at the surface and the mesoscale convective features. De and Dutta [2006] calculated that the sudden rise in the west coast rainfall is associated with a marked rise in CAPE of about 850–1500 J kg−1. Rao [1976] found that heavy rainfalls over the west coast of India are associated with a quasi-stationary trough, in which sometimes vortices are embedded.

[48] In order to investigate whether the system was warm-core or cold-core, vertical cross sections of temperature anomaly at 20°N, along 69°E–73.5°E were diagnosed from the 48 h simulations. The cross-section was taken along the line where intense echoes were simulated corresponding to the sharp cloud boundary seen from the METEOSAT cloud image in Figure 3a. The temperature anomaly was calculated from the domain averaged value at each level. Figures 18a and 18b illustrate the vertical cross sections of the temperature anomaly for 00 UTC of 27 and 28 June. The diagram shows a layer of warm temperature anomaly of about 1 degree between 400 hPa to 700 hPa, indicating that the well marked low was a warm core convective system. Study shows that the warm anomalies are usually located over the regions of higher precipitation. The core was not particularly warm as the mesoscale convective system formed by merger of a middle latitude westerly trough and a tropical monsoon low as mentioned earlier. The monsoon depression of a purely tropical origin has a much warmer core aloft.

Figure 18.

Vertical cross-section of temperature anomaly and zonal wind across the heavy rainfall band along 20°N simulated at 00UTC of 27th and 28th June 2002. Contours are drawn at −1, 0, 0.5, 1 K for temperature and −20, −15,. 0, 2, 5, 10, 12 m s−1 for the wind anomalies.

[49] The analysis of zonal wind shear during 27 and 28 June are shown in Figures 18c and 18d. The diagrams indicate strong westerlies from surface up to 200 hPa and easterlies above that level. The strength of the wind shear increased with time and reached a maximum value around 73°E–74°E on 28 June. The heavy precipitation band occurred on the southern flank of the depression with deep westerlies in that sector. Wind shear produces instability in the flow regime and, helps the convective system to grow further. Ogura and Yoshizaki [1988] simulated the orographic convective precipitation near the coast by using a cloud-resolving type model incorporating interaction of the low level flow with orography. They concluded that wind shear was essential for realistic simulation of the rainfall distribution. The strength of upper-tropospheric wind shear also increases the severity of the convective system since it differentiates the updraft from downdraft, especially when the system tilts backwards relative to the direction of propagation [Moncrieff, 1992] as in jet-like flow. It may also extend the life time of a convective cell [Raodcap and Rao, 1993; Miller and Keshavamurthy, 1968]. The vertical shear may also lead to shearing of the anvil cloud in the high energy westerly regime [Benson and Rao, 1987]. Mak [1975] attributed the genesis of baroclinic instability of a basic state to meridional as well as zonal shear. Rao and Hor [1991] studied the momentum and kinetic energy exchange associated with an offshore vortex over the west coast. They found that vertical shear and buoyancy were responsible for the changes in momentum flux and there was a net transfer of kinetic energy from the large scale to the band-scale motion.

6. Summary

[50] We examined the dynamics and cloud-microphysical properties of a deep convective system associated with a depression that formed over the west coast of India due to interaction between an extratropical westerly trough and a tropical easterly monsoon low. The two systems merged into a well marked low producing widespread heavy rainfall ranging from 2–61 cm day−1 over the west coast of India during 26–28 June 2002. Multiscale simulations were carried out using the MM5 model on a triple-nested domain at 90, 30 and 10 km resolutions. Simulations were carried out using different combinations of physical parameterization schemes and nested domains to match the rainfall with observations.

