Abstract

In this paper, the three-dimensional variational data assimilation scheme (3DVAR) in the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (Penn State–NCAR) Mesoscale Model (MM5) is used to study the impact of assimilating Atmospheric Infrared Sounder (AIRS) retrieved temperature and moisture profiles on board Aqua, a satellite that is part of NASA’s Earth Observing System. A record-breaking heavy rain event that occurred over Mumbai, India, on 26 July 2005 with 24-h rainfall exceeding 94 cm was used for the simulation.

By analyzing the data from the NCEP–NCAR reanalysis, possible causes of this heavy rainfall event were investigated. The temporal evolution of meteorological fields clearly indicates the formation of midtropospheric mesoscale vortices over Mumbai that exactly coincides with the duration of the intense rainfall. Analysis also indicated the midlevel dryness with higher temperature and moisture in the lower levels. This midlevel dryness with high temperature and moisture in the lower levels increases the conditional instability, which was conducive for the development of very severe local thunderstorms. The midtropospheric mesoscale vortices existed over Mumbai together with lower-level instability and the active monsoon conditions over the west coast resulted in intense rainfall, on the order of 94 cm in 24 h.

Numerical experiments were conducted, with two nested domains (45- and 15-km grid spacing). The assimilation of the AIRS-retrieved temperature and moisture profiles produced significant impacts on the location and intensity of the simulated rainfall. It is seen from the numerical experiments that the assimilation of AIRS data could produce the structure of mesoscale vortices, and lower-level thermodynamics and convergence much more realistically compared with the control simulation. The spatial distribution of the rainfall from the simulation using AIRS data was more realistic than that without AIRS data. To make the quantitative comparison of the predicted rainfall with the observed one, the equitable threat score and bias were calculated for different threshold values of rainfall. Inclusion of AIRS data significantly improved the precipitation as indicated by the equitable threat scores and biases for almost all of the threshold rainfall categories.

1. Introduction

Mumbai (19.0°N, 72.85°E), which was previously known as Bombay, is a major metropolitan city in India on the country’s west coast. For most of the year, Mumbai’s climate is warm and humid. Between November and February, the skies are clear, and the temperature is low. Starting in March the atmosphere becomes warm and humid till mid-June, the time of the beginning of the summer monsoon. During the summer monsoon there are torrential rains, sometimes causing the flooding of major roads and streets of Mumbai. The monsoon season [June–September (JJAS)] average rainfall for Mumbai is 180 cm.

Heavy rainfall events frequently occur in Mumbai during the summer monsoon. However, parts of Mumbai experienced an unprecedented rainstorm on 26 and 27 July 2005. The 24-h rainfall ending at 0300 UTC 27 July was 94.4 cm as per the meteorological observatory at Santa Cruz Airport (19.15°N, 72.84°E; hereafter Santa Cruz), Mumbai. On the other hand, Colaba (18.90°N, 72.81°E) the other meteorological observatory, which is about 27 km away, recorded only 7.4 cm of rain during the same period. There were reports of even heavier rainfall of 104.5 cm near Vihar Lake (19.20°N, 72.91°E) within Mumbai. This torrential rain disrupted life in the metropolis and caused a large number of deaths. According to early estimates (as reported by the media), the disaster resulted in the loss of more than $250 million (U.S. dollars). If information could have been provided, say 2–3 days in advance, it would have helped allow for better preparation by the disaster-management machinery, thus minimizing the loss of life and property. Most, if not all, operational numerical weather prediction (NWP) models were unable to forecast this heavy rainfall event. Bohra et al. (2006) reanalyzed this event using different operational numerical weather prediction (NWP) models (with 90-, 30-, and 10-km grid spacing), and confirmed the inability of the different operational models to reproduce this event satisfactorily. For example, the National Centre for Medium-Range Weather Forecasting’s (NCMRWF) global model simulated only 2 cm of rainfall over Mumbai, with the precipitation hotspot being south of Mumbai with about 4 cm of rain on 0300 UTC July 27. Jenamani et al. (2006) analyzed the different operational analyses and found that the operational Limited Area Model (LAM) and fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5) simulated only 4–7 cm of rainfall in the 24 h ending at 0000 UTC 27 July. Using reanalysis data, the Met Office (UKMO) model reported partial success in reproducing the Mumbai event with rainfall amounts of between 20 and 24 cm. At UKMO, a more advanced four-dimensional variational data assimilation (4DVAR) system is used. Moreover, UKMO uses global high-resolution (∼50 km) grids. This might have enabled the UKMO to capture the event in a slightly better way.

Accurate initial conditions are one of the foremost prerequisites for accurate numerical simulations and predictions. The lack of data over the oceans and other remote regions contributes greatly to the uncertainties in the initial state of the numerical weather prediction models. A continuing difficulty with respect to the improvement of regional weather and climate model simulations relates to the fact that the input observational information is limited and inaccurate, especially in data-sparse areas such as over large oceans.

Data assimilation has been recognized as being a useful way to obtain better “consistent” initial conditions for NWP. Although there are reasonable numbers of conventional observation stations in most inland and populated areas, the number of observations is still not enough to fully characterize complicated weather systems. The temperature and moisture distribution in the lower troposphere, especially over the Arabian Sea, is very crucial to the evolution of severe weather systems over the west coast. Particularly, moisture is more crucial because of its potential to release large amounts of latent heat. Accurate knowledge of the distribution of water vapor held in the atmosphere is indispensable for predicting the amount, the time, and the location of precipitation. Therefore, uncertainties in the initial conditions of the humidity field in numerical weather prediction models could have a significant impact on precipitation forecasts. Unfortunately, surface and upper-air data are sparse in this region (Arabian Sea and also over Indian Ocean). Satellite observations provide valuable measurements for NWP models, particularly in the locations where conventional observations are sparse or inaccessible. Since the advent of meteorological satellites in the 1960s, numerous experiments have been conducted (Gal-Chen et al. 1986; Doyle and Warner 1988; Lipton and Vander Haar 1990; Lipton et al. 1995; McNally and Vesperini 1996; Tomassini et al. 1998; Ruggiero et al. 1999; English et al. 2000; Xiao et al. 2000; Zapotocny et al. 2000; Zapotocny et al. 2002; Zhu et al. 2002; Xiao et al. 2002; Chen et al. 2004; Zavodsky et al. 2004; Zhao et al. 2005; Zapotocny et al. 2005; Chen 2007; Zhang et al. 2007; Singh et al. 2008) to evaluate the impact of these data on atmospheric analysis and prediction.

The Atmospheric Infrared Sounder (AIRS) and its companion Advanced Microwave Sounding Unit (AMSU) were launched into polar orbit on board the National Aeronautics and Space Administration (NASA) Aqua satellite in May 2002. The primary scientific achievements of AIRS have been to improve weather prediction (Le Marshall et al. 2005a–c) and to study the water and energy cycle (Tian et al. 2006). AIRS also provides new measurements of several greenhouse gases, such as CO2, CO, CH4, and O3, and SO2 and aerosols. The measurement goal of AIRS is to retrieve temperature and moisture profiles with accuracies approaching those of conventional radiosonde.

The assimilation of AIRS profiles in the regional and global models has already demonstrated significant improvements. Chou et al. (2006) assimilated AIRS profiles for a Pacific storm case and showed that the AIRS profiles impacted the analysis and resulted in a positive forecast impact on the temperature and moisture. Wu et al. (2006) assimilated the AIRS profiles in a mesoscale model (MM5) to investigate its influence on the formation and track of Hurricane Isabel (2003). They found, by incorporating the AIRS data, that the model can better simulate the large-scale flow patterns and the evolution of Hurricane Isabel in terms of the timing and location of the formation and the subsequent track of the storm. Atlas (2005) assimilated the AIRS temperature and moisture profiles in NASA’s finite-volume general circulation model (FVGCM) with NCEP’s spectral statistical interpolation (SSI) analysis. He showed the forecast’s impact, particularly in the Southern Hemisphere, and an improvement in forecasting the intensities and positions of cyclones.

Instead of AIRS profiles, some of the leading centers assimilate AIRS radiances directly (without temperature and moisture retrieval) in their operational forecasts (Le Marshall et al. 2005a–c; Chahine et al. 2006; Le Marshall et al. 2006; Jung et al. 2006; Garand et al. 2006). These centers include NCEP, the European Centre for Medium-Range Weather Forecasts (ECMWF), and the UKMO. All have reported positive impacts from AIRS radiance assimilation.

The vertical profiles of temperature and humidity from AIRS provide an unprecedented opportunity to examine the effects of atmospheric thermodynamics on the prediction of severe weather systems by allowing the initial state to be closer to the observational state. There are two objectives of the present study. The first objective is to analyze the historical heavy rainfall event of 26–27 July 2005 that took place over Mumbai to identify the meteorological conditions. The second objective is to assess the improvement in the prediction of the present heavy rainfall event due to assimilation of AIRS-retrieved vertical profiles of temperature and moisture in the initial conditions. The forecast experiments are conducted with and without AIRS-retrieved temperature and moisture–humidity profiles in the initial conditions.

