The 26 July 2005 heavy rainfall event over Mumbai: numerical modeling aspects
- 4k Downloads
The performance of the Advanced Regional Prediction System (ARPS) in simulating an extreme rainfall event is evaluated, and subsequently the physical mechanisms leading to its initiation and sustenance are explored. As a case study, the heavy precipitation event that led to 65 cm of rainfall accumulation in a span of around 6 h (1430 LT–2030 LT) over Santacruz (Mumbai, India), on 26 July, 2005, is selected. Three sets of numerical experiments have been conducted. The first set of experiments (EXP1) consisted of a four-member ensemble, and was carried out in an idealized mode with a model grid spacing of 1 km. In spite of the idealized framework, signatures of heavy rainfall were seen in two of the ensemble members. The second set (EXP2) consisted of a five-member ensemble, with a four-level one-way nested integration and grid spacing of 54, 18, 6 and 1 km. The model was able to simulate a realistic spatial structure with the 54, 18, and 6 km grids; however, with the 1 km grid, the simulations were dominated by the prescribed boundary conditions. The third and final set of experiments (EXP3) consisted of a five-member ensemble, with a four-level one-way nesting and grid spacing of 54, 18, 6, and 2 km. The Scaled Lagged Average Forecasting (SLAF) methodology was employed to construct the ensemble members. The model simulations in this case were closer to observations, as compared to EXP2. Specifically, among all experiments, the timing of maximum rainfall, the abrupt increase in rainfall intensities, which was a major feature of this event, and the rainfall intensities simulated in EXP3 (at 6 km resolution) were closest to observations. Analysis of the physical mechanisms causing the initiation and sustenance of the event reveals some interesting aspects. Deep convection was found to be initiated by mid-tropospheric convergence that extended to lower levels during the later stage. In addition, there was a high negative vertical gradient of equivalent potential temperature suggesting strong atmospheric instability prior to and during the occurrence of the event. Finally, the presence of a conducive vertical wind shear in the lower and mid-troposphere is thought to be one of the major factors influencing the longevity of the event.
During the Indian monsoon season (June through September), especially in the months of July and August, many locations along the west coast of India (towards the windward side of the Western Ghats) receive heavy rainfall. The strong moisture-laden westerlies from the Arabian Sea interact with the topography, causing heavy precipitation over this region. The other important factors leading to such intense rainfall events are the mid-tropospheric cyclones, and organized convection in the tropical convergence zone (e.g., Krishnamurti and Hawkins 1970; Benson and Rao 1987; Ogura and Yoshizaki 1988). In the past, heavy rainfall of the order of 50 cm/day has been reported at various locations along the Indian west coast (Dhar and Nandargi 1998; Kulkarni et al. 1998).
In a previous study, Doswell et al. 1996 emphasized the importance of ingredients-based methodology for flash flood forecasting. They suggested that all flash flood events have some basic ingredients in common, whose presence can be used as potential indicators of heavy precipitation. They reported that heavy precipitation rates involve high values of vertical moisture flux, associated with high vertical velocity and substantial amount of water vapor contained in the ascending air. The other important factor responsible for heavy rainfall rates is the precipitation efficiency, which in turn is a function of the relative humidity of the environment, and several other factors like wind shear. Another relevant study was by Romero et al. 1998, wherein they attempted to simulate three heavy precipitation events in the western Mediterranean region, using a mesoscale model. One of the important conclusions was that the forecasts were reasonably good for two cases in which the synoptic-scale activity was prominent; however, the model prediction was not satisfactory for the third case, which was an isolated convective storm strongly influenced by local effects. Some studies in the past have also highlighted the role of vorticity-stretching in the occurrence of heavy precipitation (e.g., Chakraborty et al. 2006).
Most of the operational numerical models failed to predict this extreme event. In one of the previous studies, Vaidya and Kulkarni (2007) reported that their model failed to simulate the 38.1 cm of rainfall that occurred during 0900–1200 UTC, 26 July 2005, due to the coarse grid spacing (40 km). In another modeling study, Deb et al. (2008) investigated the impact of the Tropical Rainfall Measurement Mission (TRMM) Microwave Imager (TMI) sea surface temperature (SST) on the simulation of the heavy rainfall episode over Mumbai, using two different mesoscale models. They found that the intensity of maximum rainfall around Mumbai was significantly improved with TMI SST as the surface boundary condition in both the models. Using the Weather Research and Forecasting (WRF) model, Kumar et al. (2008) reported that an interplay of various factors like the low-pressure area over the Bay, strong moisture convergence, and meridional temperature gradient might have contributed to an event of such magnitude.
