1. Introduction
The uncertainties in the initial conditions due to lack of observations are often considered as a major source of error in the numerical weather prediction (NWP) models. This is especially observed during the tropical cyclone (TC) track predictions, where very poor observational network over oceans can lead to errors in forecasts. Operational meteorological satellites, particularly geostationary, have the ability to sample the large-scale atmospheric environment very frequently and use them as tools for monitoring TCs. However, in the late 1980s or early 1990s, the synoptic forecasters of operational agencies were using satellite images only for visual interpretation, the high-density quantitative information from these satellites are now frequently being assimilated into the NWP models to improve the initial conditions, which in turn reduces the forecast errors. Over the past decade, one of the major areas of improvement in the NWP systems has been the more effective assimilation of high-density multispectral Geostationary Operational Environmental Satellite (GOES), Meteosat, and Geostationary Meteorological Satellite (GMS) wind observations (Velden et al. 1992, 1998; Leslie et al. 1998; Goerss et al. 1998).
Velden et al. (1992) showed that the assimilation of satellite-derived winds into the numerical model resulted in 2%–6% reductions in the mean track error. Leslie et al. (1998) demonstrated that the assimilation of high-density satellite-derived winds could significantly improve the track forecasts in a high-resolution model, although the study was limited to only two cases. Goerss et al. (1998) examined the impact of an experimental high-density GOES wind product on hurricane track predictions from the Navy Oceanographic Global Atmospheric Prediction System (NOGAPS) forecast model. Based upon their study for four tropical Atlantic storms in the year 1995, they showed that assimilation of the experimentally derived GOES winds reduced the track errors in the NOGAPS model by 12%–14%. Consequently, operational assimilation of these winds into the NOGAPS model was initiated in 1996. The Geophysical Fluid Dynamics Laboratory (GFDL) Hurricane Prediction System model (Kurihara et al. 1998) is one of the most sophisticated models for predicting hurricane tracks over the Atlantic. In contrast to the NOGAPS model, the GFDL model is a limited-area, multiply nested model designed specifically for TC prediction. Soden et al. (2001) had shown using this model that on average, assimilation of the GOES winds leads to statistically significant improvements for all forecast periods, with the relative reductions in track error ranging from 5% at 12 h to 12% at 36 h.
The initiation of recent operational derivation of water vapor winds (Kishtawal et al. 2009) from the water vapor imagers from Indian geostationary satellite Kalpana-1 at the Space Applications Centre (SAC), Ahmedabad, India, has given us an opportunity to examine the impact of these newly derived satellite water vapor winds for the track and intensity prediction of two Bay of Bengal TCs, Sidr and Nargis, using a limited-area model. The limited-area model used in this study is the Advanced Research Weather Research and Forecasting (ARW-WRF) modeling system. The performance of Kalpana-1 winds is evaluated by repeating the experiments using Meteosat-7 winds derived at the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) as well as GFS analysis to see the impact assimilating all observations. In this study, a number of experiments were conducted from different initial conditions (with and without Kalpana-1 and Meteosat-7 derived winds) for both TCs. Section 2 provides a brief description of the WRF model assimilation method used in the model and satellite winds used for this study. A very brief summary of TCs Sidr and Nargis along with experimental design is given in section 3. Section 4 presents the results of each TC along with statistical analysis, while summary and conclusions from the present study are given in section 5.
2. Description of model, assimilation system, and satellite winds
a. Forecast model: WRF modeling system
The forecast model used is the WRF-ARW (Skamarock et al. 2005) model, version 2.2, (see online at http://www.wrf-model.org). This mesoscale numerical model is designed to serve both operational forecasting and atmospheric research needs. It is a limited-area, nonhydrostatic primitive equation model with multiple options for various physical parameterization schemes. The present version of the WRF model employs Arakawa C-grid staggering for horizontal grid and a fully compressible system of equations. The terrain-following hydrostatic pressure with vertical grid stretching was followed in vertical. The time-split integration uses third-order Runge–Kutta scheme with smaller time step for acoustic and gravity wave modes. Physics options used in this study include the Kain–Fritsch (Kain and Fritsch 1990, 1993) cumulus parameterization scheme and the WRF Single-Moment 6-class graupel (WSM6) microphysics scheme. The planetary boundary layer is parameterized using the advanced version (nonlocal gradient) of the medium-range forecast (MRF) model planetary boundary layer (PBL) scheme (Hong and Pan 1996), and for soil model the multilayer Noah land surface model (LSM) is used. The longwave radiation scheme is based on the Rapid Radiative Transfer Model (RRTM) and shortwave radiation is based on Dudhia (1989).