The contents of the paper are organized in the following manner. In section 2, we briefly describe the Mumbai heavy rain event and possible causes of the event. The AIRS data used in this study are summarized in section 3. The MM5 model and its 3D variational data assimilation formalism and the experimental design are presented in section 4. Results of numerical simulations are discussed in section 5. The paper is concluded in section 6.

2. Characteristic features of the rainfall event

The spatial distribution of the 24-h (ending at 0300 UTC 27 July 2005) accumulated rainfall as observed by different rain gauge stations of the Indian Meteorological Department (IMD) over Mumbai and its neighborhood is given in Fig. 1. It may be seen from Fig. 1 that the very heavy rainfall event was not a localized one, as the 40-cm rainfall contour line extended to a distance of 50–60 km in the east–west direction. On this day, not only did Santa Cruz Airport (19.15°N, 72.84°E) record 94.4 cm, but in addition Vihar Lake, which lies northeast of Santa Cruz, recorded a higher amount of 104.9 cm. However, the amount decreased thereafter in the northeast sector of Santa Cruz, with Bandhup and Tulsi reporting rainfalls of 81.5 and 60.1 cm, respectively. The stations northeast of Santacruz, for example, Bhiwandi, Thane, and Kalyan, reported 75, 74, and 62 cm of rainfall, respectively. However, the rainfall was reduced significantly to a few centimeters beyond 60 km to the northeast of Santa Cruz at lake stations; for example, Tansa (19.35°N, 73.40°E) and Vaiterna (19.43°N, 73.51°E) reported 5 and 1 cm of rainfall, respectively. South of Santa Cruz, Colaba (18.90°N, 72.81°E) recorded only 7.4 cm of rainfall. From Fig. 1, it is also interesting to note the occurrence of another very high rainfall area lying nearly 50–60 km away to the southeast of Santa Cruz. In this area, Matheran and Karjat (18.96°N, 73.3°E) recorded 84 and 75 cm of rainfall, respectively. Figure 2 shows the temporal distribution of the 3-hourly accumulated rainfalls at Santa Cruz during 26–27 July. Almost 60% of the total rainfall at Santa Cruz occurred during 0900–1500 UTC 26 July.

Fig. 1.

The 24-h (ending at 0300 UTC 27 Jul) accumulated IMD observed rainfall (cm).

Fig. 1.

The 24-h (ending at 0300 UTC 27 Jul) accumulated IMD observed rainfall (cm).

Fig. 2.

Temporal evolution of the past 3-h accumulated rainfall (cm) over Santa Cruz.

Fig. 2.

Temporal evolution of the past 3-h accumulated rainfall (cm) over Santa Cruz.

For insight into the environmental fields inducing a heavy rain event of this kind, the large-scale meteorological features associated with this rain event were analyzed during 25–27 July 2005. The atmospheric fields were made from the global analyses produced every 6 h by the NCEP Global Forecast System (GFS). The horizontal resolution of the GFS is about 100 km. Figure 3 shows the large-scale flow pattern of the NCEP-analyzed mean sea level pressure and low-level (850 hPa) wind and temperature fields before (Fig. 3a), just before (Fig. 3b), and during (Fig. 3c) this event. Figure 4 shows the large-scale NCEP-analyzed midlevel (700 hPa) wind and temperature fields before (Fig. 4a), just before (Fig. 4b), and during (Fig. 4c) this event. Large-scale synoptic fields like circulation and pressure (Figs. 3 and 4) clearly characterize the active phase of the monsoon over India, particularly over the west coast and peninsular parts of India on 26 July. A low pressure area formed over the northern Bay of Bengal off of the Gangetic region of West Bengal and the Orissa coast on 24 July (not shown), and intensified into a well-marked low as it moved inland. It moved inland and laid over Orissa and adjoining Jharkhand on 26 July (Fig. 3b). Subsequently, it moved northwest and over eastern Madhya Pradesh and adjoining Vidarbha on 27 July (Fig. 3c). It moved northwest and was located over northern Gujarat on 28 July (not shown). The strong cyclonic circulation is seen up to 700 hPa (Fig. 4). In summary, the main synoptic features on 25 and 26 July, shown in Figs. 3 and 4, are (i) a well-marked low pressure area over the northern Bay of Bengal on 25 July over Orissa and adjoining Jharkhand on 26 July and over eastern Madhya Pradesh and adjoining Vidarbha on 27 July, (ii) a trough off of the west coast extending from Konkan to the Karnataka coast, and (iii) cyclonic circulation that extends up to midtropospheric level over Mumbai. These synoptic-scale features are highly favorable (Srinivasan 1972) for the occurrence of heavy to very heavy rainfall over the west coast of India. But it is apparent that large-scale features alone cannot explain an extraordinary rainfall event of this kind.

Fig. 3.

Spatial distribution of 850-hPa temperature (K, shaded), mean sea level pressure (hPa, contour), and 850-hPa streamlines at 0000 UTC (a) 25, (b) 26, and (c) 27 Jul 2005.

Fig. 3.

Spatial distribution of 850-hPa temperature (K, shaded), mean sea level pressure (hPa, contour), and 850-hPa streamlines at 0000 UTC (a) 25, (b) 26, and (c) 27 Jul 2005.

Fig. 4.

Spatial distribution of 700-hPa temperature (K, shaded) and streamlines at 0000 UTC (a) 25, (b) 26, and (c) 27 Jul 2005.

Fig. 4.

Spatial distribution of 700-hPa temperature (K, shaded) and streamlines at 0000 UTC (a) 25, (b) 26, and (c) 27 Jul 2005.

We have analyzed thermodynamic changes that took place during 25–27 July around Mumbai. As mentioned above, the temperature fields of the lower and middle troposphere are shown in Figs. 3 and 4, while midlevel (700 hPa) relative humidity (RH) is shown in Fig. 5 for 25–27 July. By looking at Figs. 3 –5, it is seen that there are significant changes in the thermodynamic conditions near Mumbai from 25 to 27 July. The lower-tropospheric (850 hPa) temperature upstream of Mumbai (i.e., over Arabian Sea) and to the north of Mumbai was higher by about 1–2 K on 26 July (Fig. 3b) as compared with 25 July (Fig. 3a). This higher temperature might have been advected from the African coast by low-level westerlies or may be due to subsidence over this region. The Bay of Bengal was also warmer on 26 July as compared to 25 July. The higher temperature that was seen on 26 July was still present over the Arabian Sea and Bay of Bengal on 27 July, while the higher temperature to the north of Mumbai disappeared by that time. The Bay of Bengal and Arabian Sea were also warmer at 700 hPa (Fig. 4) on 26 and 27 July as compared to 25 July. Very interesting features were also seen in midtropospheric moisture fields on 26 July around Mumbai (Fig. 5b). The relative humidity decreased on 26 July as compared to 25 July. The circulation patterns above 700 hPa show that the winds over Mumbai were northwesterly on 26 July, but were northerly on 25 July. These northwesterly winds brought drier air from Pakistan toward Mumbai and resulted in a lower RH value. The higher temperature at the lower level and the dryness of the midlevel were very conducive for the development of severe mesoscale convective cells. Figure 6 shows the temporal variation of the 950-hPa temperatures, specific humidity, and 700-hPa relative humidity over Santa Cruz. From Fig. 6 it can be seen that just before the initiation of the heavy rain event (0900 UTC 26 July) the 950-hPa temperature (Fig. 6a) and moisture (Fig. 6b) became higher and the midtroposphere turned drier (Fig. 6c). The strong vorticity (Fig. 7) was also present in the middle troposphere just before the time of the strong precipitation. Our analysis indicates that the large vorticity present over Mumbai was not generated locally but rather it was advected (Fig. 7b) from northeast of Mumbai and gradually percolated to the lower-tropospheric levels. This midlevel dryness and mesoscale vortices with higher temperature and moisture at lower levels coupled with active monsoon flow might have caused this heavy rain event.

Fig. 5.

Spatial distribution of 700-hPa relative humidity (%, shaded) and streamlines at 0000 UTC (a) 25, (b) 26, and (c) 27 Jul 2005.

Fig. 5.

Spatial distribution of 700-hPa relative humidity (%, shaded) and streamlines at 0000 UTC (a) 25, (b) 26, and (c) 27 Jul 2005.

Fig. 6.

Temporal evolutions of (a) 950-hPa temperature (K), (b) 950-hPa specific humidity (g kg−1), and (c) 700-hPa relative humidity (%) over Santa Cruz.

Fig. 6.

Temporal evolutions of (a) 950-hPa temperature (K), (b) 950-hPa specific humidity (g kg−1), and (c) 700-hPa relative humidity (%) over Santa Cruz.

Fig. 7.

(a) Temporal evolution of the relative vorticity (×10−5 s−1) with height over Santa Cruz and (b) temporal variation of 600-hPa relative vorticity (×10−5 s−1) with latitude over Santa Cruz.

Fig. 7.

(a) Temporal evolution of the relative vorticity (×10−5 s−1) with height over Santa Cruz and (b) temporal variation of 600-hPa relative vorticity (×10−5 s−1) with latitude over Santa Cruz.