A major feature of this event was that around 65 cm of rainfall was accumulated in a span of 6 h, which amounts to 70% of the total rainfall being accumulated in just 25% of the time, hinting at the strong intensities of rainfall that might have occurred during the period. In the present work we explore the ability of the Advanced Regional Prediction System (ARPS) in simulating the event, and subsequently investigate some of the important physical mechanisms involved. We start with an idealized model integration, and then conduct further simulations in a realistic framework using two kinds of ensemble techniques, namely, the standard method for ensemble construction (model integrations starting at different times) and the Scaled Lagged Average Forecasting (SLAF) technique. We evaluate the performance of the model, based on three attributes, namely, location, timing and intensity of the simulated rainfall maximum, and then choose the best member from each of the two realistic sets of experiments, and analyze some of the physical aspects of the event.
The following is the structure of the paper: details of the modeling strategy and the data sets used are presented in Sect. 2. In Sect. 3, we present the main results from this work, followed by conclusions in Sect. 4.
2 Data and modeling details
2 km: explicit convection; initial and boundary conditions from 6 km explicit convection.
2 km: explicit convection; initial and boundary conditions from 6 km implicit convection.
2 km: implicit convection; initial and boundary conditions from 6 km implicit convection.
3 Results and discussion
3.1 Idealized set of experiments
3.2 Experiments using the standard ensemble technique
3.3 Experiments using the SLAF technique
Finally, we analyze the model output using the 2 km grid spacing. It is to be noted that a grid spacing of 2 km was chosen as opposed to 1 km (finest grid for EXP2) because at 1 km the model was unable to simulate the spatial structure of the event, probably due to the strong influence of the boundary conditions obtained from the 6 km model. The two model simulations with explicit convection show that the rainfall follows the topography of the Western Ghats mountain range and there is no localized heavy rainfall around the region of interest (figure not shown). For the third set with the 2 km grid spacing, it was seen that the use of the cumulus scheme leads to a smearing of rainfall to other locations (figure not shown). This suggests that even with a 2 km grid spacing the model was not able to produce a realistic spatial distribution of rainfall accumulation. It is to be noted here that the inability of the model to produce a satisfactory spatial structure at higher resolutions may be specific to this event, and hence should not be treated as a generic conclusion.
3.4 Analysis of the circulation and thermodynamic features
We first choose the “best” member from the two ensembles based on the spatial distribution of rainfall and the timing of the maximum intensity. Based on these criteria, M3 from EXP2 (Fig. 7c, Fig. 8) and M2 from EXP3 (Fig. 10b, Fig. 12) were chosen for further analysis.
The purpose of this paper is to evaluate the performance of the ARPS model in simulating the spatio-temporal characteristics of a rainfall event of unprecedented intensity, and subsequently gain insights into the initiation and sustenance of that event. In this regard, three sets of ensemble simulations were conducted using ARPS, with three different experiment setups. The Mumbai rainfall event of 26 July 2005 that led to a record 65 cm of rainfall accumulation in a span of just 6 h was chosen as a case study. Three parameters, namely, location, timing, and intensity were chosen to evaluate the model performance.
Even in an idealized framework (with horizontally homogeneous initial conditions) signatures of localized heavy rain were present in the model simulations. For a more detailed analysis, two more sets of ensemble experiments were conducted, one using the standard ensemble technique (EXP2), and the other using SLAF (EXP3) for ensemble construction. It was seen that the time of rainfall peak was well simulated by the model in both sets of ensemble simulations. In addition to the simulated timing of maximum rainfall intensity, it was also found that the model simulations using SLAF were able to simulate the abrupt increase in rainfall intensity that was seen from the observed rainfall time series. The other parameter used for the evaluation of model performance was the intensity of rainfall and the 6-hourly accumulation. In this case the model simulations using SLAF were found to be superior to those using the standard ensemble technique. The location of rainfall peak simulated by the model had spatial shifts (compared to observations) in almost all the simulations. This shift is apparently because of two reasons: (1) deficiency of the moist physics (cumulus scheme and cloud microphysics), and (2) errors present in the initial and boundary conditions used to integrate the model. Furthermore, in regard to the use of convection scheme at higher spatial resolutions, we found from this study that the use of cumulus scheme at finer resolutions of around 6 km or less appears to simulate a more spatially homogeneous distribution of rainfall.