[51] While none of the experiments produced the observed distribution of rainfall, the combination of GRL + GSFC + MYA + CCM2 + MSM schemes produced relatively better results. Compared to observations, the precipitation was underestimated by almost a factor of 10. Further experiments operating four dimensional data assimilation (nudging) ingested measurements collected during the Arabian Sea Monsoon Experiment (ARMEX). We examined the impact of the duration of the nudging period on the rainfall forecasts in the multiple domains. Unless nudging is carried out in all domains for a sufficiently long time to achieve dynamical balance, the simulated wind fields and rainfall distributions do not match with observations. The best results were obtained when the nudging was carried out continuously for a period of about 10 days. Continuous nudging of observations produced better analysis, a balanced atmosphere and reduced spin-up effects. Thereby, accurate simulations of the location of the vortex, stronger low-level jet and better distribution of rainfall were achieved. Time series of the vertical velocity and rainfall showed that MM5 simulated the episodes of suppressed and enhanced convective activities accurately when the FDDA was operated for a sufficiently long period, i.e., 10 days herein. The nudging is required for a longer period to achieve proper circulations in tropical regions where the large-scale balance is weaker than in the midlatitudes. These results emphasize the importance of mesoscale data assimilation for real-time prediction of intense precipitating events in the tropics. On a larger perspective, such results support the view that focused field campaigns are required for understanding the intricacies of convective systems in tropics.

[52] The reanalyzed fields with ARMEX observations were used to generate boundary conditions for initializing MM5 in cloud-system-resolving mode (2 km grid-resolution) over the west coast of India in order to study detailed microphysical and dynamical characteristic of the depression. There is no significant difference in the hydrometeor profiles simulated by different cloud microphysics schemes, although the GSFC scheme did produce relatively higher values of the total condensate compared to the other three schemes. The profile of cloud liquid water had a maximum in the lower troposphere and a secondary maximum in the middle troposphere. The maximum values of ice, snow and graupel were in the upper and middle troposphere. Results indicate that the hydrometeor profiles observed by TRMM (2A12 products) are usually higher than the simulated values. The profile of the total condensate showed a maximum value in the upper troposphere implying dominance of ice, snow and graupel, and the presence of deep clouds with anvils during the depression.

[53] The simulated deep precipitating convection resulted in intense echoes (reflectivity) and low OLR matching the cloud bands in the satellite cloud imagery. The simulated cloud fractions indicated the presence of multilayers of high, medium and low clouds along the observed rainfall band and only low and middle clouds over rest of the domain. The depression was a warm core system having a temperature anomaly of about 0.5 to 1 K around 500–600 hPa. The southern flank of the convective system was associated with deep strong westerlies up to about 300 hPa and increased wind shear led to strong moisture convergence and heavy precipitation.

[54] Figure 19 conceptualizes the life-cycle of the intense system based upon the present study in terms of a system-relative circulation, which involves interaction between midlatitude westerlies and tropical easterlies in three stages. In the first stage, a westerly trough at upper level (500 hPa) and a monsoon low at the surface approach each other along the monsoon trough [(Figure 19a)]. In the second stage, the two systems merge and produce deep convective clouds carrying moisture from the Arabian Sea and the Bay of Bengal. Heavy precipitation occurs at the location of merger [(Figure 19b)]. Heavy precipitation is caused by latent heating, increased CAPE, resultant wind shear distribution, exchange of momentum and kinetic energy between the upper upper-level westerly flow, and the low at the surface. In the third stage [(Figure 19c)] the westerly trough moves away and weakens over the Tibetan plateau. The monsoon low weakens over the deserts of India/Pakistan, where the source of moisture is cutoff and the system decays.

Figure 19.

Schematic diagram indicating the life cycle of the system in three stages.


[55] Results reported in this paper are based on the collaborative works under a MOU signed between the National Center for Medium Range Weather Forecasting, of the Department of Science & Technology, Govt. of India and the National Center for Atmospheric Research, sponsored by the National Science Foundation, USA. Thanks are due to many colleagues who provided help during this study, in particular, Bob Gall (NCAR), Cindy Bruyere (NCAR), the late S. V. Singh (NCMRWF), A. K. Das (IMD) D.R. Patnaik (IMD) and S. K. Dutta. Rain gauge observations obtained from IMD is duly acknowledged. We would also like to thank the three anonymous referees whose comments have led to substantial improvement of the manuscript. Simulations carried out in this study were made at NCMRWF and NCAR supercomputers.