3. AIRS temperature and humidity profiles

AIRS is currently the most advanced and sophisticated space-borne atmospheric profiler. AIRS is one of six instruments on board the NASA Aqua spacecraft, and has been operational since September 2002 (Aumann et al. 2003). AIRS on Aqua is in a sun-synchronous polar orbit with equatorial crossings at ∼1330 and ∼0130 local time. The AIRS instrument has 2378 channels in the thermal infrared spectrum, ranging from 3.7 to 15.4 μm (650 to 2675 cm−1). The AIRS is accompanied by two microwave sounding radiometers, the Advanced Microwave Sounding Unit-A (AMSU-A) and the Humidity Sounder for Brazil (HSB). AMSU-A is a 15-channel temperature sounder utilizing the 55-GHz oxygen absorption band. HSB (which failed in February 2003) is mainly a humidity sounder with channels centered on the water vapor line at 183.31 GHz. AIRS coupled with the AMSU-A form an integrated temperature and humidity sounding system (Susskind et al. 2003). Because of its hyperspectral nature, AIRS can provide almost rawinsonde quality atmospheric temperature profiles with the ability to resolve some small-scale vertical features.

AIRS footprints coincide with AMSU-A footprints, allowing AMSU-A data to be used in the retrieval process (Susskind et al. 2003), particularly over cloudy regions. This produces a uniform distribution of AIRS retrievals in both clear and cloudy conditions at a spatial resolution of approximately 50 km. The superior vertical resolution and sounding accuracy make the instrument very appealing as a complement to rawinsonde measurements in data-sparse regions. The requirement for temperature and moisture soundings from space with rawinsonde accuracy has been the driving force for the development of AIRS. An intensive effort to validate these soundings showed that global satellite retrievals are in many ways superior to conventional radiosondes, although their vertical resolution cannot quite match that of the point measurement from balloon-borne instruments. The global coverage and high accuracy of the temperature and moisture profiles from space have shown to be good additions to traditional radiosondes, particularly over oceans.

For this study, we have used AIRS level 2, version 4.0, atmospheric moisture and temperature profiles during 25 and 26 July 2005 (Table 1). The data have a horizontal resolution of ∼45 km, and the vertical grid is based on the World Meteorological Organization (WMO) standard pressure levels from 1000 to 50 hPa. Globally, the AIRS version 4.0 retrieved profiles compared to rawinsondes collocated in time and space exhibit root-mean-square errors (rmse’s) of 1 K in 1-km layers for temperature and 10%–15% RH in 2-km layers for water vapor (Tobin et al. 2006, Divakarla et al. 2006; Fetzer et al. 2003; Fetzer 2006; Susskind et al. 2006). The lowest errors occur for clear-sky cases over water with a degradation in profile accuracy in cloudy regions and/or over land.

Table 1.

AIRS granules used in the present study.

AIRS granules used in the present study.
AIRS granules used in the present study.

4. Model descriptions and experimental design

a. Model description

The MM5 and its 3DVAR system are utilized in this study. The MM5 is a limited-area, nonhydrostatic primitive equation model with multiple options for various physical parameterization schemes (Dudhia 1993; Grell et al. 1994). The MM5 was employed within a two-way interacting nested (Fig. 8) configuration with 200 × 200 grid points at 45-km and 52 × 52 grid points at 15-km resolutions, with its high-resolution-sized domain covering the region around Mumbai. The model has 28 vertical levels with the top of the atmosphere located at 50 hPa. The major physics options in the experiments include the Grell cumulus parameterization (Grell 1993), explicit treatment (Goddard’s scheme) for ice/graupel physics, the Medium Range Forecast (MRF) model scheme for PBL parameterization (Hong and Pan 1996), and rapid radiative transfer model (RRTM) longwave (Mlawer et al. 1997) and Dudhia shortwave atmospheric radiation schemes (Dudhia 1989).

Fig. 8.

Two nested domains used in MM5 simulations. Resolutions are 45 and 15 km for domains 1 and 2, respectively. The symbol in domain 2 shows the position of Mumbai.

Fig. 8.

Two nested domains used in MM5 simulations. Resolutions are 45 and 15 km for domains 1 and 2, respectively. The symbol in domain 2 shows the position of Mumbai.

b. Assimilation methodology

The 3DVAR was conducted by minimizing the cost function. The MM5 3DVAR method (Barker et al. 2004) is based on the minimization of a cost function defined as (Ide et al. 1997)

 
formula

where x is the analysis variables vector (n dimensional), xb the background variables vector (n dimensional), yo the observation vector (m dimensional), 𝗕 the background error covariances matrix (n × n), and 𝗥 the observation error covariances matrix (m × m). In (1), the analyses x = xa represent the a posteriori maximum likelihood (minimum variance) estimate of the true state of the atmosphere given two sources (xb and yo) of data. The analyses fit to this data are weighed by estimates of their errors (𝗕, 𝗥). The cost function (1) assumes that observational and background error covariances are described using Gaussian probability density functions with zero mean error.

The configuration of the MM5 3DVAR system is based on an incremental formulation producing a multivariate incremental (Courtier et al. 1994) analysis in the MM5 model space. The incremental cost function minimization is performed in a preconditioned control variables space.

The preconditioned control variables are the streamfunction, velocity potential, unbalanced pressure, and relative humidity. The background covariance matrix 𝗕 is prescribed as the monthly mean forecast error variances derived from the yearly MM5 forecasts. The statistics of the differences between 24- and 12-h forecasts valid at 1200 UTC are used to estimate background error covariances by applications of the National Meteorological Center (NMC, now known as NCEP) method to the MM5 forecast. The representation of the horizontal components of the background error is via horizontally isotropic and homogeneous recursive filters. The vertical component is applied through the projection onto climatologically averaged eigenvectors of the vertical error estimated via the NMC method. Horizontal/vertical errors are nonseparable, in that horizontal scales vary with vertical eigenvectors. A detailed description of the 3DVAR system can be found in Barker et al. (2004).

In the MM5 3DVAR method, all observation errors are assumed to be uncorrelated in space and time. Since observation errors are assumed to be uncorrelated, the matrix 𝗥 is a simple diagonal with observation error variances as elements. In this study, these variances are taken as constant in space and time. The observation error estimates for AIRS temperature profiles varies in the vertical, with maxima in the lower and upper levels (1.5 K) and minima in the middle levels (1.2 K). The observation error estimates for the AIRS moisture profiles also varies in the vertical, with maxima in the middle and upper levels (20%) and a minima in the lower levels (15%). These errors are defined based on the available literature about the global accuracy of the AIRS-retrieved temperature and moisture profiles (Tobin et al. 2006; Divakarla et al. 2006; Fetzer 2006; Susskind et al. 2006). However, the choice of the errors for the minimization processes is always a challenge, and more studies are required to adequately define the errors in AIRS-retrieved profiles for assimilation studies.

c. Experiment design

Three analyses are made: one at 0000 UTC and two at 0800 UTC 25 July. The interpolated GFS analysis at 0000 UTC 25 July is used as a first guess for the MM5 3DVAR method (cold start). During the cold start (0000 UTC 25 July), we assimilated conventional data, Quick Scatterometer (QuikSCAT) ocean surface wind vectors (QW; Shirtliffe 1999), Special Sensor Microwave Imager (SSM/I) wind speed (WS) and precipitable water vapor (PW; Hollinger 1989), and Meteosat-5-derived atmospheric motion vectors (AMVs; information online at www.archive.eumetsat.org). The conventional data include surface station reports as well as upper-air observations. The University of Wyoming (see their Web site at http://weather.uwyo.edu) provides surface and upper-air data over the land, while the surface marine observations (ships and buoys) are taken from the International Comprehensive Ocean–Atmosphere Dataset (ICOADS). The National Climate Data Center (NCDC) provided the ICOADS data (information online at www.ncdc.noaa.gov/oa/marine.html). The QuikSCAT and SSM/I observation are available at 25 km only in the nonprecipitating environment over the oceanic region. More details about the QuikSCAT and SSM/I data products can be found online (ftp.ssmi.com).

There is a possibility that the above-mentioned data (conventional and satellite) were already ingested by the GFS assimilation cycle at 0000 UTC 25 July. Most of these observations (except the conventional) are at a finer scale (25 km) compared with the GFS resolution (100 km). We have conducted an assimilation at higher spatial resolution (45 km) as compared to that of the GFS (100 km) and decided to assimilate these observations at 0000 UTC 25 July to incorporate some of the unresolved features in the GFS analysis. The 8-h MM5 forecast (from the first 3DVAR analysis) ending at 0800 UTC 25 July serves as the first guess for the second and third MM5 3DVAR analyses. We assimilated conventional (surface only), Meteosat-5 AMVs, and AMSR-E-derived WS and PW in the case of the second MM5 3DVAR analysis (hereafter CNT) at 0800 UTC 25 July. The third MM5 3DVAR analysis (hereafter EXP) is identical to CNT in every aspect except that AIRS-derived temperature and moisture profiles are assimilated in addition to AMSR-E, Meteosat-5, and conventional data. In all of the analyses, we have used the observation errors defined by NCAR in the MM5 3DVAR system for conventional, QuikSCAT, SSM/I, and Meteosat observations. The observation error for AMSR-E is similar to that of SSM/I, while the observation error used for the AIRS is mentioned in the earlier section.