We also investigated the role of various other diagnostics important in the initiation and sustenance of this particular rain event. To start with, the circulation and thermodynamic features were analyzed by choosing the “best” member from ensemble set of EXP2 and EXP3. From our analysis, it was concluded that all the necessary ingredients needed for a severe thunderstorm were present prior to and during the occurrence of the event. Specifically, there was strong convective instability in the atmosphere, accompanied by high values of specific humidity, and a conducive vertical wind shear. There was a high negative gradient of equivalent potential temperature in the vertical, prior to the onset of the event, thus confirming the presence of strong convective instability present in the atmosphere triggering deep convection. Strong convergence of moist air initiated at the mid-troposphere (~8 km) and gradually propagated to the lower altitudes. From the analysis of specific humidity, it was seen that a positive feedback cycle formed by the moisture convergence, rainfall and its associated latent heat release added to the severity of the event.
The most striking feature of this thunderstorm was its sustenance for a long period of time, since one would intuitively expect an event of such intensity to be short lived. From the analysis of vertical wind shear, it was found that a strong vertical shear existed, both in magnitude as well as direction of the wind, which was one of the primary reasons leading to the longer lifetime of the thunderstorm. The vertical profile of hydrometeors revealed that the model-simulated cloud-top heights were around 16–17 km and in very good agreement with the observed value (~17 km) over the location of maximum rainfall.
The authors would like to thank Fanyou Kong of the Center for Analysis and Prediction of Storms, University of Oklahoma and S.S. Vaidya of the Indian Institute of Tropical Meteorology for their useful suggestions and help during the course of this work. The use of Advanced Regional Prediction System (ARPS) developed by the Center for Analysis and Prediction of Storms, University of Oklahoma, and rainfall data from the India Meteorological Department is thankfully acknowledged. Finally, we would like to thank the two anonymous reviewers for their useful suggestions which helped improve the manuscript.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- Chakraborty A, Mujumdar M, Behera SK, Ohba R, Yamagata T (2006) A cyclone over Saudi Arabia on 5 January 2002: a case study. Meteorol Atmos Phys 93:155–122Google Scholar
- Dhar ON, Nandargi SS (1998) Rainfall magnitudes that have not exceeded in India. Weather 53:145–151Google Scholar
- Droegemeier, KK, Bassett G, Xue M (1994) Very high-resolution, uniform-grid simulations of deep convection on a massively parallel processor: Implications for small-scale predictability. Preprints, 10th Conference on Numerical Weather Prediction, Portland, OR, Am Meteor Soc, 376–379Google Scholar
- Huffman GJ, Adler RF, Bolvin DT, Gu G, Nelkin EJ, Bowman KP, Hong Y, Stocker EF, Wolff DB (2007) The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J Hydrometeorol 8:38–55. doi:10.1175/JHM560.1 CrossRefGoogle Scholar
- Jenamani RK, Bhan SC, Kalsi SR (2006) Observational/forecasting aspects of the meteorological event that caused a record highest rainfall in Mumbai. Curr Sci 90:1344–1362Google Scholar
- Kalnay E (2003) Atmospheric Modeling, Data Assimilation and Predictability. Cambridge Press, Cambridge 341 ppGoogle Scholar
- Kong F, Droegemeier KK, Venugopal V, Foufoula-Georgiou E (2004) Application of scale-recursive estimation to ensemble forecasts: a comparison of coarse and fine resolution simulations of a deep convective storm. 20th conference on weather analysis and forecasting and 16th conference on numerical weather predication, Seattle, WA, Am Meteor Soc (Preprint) Google Scholar
- Kulkarni AK, Mandal BN, Sangam RB (1998) Some facts about heavy rainfall distribution over the Konkan region. Proc Workshop on Scientific Aspects of Water Management, April 1998, 12–21Google Scholar
- Xue M, Droegemeier KK, Wong V, Shapiro A, Brewster K, Carr F, Weber D, Liu Y, Wang D (2001) The advanced regional prediction system (ARPS)—a multiscale non-hydrostatic atmospheric simulation and prediction tool. Part II: Model physics and application. Meteorol Atmos Phys 76:143–165CrossRefGoogle Scholar