The resolution of AMSR-E is 25 km, which is same as that of QuikSCAT and SSM/I. Like SSM/I, AMR-E-retrieved precipitable water vapor and oceanic surface winds are available over nonprecipitating oceanic environments only. There are two swaths (0712 and 0748 UTC 25 July), from AMSR-E, which passes over the Bay of Bengal and the Arabian Sea. Three levels of Meteosat-5 AMVs are available at 0800 UTC. The spatial distribution of AMSR-E-retrieved WS, PW, and Meteosat-5-derived upper-level (above 300 hPa) winds are shown in Fig. 9. The Meteosat-5-derived lower- and middle-level winds, which are also assimilated, are not shown in Fig. 9. The spatial distribution of the AIRS observations, which are assimilated at 0800 UTC, is shown in the next section (Fig. 13). The data are assimilated on the coarsest, 45-km-resolution domain (domain 1). The model initial condition for the second domain (15-km resolution) is interpolated from domain 1. Prior to data assimilation, data underwent the following quality checking processes in order to reduce the possibility of assimilating bad observations.

Fig. 9.

The spatial coverage of (a) AMSR-E wind speed (m s−1, shaded region) and precipitable water vapor (g cm−2, contour lines), and (b) Meteosat-5-derived upper-level atmospheric motion vectors (AMVs) at 0800 UTC 25 Jul 2005.

Fig. 9.

The spatial coverage of (a) AMSR-E wind speed (m s−1, shaded region) and precipitable water vapor (g cm−2, contour lines), and (b) Meteosat-5-derived upper-level atmospheric motion vectors (AMVs) at 0800 UTC 25 Jul 2005.

Fig. 13.

The spatial distribution of vertically averaged rmse (K) (a) between AIRS data and CNT and (b) between AIRS data and EXP, and (c) the improvement parameters (η = ab), valid at 0800 UTC 25 Jul 2005.

Fig. 13.

The spatial distribution of vertically averaged rmse (K) (a) between AIRS data and CNT and (b) between AIRS data and EXP, and (c) the improvement parameters (η = ab), valid at 0800 UTC 25 Jul 2005.

First, the rain-contaminated data from microwave sensors (SSM/I, QuikSCAT, and AMSR-E) were excluded. The rainfall probability parameter p along with QuikSCAT winds is used to exclude the rain-contaminated observations from QuikSCAT winds. The SSM/I and AMSR-E retrievals were already flagged for the rain-contaminated pixels by the SSM/I and AMSR data products generation team and we have not done further flagging. A set of quality assurance (QA) flags provided by AIRS science team members (Susskind et al. 2006) were used to filter the cloud-contaminated temperature and moisture data. The quality flags that are relevant for the temperature and moisture profiles are 1) Qual_Temp_Profile_Top, 2) Qual_Temp_Profile_Mid, 3) Qual_Temp_Profile_Bot, and 4) Qual_Surf. These four quality flags assure the quality of the temperature as well as the moisture values at various pressure levels pertaining to the stratosphere and upper troposphere, middle troposphere, lower troposphere, and surface, respectively. Each profile of AIRS contains level-specific quality indicators (as mentioned above) allowing us to select the highest quality of data (products quality flag = 0). This fact is reflected in the data gaps shown later (Fig. 13). Further, a gross error quality control was performed in which observations that differed from the model’s first guess by more than 5 times the observation error were removed.

MM5 was integrated for 48 h from these two (CNT, EXP) analyses. The NCEP analyses with 1° × 1° resolution are used to provide the lateral boundary conditions for MM5. The differences in the forecasts from these two initial conditions is due to the use of AIRS temperature and moisture profiles.

5. Results

a. Impact of AIRS data on the initial analysis

The impact of the AIRS profiles on the temperature and moisture analyses is examined at 850 and 700 hPa. AIRS observations (Table 1; for 25 July) along with other conventional and satellite (AMSR-E and Meteosat-5) data are assimilated into the model initial conditions (0800 UTC 25 July) in the AIRS data sensitivity experiment (EXP), while in the case of CNT we have assimilated only conventional, AMSR-E, and Meteosat-5-derived AMVs. Figure 10 shows the distribution of the 850-hPa initial temperature field at 0800 UTC 25 July for CNT (Fig. 10a) and the difference field (EXP − CNT; Fig. 10b). Differences between the EXP and CNT analyses are typically of the order of ±0.5 K across the domain. Over some places, there are major differences between the CNT analysis and the analysis obtained from EXP. A general area of warming in excess of 1.5 K is found in northwest India and adjoining Pakistan and the southeast Arabian Sea, while an area of cooling in excess of 1 K is seen over the central Arabian Sea and the eastern Bay of Bengal. Figure 11 shows the distribution of the 700-hPa temperature field at 0800 UTC 25 July for CNT (Fig. 11a) and the difference field (EXP − CNT; Fig. 11b). Over most of the places the differences are within ±0.5 K. Significant warming (∼1.5) is seen in the northwestern part of the domain (Pakistan, Iran, Saudi Arabia), over north India, the central Arabian Sea, and the northern and southeastern Bay of Bengal. Another small pocket of high warming (1.5 K) is seen to the northwest of Mumbai. Some scattered pockets of cooling with magnitudes of the order of 1 K are seen over central India, the southwest Arabian Sea, the northwest Arabian Sea and the southeast Bay of Bengal. Figure 12 shows the spatial distribution of the 850-hPa specific humidity (g kg−1) at 0800 UTC 25 July for CNT (Fig. 12a) and the difference field (EXP − CNT; Fig. 12b). The assimilation of the AIRS profiles makes the atmosphere more humid as compared to that of CNT. The positive differences as large as 1.5 g kg−1 are particularly evident over the observation locations. The negative differences of the order of 0.5 g kg−1 are seen mostly over land. The MM5 3DVAR scheme also produces increments of the wind and pressure (not shown) in response to the temperature and moisture fields due to its multivariate nature.

Fig. 10.

(a) The initial temperature field (K) at 850 hPa in CNT and (b) the difference (K) between EXP and CNT valid at 0800 UTC 25 Jul 2005.

Fig. 10.

(a) The initial temperature field (K) at 850 hPa in CNT and (b) the difference (K) between EXP and CNT valid at 0800 UTC 25 Jul 2005.

Fig. 11.

(a) The initial temperature field (K) at 700 hPa in CNT and (b) difference (K) between EXP and CNT valid at 0800 UTC 25 Jul 2005.

Fig. 11.

(a) The initial temperature field (K) at 700 hPa in CNT and (b) difference (K) between EXP and CNT valid at 0800 UTC 25 Jul 2005.

Fig. 12.

(a) The initial specific humidity (g kg−1) at 850 hPa in CNT and (b) difference (g kg−1) between EXP and CNT valid at 0800 UTC 25 Jul 2005.

Fig. 12.

(a) The initial specific humidity (g kg−1) at 850 hPa in CNT and (b) difference (g kg−1) between EXP and CNT valid at 0800 UTC 25 Jul 2005.

The spatial distributions of the vertically averaged rmse between the AIRS data and the CNT analysis are shown in Fig. 13a, while Fig. 13b shows the spatial distribution of the vertically averaged rmse between the AIRS and EXP analyses. Comparisons between the AIRS- and CNT-analyzed temperature fields demonstrate that the domain-averaged rmse is about 2.2 K (Fig. 13a). Comparisons between the AIRS- and EXP-analyzed temperature show (Fig. 13b) better agreement (rmse of 1.1 K). The rmse’s are generally higher over the land as compared to oceans in both the cases. The AIRS − CNT rmse of 2.2 K reduces to 1.1 K for AIRS − EXP. This shows that 3DVAR produces an analysis that fits the AIRS observations well. Further, to see the spatial distribution of the improvement after the assimilation of the AIRS data, we have computed an improvement parameter η defined as η = (rmse between AIRS and CNT) − (rmse between AIRS and EXP). The distribution of the vertically averaged η is given in Fig. 13c. The positive value of η indicates real improvement. Over all the observation locations with the exception of some isolated pockets the value of η is positive, which indicates successful assimilation of AIRS data. A similar analysis has been done (not shown) for humidity. The analysis of the humidity fields also shows a successful assimilation of the AIRS moisture profiles. The domain-averaged rmse of the EXP- and CNT-analyzed temperature fields with respect to AIRS data (Fig. 14) reveals that the EXP-analyzed temperature has improved significantly over all the levels as compared to CNT.

Fig. 14.

The vertical variation of the domain-averaged rmse for CNT and EXP at 0800 UTC 25 Jul 2005. The rmse’s (K) are based on the comparison of CNT and EXP with AIRS data.

Fig. 14.

The vertical variation of the domain-averaged rmse for CNT and EXP at 0800 UTC 25 Jul 2005. The rmse’s (K) are based on the comparison of CNT and EXP with AIRS data.

To see the closeness of the EXP-analyzed temperature and moisture fields with the independent observations (not assimilated), we have compared the EXP-analyzed temperature and moisture fields with Moderate Resolution Imaging Spectroradiometer (MODIS) retrieved temperature and moisture profiles at 0800 UTC 25 July. Two granules (0840 and 0845 UTC 25 July) from MODIS/Aqua (Zavodsky et al. 2004) are available over the Arabian Sea. We have computed an improvement parameter η defined as η = (rmse between MODIS and CNT) − (rmse between MODIS and EXP). The distribution of the vertically averaged η is given in Fig. 15a (for temperature) and Fig. 15b (for humidity). The positive value of η indicates improvement. Over all the observation locations, except for some isolated pockets, the value of η is positive, which indicates the improved thermodynamics state of the atmosphere at 0800 UTC 25 July in the EXP analysis. This shows that the assimilation of AIRS data brings our analysis closer to the observations (which is independent of the AIRS).

Fig. 15.

The spatial distribution of the vertically averaged improvement parameter (η) for (a) temperature and (b) specific humidity, valid at 0800 UTC 25 Jul 2005. This is based on the comparison of CNT and EXP with MODIS-observed temperature and moisture profiles.

Fig. 15.

The spatial distribution of the vertically averaged improvement parameter (η) for (a) temperature and (b) specific humidity, valid at 0800 UTC 25 Jul 2005. This is based on the comparison of CNT and EXP with MODIS-observed temperature and moisture profiles.

b. Impact of the AIRS data on numerical simulation

1) Rainfall prediction

To assess the impact of the AIRS-observed temperature and moisture profiles on the numerical forecast, the CNT experiment is carried out to serve as a benchmark. The 24-h (ending at 0300 UTC 27 July 2005) accumulated Tropical Rainfall Measuring Mission (TRMM) observed rainfall (cm) is plotted in Fig. 16 (top). The middle and bottom panels of Fig. 16 show the 24-h predicted (ending at 0300 UTC 27 July 2005) accumulated rainfall from CNT and EXP. The TRMM 3B42 (Adler et al. 2000) rainfall is at 3-h temporal resolution and 0.25° × 0.25° spatial resolution in a global belt extending from 50°S to 50°N. The model-predicted rainfall for the 15-km domain was resampled to 25 km (using bilinear interpolation) in order to compare it with the TRMM rainfall. Also it should be noted that the TRMM 3B42 algorithm’s 3-h rain rate is converted to accumulated rainfall assuming constant rain rate over 3 h. From Fig. 16, it can be seen that the heavy rainfall event is well captured by the TRMM data with the 24-h accumulated rainfall being of the order of 40 cm over Mumbai. Although, the rainfall is drastically underestimated by TRMM, the location perfectly matches with the observed rainfall (Fig. 1). Rainfall prediction was significantly improved (relative to the control simulation; Fig. 16, middle panel) by incorporation of the AIRS data (Fig. 16, bottom panel). The control simulation has hardly predicted any significant rainfall over Mumbai. The rainfall (∼55 cm) in the control simulation is found to be confined to the far northeast (20.2°N, 73.2°E) of the observed position (19.15°N, 72.84°E). In contrast, the rainfall predicted by the assimilation experiment (Fig. 16, bottom panel) is better matched (spatial distribution) with the TRMM data than with the control run. The magnitude and spatial pattern of the model-simulated rainfall in the case of EXP are in close agreement with the IMD-observed rainfall (Fig. 1). The pockets of highest rainfall amount (95 cm) in EXP exactly match the location (19.15°N. 72.84°E) and intensity of the IMD-observed maxima rainfall (94 cm). The temporal evolution of the model-simulated rainfall indicates (not shown) that a major fraction of the rainfall occurred around 0000 UTC 27 July instead of around 1200 UTC 26 July (as seen in the observations; Fig. 2).

Fig. 16.

The 24-h accumulated rainfall (cm) valid for 0300 UTC 27 Jul 2005 for (a) TRMM 3B42 observed, (b) CNT predicted, and (c) EXP predicted.

Fig. 16.

The 24-h accumulated rainfall (cm) valid for 0300 UTC 27 Jul 2005 for (a) TRMM 3B42 observed, (b) CNT predicted, and (c) EXP predicted.

Another way of verifying a precipitation forecast is through the use of the equitable threat score (ETS) and bias statistics, both of which are based on a contingency table approach (Colle et al. 1999). This contingency table is shown in Table 2. It is a 2 × 2 matrix, where each element of the matrix holds the number of occurrences in which the model and the observations did or did not reach a certain threshold amount of accumulated precipitation.

Table 2.

A contingency table for ETS and bias skill score metrics. The letter A, B, C, and D represent the number of occurrences for which the model forecast precipitation or the observed precipitation did (yes) or did not (no) reach/exceed a given threshold value.

A contingency table for ETS and bias skill score metrics. The letter A, B, C, and D represent the number of occurrences for which the model forecast precipitation or the observed precipitation did (yes) or did not (no) reach/exceed a given threshold value.
A contingency table for ETS and bias skill score metrics. The letter A, B, C, and D represent the number of occurrences for which the model forecast precipitation or the observed precipitation did (yes) or did not (no) reach/exceed a given threshold value.

Based on the contingency table, a bias score is defined as

 
formula

where F is the number of the forecasts at the observation stations with precipitation equal to or exceeding a given threshold and O is the number of occurrences in the observations that meet or exceed the threshold. Thus, the bias score indicates how well the model predicts the precipitation coverage. A bias score of 1 indicates perfect precipitation coverage while a value of less (more) than 1 indicates under- (over-) forecasting of precipitation over the grid.

The ETS indicates how well the forecasted precipitation region matches the observed precipitation region that exceeds a given threshold. A higher ETS indicates a more accurate forecast of precipitation location and intensity. An ETS of 1 indicates that the precipitation fields are perfectly aligned and an ETS of 0 means there are no matches at all. The ETS is defined by

 
formula

where H is the number of the forecast “hits,” a hit being defined as an occurrence of both the simulated and observed precipitation meeting or exceeding a given precipitation threshold at a point. In Eq. (3), F and O are defined as above for Eq. (2) and E is defined as

 
formula

where N is the total number of the observations utilized in the verification process.

For the quantitative comparison the ETS and bias scores are computed using both sets (TRMM and IMD) of observed rainfall data. The area of the domain used for the ETS and bias scores computation is 16°–22°N, 70°–75°E. The ETS and bias score are computed by resampling the model output to 25-km grid resolution when TRMM rainfall data are used as the observed rainfall, while 15-km grid resolution is used when the IMD rainfall data are used as the observed rainfall. Figure 17 shows the ETS and bias scores computed using TRMM data as the observed rainfall for different thresholds for the control and assimilation runs. It can be seen from Fig. 17 that for the 24-h accumulated rainfall (ending at 0300 UTC 27 July 2005) the improvement in the ETS for the experiment with the AIRS data (EXP) is reflected in all of the rainfall threshold categories except the categories below 5 cm, in which CNT shows slightly higher skill. Compared with TRMM, both assimilations (CNT and EXP) shows higher bias. The bias is very high in the case of EXP as compared with CNT and this is because TRMM underestimates the rainfall (Fig. 16a) as compared with the observations (Fig. 1). Figure 18 shows the ETS and bias scores computed using IMD data as the observed rainfall for different thresholds for the CNT and EXP runs. The model with the AIRS data assimilation shows better skill in predicted rainfall over all the thresholds than without data assimilation. This confirms the positive impact of the AIRS data on the simulation of Mumbai’s heavy rain event in terms of the magnitude and spatial distribution of the predicted rainfall.

Fig. 17.

(a) ETS and (b) bias score of 24-h accumulated rainfall (ending at 0300 UTC 27 Jul) compared with TRMM 3B42 observed rainfall for different thresholds, for CNT and EXP. The model output of 15 km is degraded to 25 km to match up with the TRMM 3B42 resolution.

Fig. 17.

(a) ETS and (b) bias score of 24-h accumulated rainfall (ending at 0300 UTC 27 Jul) compared with TRMM 3B42 observed rainfall for different thresholds, for CNT and EXP. The model output of 15 km is degraded to 25 km to match up with the TRMM 3B42 resolution.

Fig. 18.

(a) ETS and (b) bias score of 24-h accumulated rainfall (ending at 0300 UTC 27 Jul) compared with IMD-observed rainfall for different thresholds, for CNT and EXP.

Fig. 18.

(a) ETS and (b) bias score of 24-h accumulated rainfall (ending at 0300 UTC 27 Jul) compared with IMD-observed rainfall for different thresholds, for CNT and EXP.

2) Why is EXP-predicted rainfall better than that of CNT?

Why did the AIRS data assimilation experiment predicted rainfall in a better way as compared to the control experiment? We have analyzed the predicted meteorological parameters over Mumbai and the surrounding region in order to answer the above question.

Figure 19 shows the model-predicted mean sea level pressure (MSLP) around Mumbai at 0000 UTC 26 and 27 July from CNT (Figs. 19a and 19b) and EXP (Figs. 19c and 19d). At 0000 UTC 26 July, the west coast trough (indicative of active monsoon conditions) is well (intensity as well as structure) simulated by the assimilation experiment (Fig. 19c) as compared to the control run (Fig. 19a). The west coast trough was also very active in the sea level charts prepared by the IMD. At 0000 UTC 27 July (Figs. 19b and 19d), over Mumbai, the EXP-simulated MSLP is 995 hPa, which is lower than the result for CNT (996 hPa). At 0000 27 July, the MSLP is very low to the north of Mumbai in CNT (Fig. 19b) as compared to EXP (Fig. 19d). The simulated low MSLP to the north of Mumbai in CNT caused more convergence of moisture from the surrounding region and might have prevented the rainfall occurrence over Mumbai by shifting the center of convergence.

Fig. 19.

Spatial distribution of model-predicted MSLP (hPa) valid at 0000 UTC (a) 26 Jul from CNT, (b) 27 Jul from CNT, (c) 26 Jul from EXP, and (d) 27 Jul from EXP.

Fig. 19.

Spatial distribution of model-predicted MSLP (hPa) valid at 0000 UTC (a) 26 Jul from CNT, (b) 27 Jul from CNT, (c) 26 Jul from EXP, and (d) 27 Jul from EXP.

Figure 20 shows the model-predicted near-surface (10-m height) horizontal (streamlines) and 700-hPa vertical wind fields (shaded) at 0000 UTC 26–27 July for the control run (Figs. 20a and 20b) and assimilation (Figs. 20c and 20d) experiments. In the case of the assimilation experiment, the surface wind field on 26 July (Fig. 20c) shows an area of strong convergence to the north of Mumbai, where the westerly flow from the Arabian Sea is forced to turn abruptly, becoming southerly (to the northeast of Mumbai) ahead of topography (thick white contours), nearly parallel to the terrain contours. The convergence area near Mumbai is also associated with larger vertical velocity (shaded region) with RH close to saturation (RH not shown). The vertical velocities are very high along the west coast in EXP (Fig. 20c). In CNT, the convergence of the wind is missing and little vertical velocity is seen along the west coast (Fig. 20a). On 27 July, EXP exhibits cyclonic circulation (Fig. 20d) with large vertical velocity (>2.5 m s−1) over Mumbai. This cyclonic circulation is completely missing in the control run (Fig. 20b). In fact, in the control run, the westerly wind blowing from the Arabian Sea is not abruptly deflected between the coast and the mountains but, rather, undergoes a weak northward deviation (Fig. 20b) once well inland. One pocket of high vertical velocities is seen to the northeast of Mumbai in CNT, which is consistent with the location of the maximum precipitation simulated by CNT (Fig. 16b). Figure 21 shows the model-predicted 850-hPa specific humidity around Mumbai at 0000 UTC 26 and 27 July from CNT (Figs. 21a and 21b) and EXP (Figs. 21c and 21d). Compared with CNT, EXP shows higher amounts of moisture around Mumbai on both days. The moisture distribution at 850 hPa looks very similar (Fig. 21) to that of the 700-hPa vertical velocity (Fig. 20). Analysis of the 700-hPa moisture (not shown) also indicates that the moisture is higher in EXP than CNT. This shows that the probability of more precipitation is clearly seen in the meteorological conditions simulated by the assimilation experiment. So, the initial condition, obtained by assimilating the AIRS data, leads to a simulation that is very much favorable for the heavy rain.

Fig. 20.

Spatial distribution of model-predicted 10-m horizontal (streamlines) and 700-hPa vertical (m s−1, shaded) wind field, valid at 0000 UTC (a) 26 Jul from CNT, (b) 27 Jul from CNT, (c) 26 Jul from EXP, and (d) 27 Jul from EXP.

Fig. 20.

Spatial distribution of model-predicted 10-m horizontal (streamlines) and 700-hPa vertical (m s−1, shaded) wind field, valid at 0000 UTC (a) 26 Jul from CNT, (b) 27 Jul from CNT, (c) 26 Jul from EXP, and (d) 27 Jul from EXP.

Fig. 21.

Spatial distribution of model-predicted 850-hPa specific humidity (g kg−1), valid at 0000 UTC (a) 26 Jul from CNT, (b) 27 Jul from CNT, (c) 26 Jul from EXP, and (d) 27 Jul from EXP.

Fig. 21.

Spatial distribution of model-predicted 850-hPa specific humidity (g kg−1), valid at 0000 UTC (a) 26 Jul from CNT, (b) 27 Jul from CNT, (c) 26 Jul from EXP, and (d) 27 Jul from EXP.

The 24-h model-predicted temperature and moisture profiles from CNT and EXP are compared with AIRS observations (Table 2; 26 July) at 0800 UTC 26 July 2005. Figure 22 shows the improvement parameter [η = (rmse between AIRS and CNT) − (rmse between AIRS and EXP)]. The positive value of η indicates improvement in the temperature forecast by EXP as compared with CNT. A large area of positive values is seen over the southern Indian Ocean, southwestern Arabian Sea, India, and extreme northern part of the domain. There are some regions (with negatives values) where the accuracy of the temperature prediction in EXP is degraded due to the assimilation of AIRS data. The area covered by the negative values is much smaller than the area covered by the positive values. This shows that the assimilation of AIRS data in the initial conditions improved the 24-h temperature forecast over the Indian landmass, as well as other data-sparse regions (e.g., the Indian Ocean and Asian landmass). The analysis of humidity also indicates (not shown) the improvement in humidity prediction due to the assimilation of the AIRS data.

Fig. 22.

Spatial distribution of vertically averaged improvement parameter (η) computed using AIRS-observed and 24-h model-predicted temperature (K) profiles in CNT and EXP.

Fig. 22.

Spatial distribution of vertically averaged improvement parameter (η) computed using AIRS-observed and 24-h model-predicted temperature (K) profiles in CNT and EXP.

Figure 23 shows the height–temperature variation of the vorticity over Santa Cruz from the CNT and EXP runs. Strong vorticity is present in the middle levels before the time of the heavy rainfall event in CNT as well as in the EXP runs. The vorticity is very strong and persists for a longer time in EXP compared with CNT. It is very interesting to see that the midlevel vorticity underwent a gradual downward shift after 0600 UTC 26 July and finally merged with another strong vorticity cell that extended from the surface to upper levels. This downward shifting of the midlevel vorticity is not as prominent in the CNT run. The strong vertical velocities (not shown) are seen throughout the troposphere near the end of the simulation in the assimilation run. In the control run, the vertical velocities are weak and confined up to midtropospheric levels.

Fig. 23.

Temporal–height variation of model-predicted relative vorticity (×10−5 s−1) over Santa Cruz for (a) CNT and (b) EXP.

Fig. 23.

Temporal–height variation of model-predicted relative vorticity (×10−5 s−1) over Santa Cruz for (a) CNT and (b) EXP.

The temporal evolution of the model-predicted surface forcing and stability parameters over Santa Cruz is shown in Fig. 24. Figure 24 reveals that the surface temperature (Fig. 24a) and moisture (Fig. 24b) were more favorable (around 0000 UTC 26 July) to severe convection in the assimilation experiment as compared with the control run. The total totals index (TT) is a measure of stability and is used as a severe weather forecast tool. The TT is a simple index derived from the temperature lapse rate between 850 and 500 hPa, and moisture content at 850 hPa. The TT is actually a combination of vertical totals, VT = T850T500, and the cross totals, CT = Td850T500, so that the sum of the two products is the total total. Therefore, TT is equal to the temperature at 850 hPa plus the dewpoint at 850 hPa, minus twice the temperature at 500 hPa. The TT around 44 K indicates the chance of an isolated thunderstorm. As the value of TT increases from 44 K the weather becomes increasingly severe. The temporal evolution (Fig. 24c) of the TT indicates that the EXP-simulated thermodynamics of the atmosphere is nearly favorable for the convection later on 25 July and early on 26 July, while in CNT the values are far below what is needed for the convection.

Fig. 24.

Temporal evolutions of model-predicted (CNT and EXP) (a) temperature (K) at 2 m, (b) specific humidity (g kg−1) at 2 m, and (c) TT (K) over Santa Cruz.

Fig. 24.

Temporal evolutions of model-predicted (CNT and EXP) (a) temperature (K) at 2 m, (b) specific humidity (g kg−1) at 2 m, and (c) TT (K) over Santa Cruz.

To investigate the effects of the AIRS data on the diabatic forcing of the model atmosphere, the difference between the EXP- and CNT-simulated surface latent and sensible heat fluxes at 1600 UTC 26 July (time of strong convection) around Mumbai is shown in Fig. 25. It can be seen from Fig. 25 that in EXP Mumbai is associated with much higher latent (difference is about 120 W m−2) and sensible (difference is about 60 W m−2) heat fluxes as compared to those of CNT. The response of the model atmosphere to the inclusion of the AIRS data is thus reflected in the surface heat fluxes, which in turn could affect the convection in the subsequent forecast hours. The analysis of the large-scale, as well local, meteorological features indicates that, as compared to the control run, the assimilation run shows more favorable conditions for the occurrence of heavy rain around Mumbai.

Fig. 25.

Difference between EXP- and CNT-predicted (a) latent heat flux (W m−2) and (b) sensible heat flux (W m−2) valid at 1600 UTC 26 Jul 2005.

Fig. 25.

Difference between EXP- and CNT-predicted (a) latent heat flux (W m−2) and (b) sensible heat flux (W m−2) valid at 1600 UTC 26 Jul 2005.

6. Conclusions

The present study utilizes 3DVAR to assimilate AIRS-retrieved temperature and moisture profiles into mesoscale model (MM5) initial conditions for simulation of a record-breaking heavy rain event (with 24-h rainfall exceeding 94 cm) that occurred over Mumbai, India, on 26 July 2005.

NCEP analyses are used to study the possible causes of this very heavy rain event by analyzing the temporal evolution of the meteorological features around Mumbai. The temporal evolution of the meteorological features clearly indicates the formation of midtropospheric mesoscale vortices over Mumbai that exactly coincide with the duration of the intense rainfall. Analysis also indicated the midlevel dryness with higher low-level temperature and moisture. The midlevel dryness associated with high temperature and moisture in the lower levels increase the conditional instability, which is conducive for the development of very severe local thunderstorms. It is seen from the above study that the exceptional rainfall event of 26 July 2005 over Mumbai occurred due to interaction between midtropospheric mesoscale vortices, lower-level instability, and the active monsoon conditions.

It is seen from the numerical experiments that the assimilation of AIRS data could produce the environmental features responsible for heavy rainfall, which were not produced by our control simulation. The west coast trough simulated by our assimilation experiment is better matched with the observations than is CNT. Similarly, the lower- and upper-level moisture around and upstream of Mumbai was higher in EXP as compared to CNT. The temporal evolution of the surface forcing is better simulated by the assimilation experiment. It is seen that sensible and latent heat fluxes are considerably higher over Mumbai in the experiment with AIRS data than in CNT. This enhancement of the heat and moisture fluxes appears to have influenced the mesoscale convection, which in turn had a positive impact on the precipitation forecast. Also, the midlevel mesoscale vortices were very strong and lasted longer in the assimilation run as compared with the control run.

The simulation of rainfall has shown qualitative and quantitative improvements with the inclusion of AIRS data in the initial conditions. The experiment using AIRS data in the initial conditions (EXP) performed better than did CNT (without AIRS data in the initial conditions) both in term of rainfall amount and distribution. CNT reproduced about 55 cm of rain and the location was far northeast of Mumbai, whereas EXP could reproduce 95 cm of rain and the location was over Mumbai. The ETS and the bias score suggest that the improvement in the prediction of the location and amount of heavy rainfall is achieved by incorporating the AIRS data into the initial conditions. Here, the threat score is used to show the relative improvement in this particular case. For significance, many case studies have to be done.

Acknowledgments

MM5 data are made publicly available and supported by the Mesoscale and Microscale Meteorology Division at the National Center for Atmospheric Research (NCAR/MMM). The dedication and hard work of the MMM staff are gratefully acknowledged. The authors sincerely thank Dr. S. Rizvi and Dale Barker at NCAR for their interests and advice about variational assimilation in MM5. The authors would like to acknowledge the National Centers for Environmental Prediction (NCEP) for making analysis data available at their Web site. The AIRS and TRMM data were obtained online (http://daac.gsfc.nasa.gov/data/datasets). The Meteosat-5 AMVs were obtained from EUMETSAT (information online at www.archive.eumetsat.org). The SSM/I, QuikSCAT, and AMSR-E data were obtained online (ftp.ssmi.com).

REFERENCES

REFERENCES
Adler
,
R. F.
,
D. T.
Bolun
,
S.
Curtis
, and
E. J.
Nelkin
,
2000
:
Tropical rainfall distributions determined using TRMM combined with other satellite and rain gauge information.
J. Appl. Meteor.
,
39
,
2007
2023
.
Atlas
,
R.
,
2005
:
The impact of AIRS data on weather prediction.
Proc. SPIE
,
5806
,
599
606
.
Aumann
,
H. H.
, and
Coauthors
,
2003
:
AIRS/AMSU/HSB on the Aqua mission: Design, science objectives, data products, and processing systems.
IEEE Trans. Geosci. Remote Sens.
,
41
,
253
264
.
Barker
,
D. M.
,
W.
Huang
,
Y. R.
Guo
,
A. J.
Bourgeois
, and
Q. N.
Xiao
,
2004
:
A three-dimensional variational data assimilation system for MM5: Implementation and initial results.
Mon. Wea. Rev.
,
132
,
897
914
.
Bohra
,
A. K.
,
S.
Basu
,
E. N.
Rajagopal
,
G. R.
Iyengar
,
M.
Das Gupta
,
R.
Ashrit
, and
B.
Athiyaman
,
2006
:
Heavy rainfall episode over Mumbai on 26 July 2005: Assessment of NWP guidance.
Current Sci.
,
90
,
1188
1194
.
Chahine
,
M. T.
, and
Coauthors
,
2006
:
AIRS: Improving weather forecasting and providing new data on greenhouse gases.
Bull. Amer. Meteor. Soc.
,
87
,
911
926
.
Chen
,
S. H.
,
2007
:
The impact of assimilating SSM/I and QuikSCAT satellite winds on Hurricane Isidore simulation.
Mon. Wea. Rev.
,
135
,
549
566
.
Chen
,
S. H.
,
F.
Vandenberghe
,
G. W.
Petty
, and
J. F.
Bresch
,
2004
:
Application of SSM/I satellite data to a hurricane simulation.
Quart. J. Roy. Meteor. Soc.
,
130
,
801
825
.
Chou
,
S-H.
,
B.
Zavodsky
,
G.
Jedlovec
, and
W.
Lapenta
,
2006
:
Assimilation of Atmospheric Infrared sounder (AIRS) data in regional model. Preprints, 14th Conf. on Satellite Meteorology and Oceanography, Atlanta, GA, Amer. Meteor. Soc., P5.12. [Available online at http://ams.confex.com/ams/pdfpapers/103317.pdf.]
.
Colle
,
B. A.
,
K. J.
Westrick
, and
C. F.
Mass
,
1999
:
Evaluation of MM5 and Eta 10 precipitation forecast over the Pacific Northwest during the cool season.
Wea. Forecasting
,
14
,
137
154
.
Courtier
,
P.
,
J. N.
Thepaut
, and
A.
Hollingsworth
,
1994
:
A strategy for operational implementation of 4DVAR using an incremental approach.
Quart. J. Roy. Meteor. Soc.
,
120
,
1367
1387
.
Divakarla
,
M. G.
,
C. D.
Barnet
,
M. D.
Goldberg
,
L. M.
McMillin
,
E.
Maddy
,
W.
Wolf
,
L.
Zhou
, and
X.
Liu
,
2006
:
Validation of Atmospheric Infrared Sounder temperature and water vapor retrievals with matched radiosonde measurements and forecasts.
J. Geophys. Res.
,
111
.
D09S15, doi:10.1029/2005JD006116
.
Doyle
,
J. D.
, and
T. T.
Warner
,
1988
:
Verification of mesoscale objective analysis of VAS and rawinsonde data using the March 1982 AVE/VAS special network data.
Mon. Wea. Rev.
,
116
,
358
367
.
Dudhia
,
J.
,
1989
:
Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model.
J. Atmos. Sci.
,
46
,
3077
3107
.
Dudhia
,
J.
,
1993
:
A nonhydrostatic version of the Penn State–NCAR Mesoscale Model: Validation tests and simulation of an Atlantic cyclone and cold front.
Mon. Wea. Rev.
,
121
,
1493
1513
.
English
,
S. J.
,
R. J.
Renshaw
,
P. C.
Dibben
,
A. J.
Smith
,
P. C.
Rayer
,
C.
Poulsen
,
F. W.
Saunders
, and
J. R.
Eyre
,
2000
:
A comparison of the impact of TOVS and ATOVS satellite sounding data on the accuracy of numerical weather forecasts.
Quart. J. Roy. Meteor. Soc.
,
126
,
2911
2931
.
Fetzer
,
E. J.
,
2006
:
Preface to special section: Validation of Atmospheric Infrared Sounder observations.
J. Geophys. Res.
,
111
.
D09S01, doi:10.1029/2005JD007020
.
Fetzer
,
E. J.
, and
Coauthors
,
2003
:
AIRS/AMSU/HSB validation.
IEEE Trans. Geosci. Remote Sens.
,
41
,
418
431
.
Gal-Chen
,
T.
,
B. D.
Schmidt
, and
L. W.
Uccellini
,
1986
:
Simultaneous experiments for testing the assimilation of geostationary satellite temperature retrievals into a numerical prediction model.
Mon. Wea. Rev.
,
114
,
1213
1230
.
Garand
,
L.
,
A.
Beaulne
, and
N.
Wagneur
,
2006
:
Assimilation of AIRS hyperspectral radiances at MSC. Preprints, 14th Conf. on Satellite Meteorology and Oceanography, Atlanta, GA, Amer. Meteor. Soc., P5.13. [Available online at http://ams.confex.com/ams/pdfpapers/98316.pdf.]
.
Grell
,
G. A.
,
1993
:
Prognostic evaluation of assumptions used by cumulus parameterizations.
Mon. Wea. Rev.
,
121
,
764
787
.
Grell
,
G. A.
,
J.
Dudhia
, and
D. R.
Stauffer
,
1994
:
A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR/TN-398+STR, 131 pp
.
Hollinger
,
J.
,
1989
:
DMSP Special Sensor Microwave/Imager calibration/validation. Naval Research Laboratory Final Rep., Vol. 1, 153 pp
.
Hong
,
S-Y.
, and
H-L.
Pan
,
1996
:
Nonlocal boundary layer vertical diffusion in a medium-range forecast model.
Mon. Wea. Rev.
,
124
,
2322
2339
.
Ide
,
K.
,
P.
Courtier
,
M.
Ghil
, and
A. C.
Lorenc
,
1997
:
Unified notation for data assimilation: Operational, sequential and variational.
J. Meteor. Soc. Japan
,
75
,
180
189
.
Jenamani
,
R. K.
,
S. C.
Bhan
, and
S. R.
Kalsi
,
2006
:
Observational forecasting aspects of the meteorological event that caused a record highest rainfall in Mumbai.
Current Sci.
,
90
,
1344
1363
.
Jung
,
J. A.
,
T. H.
Zapotocny
,
R. E.
Treadon
, and
J. F.
Le Marshall
,
2006
:
Atmospheric Infrared Sounder assimilation experiments using NCEP’s GFS. Preprints, 14th Conf. on Satellite Meteorology and Oceanography, Atlanta, GA, Amer. Meteor. Soc., P5.11. [Available online at http://ams.confex.com/ams/pdfpapers/103965.pdf.]
.
Le Marshall
,
J.
, and
Coauthors
,
2005a
:
AIRS hyperspectral data improves Southern Hemisphere forecasts.
Aust. Meteor. Mag.
,
54
,
57
60
.
Le Marshall
,
J.
, and
Coauthors
,
2005b
:
Impact of Atmospheric Infrared Sounder observations on weather forecasts.
Eos, Trans. Amer. Geophys. Union
,
86
.
109, 115, 116
.
Le Marshall
,
J.
, and
Coauthors
,
2005c
:
AIRS associated accomplishments at the JCSDA—First use of full spatial resolution hyperspectral data show significant improvements in global forecasts.
Proc. SPIE
,
5890
.
doi:10.1117/12.615757
.
Le Marshall
,
J.
, and
Coauthors
,
2006
:
Improving global analysis and forecasting with AIRS.
Bull. Amer. Meteor. Soc.
,
87
,
891
894
.
Lipton
,
A. E.
, and
T. H.
Vonder Haar
,
1990
:
Mesoscale analysis by numerical modeling coupled with sounding retrieval from satellites.
Mon. Wea. Rev.
,
118
,
1308
1329
.
Lipton
,
A. E.
,
G. D.
Modica
,
S. T.
Heckman
, and
A. J.
Jackson
,
1995
:
Satellite–model coupled analysis of convective potential in Florida with VAS water vapor and surface temperature data.
Mon. Wea. Rev.
,
123
,
3292
3304
.
McNally
,
A. P.
, and
M.
Vesperini
,
1996
:
Variational analysis of humidity information from TOVS radiances.
Quart. J. Roy. Meteor. Soc.
,
122
,
1521
1544
.
Mlawer
,
E. J.
,
S. J.
Taubman
,
P. D.
Brown
,
M. J.
Lacono
, and
S. A.
Clough
,
1997
:
Radiative transfer for homogeneous atmosphere: RRTM, a validated correlated-k model for the long-wave.
J. Geophys. Res.
,
102
,
D14
.
16663
16682
.
Ruggiero
,
F. H.
,
K. D.
Sashegyi
,
A. E.
Lipton
,
R. V.
Madala
, and
S.
Raman
,
1999
:
Coupled assimilation of geostationary satellite sounder data into a mesoscale model using the Bratseth analysis approach.
Mon. Wea. Rev.
,
127
,
802
820
.
Shirtliffe
,
G.
,
1999
:
QuikSCAT science data products user’s manual. Jet Propulsion Laboratory Publ. D-18053, Pasadena, CA, 90 pp
.
Singh
,
R.
,
P. K.
Pal
,
C. M.
Kishtawal
, and
P. C.
Joshi
,
2008
:
The impact of variational assimilation of SSM/I and QuikSCAT satellite observations on the numerical simulation of Indian Ocean tropical cyclones.
Wea. Forecasting
,
23
,
460
476
.
Srinivasan
,
V.
,
1972
:
Discussion of typical synoptic situations over Konkan and coastal Karnataka.
Forecasting Manual III-3.7, India Meteorological Department, New Delhi, India, 3–7
.
Susskind
,
J.
,
C.
Barnet
, and
J.
Blaisdell
,
2003
:
Retrieval of atmospheric and surface parameters from AIRS/AMSU/HSB under cloudy conditions.
IEEE Trans. Geosci. Remote Sens.
,
41
,
390
409
.
Susskind
,
J.
,
C.
Barnet
,
J.
Blaisdell
,
L.
Iredell
,
F.
Keita
,
L.
Kouvaris
,
G.
Molnar
, and
M.
Chahine
,
2006
:
Accuracy of geophysical parameters derived from Atmospheric Infrared Sounder/Advanced Microwave Sounding Unit as a function of fractional cloud cover.
J. Geophys. Res.
,
111
.
D09S17, doi:10.1029/2005JD006272
.
Tian
,
B.
,
D. E.
Waliser
,
E. J.
Fetzer
,
B. H.
Lambrigtsen
,
Y.
Yung
, and
B.
Wang
,
2006
:
Vertical moist thermodynamic structure and spatial–temporal evolution of the MJO in AIRS observations.
J. Atmos. Sci.
,
63
,
2462
2485
.
Tobin
,
D. C.
, and
Coauthors
,
2006
:
Atmospheric Radiation Measurement site atmospheric state best estimates for Atmospheric Infrared Sounder temperature and water vapor retrieval validation.
J. Geophys. Res.
,
111
.
D09S14, doi:10.1029/2005JD00610
.
Tomassini
,
M. D.
,
X.
LeMeur
, and
R. W.
Saunders
,
1998
:
Near-surface satellite wind observations of hurricanes and their impact on ECMWF model analyses and forecast.
Mon. Wea. Rev.
,
126
,
1274
1286
.
Wu
,
L.
,
S. A.
Braun
,
J. J.
Qu
, and
X.
Hao
,
2006
:
Simulating the formation of Hurricane Isabel (2003) with AIRS data.
Geophys. Res. Lett.
,
33
.
L04804, doi:10.1029/2005GL024665
.
Xiao
,
Q.
,
X.
Zou
, and
Y. H.
Kuo
,
2000
:
Incorporating the SSM/I-derived precipitable water and rainfall rate into a numerical model: A case study for the ERICA IOP-4 cyclone.
Mon. Wea. Rev.
,
128
,
87
108
.
Xiao
,
Q.
,
X.
Zou
,
M.
Pondeca
,
M. A.
Sharpiro
, and
C.
Velden
,
2002
:
Impact of GMS-5 and GOES-9 satellite-derived winds on the prediction of a NORPEX extratropical cyclone.
Mon. Wea. Rev.
,
130
,
507
528
.
Zapotocny
,
T. H.
, and
Coauthors
,
2000
:
A case study of the sensitivity of the Eta Data Assimilation System.
Wea. Forecasting
,
15
,
603
621
.
Zapotocny
,
T. H.
,
W. P.
Menzel
,
J. P.
Nelson
III
, and
J. A.
Jung
,
2002
:
An impact study of five remotely sensed and five in situ data types in the Eta Data Assimilation System.
Wea. Forecasting
,
17
,
263
285
.
Zapotocny
,
T. H.
,
W. P.
Menzel
,
J. A.
Jung
, and
J. P.
Nelson
III
,
2005
:
A four-season impact study of rawinsonde, GOES, and POES data in Eta Data Assimilation System. Part I: The total contribution.
Wea. Forecasting
,
20
,
161
177
.
Zavodsky
,
B. T.
,
S. M.
Lazarus
,
P. F.
Blotiman
, and
D. W.
Sharp
,
2004
:
Assimilation of MODIS temperature and water vapor profiles into a mesoscale analysis system. Preprints, 20th Conf. on Weather Analysis and Forecasting/16th Conf. on Numerical Weather Prediction, Seattle, WA, Amer. Meteor. Soc., 2.5. [Available online at http://ams.confex.com/ams/pdfpapers/68594.pdf.]
.
Zhang
,
X.
,
Q.
Xiao
, and
F.
Patrick
,
2007
:
The impact of multisatellite data on the initialization and simulation of Hurricane Lili’s (2002) rapid weakening phase.
Mon. Wea. Rev.
,
135
,
526
548
.
Zhao
,
Y.
,
B.
Wang
,
Z.
Ji
,
X.
Liang
,
G.
Deng
, and
X.
Zhang
,
2005
:
Improved track forecasting of typhoon reaching landfall from four-dimensional variational data assimilation of AMSU A retrieved data.
J. Geophys. Res.
,
110
.
D14101, doi:10.1029/2004JD005267
.
Zhu
,
T.
,
D. L.
Zhang
, and
F.
Weng
,
2002
:
Impact of Advanced Microwave Sounding Unit measurements on hurricane prediction.
Mon. Wea. Rev.
,
130
,
2416
2432
.

Footnotes

Corresponding author address: Randhir Singh, Atmospheric Sciences Division, Meteorology and Oceanography Group, Space Applications Centre (ISRO), Ahmedabad-380015, India. Email: randhir_h@yahoo.com