b. Assimilation system
The WRF three-dimensional variational data assimilation system, WRF 3D-Var (Skamarock et al. 2005) was used in the present study. The WRF 3D-Var is originated and evolved from the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) 3D-Var (Barker et al. 2004), but the basic software interface and coordinate framework are fully updated for the WRF model. The background covariances matrix was estimated using the so-called NMC method (Parrish and Derber 1992; Wu et al. 2002) provided in WRF-VAR package. The observation errors were assumed to be uncorrelated in space and time. Since observation errors were assumed uncorrelated, the observational error covariances matrices were simply diagonal with satellite water vapor wind observation error variances as elements. In this study, observation errors were taken as constant in space and time. The horizontal resolution of both Kalpana-1 and Meteosat-7 water vapor winds is 160 km. The observation error for Meteosat-7 water vapor winds is approximately 6.5 m s−1 (EUMETSAT 2009, personal communication), which is the same error assumed for Kalpana-1 water vapor winds as well for this study. Prior to data assimilation, the satellite winds underwent quality-checking processes in order to reduce the possibility of assimilating bad observations. A gross error quality control was performed in which observations that differed from the model first guess by more than 5 times the observational errors were removed.
c. Satellite winds
This section provides a brief description of the estimation methodology used to derive the water vapor winds from Kalpana-1 (at SAC) as well as Meteosat-7 (at EUMETSAT). Since Kalpana-1 water vapor wind is a very new satellite-derived product, its methodology is described in a bit more detail, while for Meteosat-7, the description is very sparse.
1) Kalpana-1 water vapor winds
2) Meteosat-7 water vapor winds
At EUMETSAT water vapor winds are extracted operationally (Holmlund 1995) on an equidistant grid (baseline 32 × 32 pixels) with a target size equivalent to the grid size from first generation Meteosat-5/Meteosat-7 imagers over the Indian Ocean region. The algorithm consists of several steps: in the first step, the tracer is used for the extraction of features. The tracers in water vapor images are chosen by considering the medium- and high-level cloud cluster, which has the coldest (lowest) WV mean count. The height assignment is also performed during this step. The cloud brightness temperature are then obtained using the inverse Plank’s function and were compared with the temperature soundings from the forecast from the operational NWP models to obtain the height of the derived wind fields. The determination of the height of WV image tracers is similarly performed but based on the potentially semitransparency-corrected WV mean count. It is to be noted that an inverse atmospheric absorption correction is not carried out, which might be a deficiency. The maximum cross-correlation (MCC) technique is used operationally for tracking features both infrared and water vapor channels at EUMETSAT (Rattenborg 2000). The height assignment of water vapor winds is done using the single-level height assignment based on the cluster equivalent blackbody temperature (EBBT) method. Another method based on WV contribution function calculated from the radiative transfer model is also used to calculate the WV tracer’s height (Rattenborg and Holmlund 1996).
3. Overview of numerical experiments and data
Two Bay of Bengal TCs, Sidr and Nargis, are considered for the present study. TC Sidr developed in the Bay of Bengal on 9 November 2007 and intensified to a very severe cyclonic storm, with maximum wind approximately 75 m s−1 on 14 November 2007. Sidr made a landfall at the coast of Bangladesh at 1200 UTC 15 November 2007 (Fig. 1) with maximum sustained winds near 240 km h−1 (150 mph) and minimum sea level pressure (MSLP) of about 918 hPa, producing heavy rains and high tidal surges that caused widespread flooding. On 2 May 2008, TC Nargis (Fig. 1) caused the deadliest natural disaster in the recorded history of Myanmar, causing catastrophic destruction and at least 80 000 fatalities with an additional 56 000 people still missing. In the last week of April 2008, Nargis originated over an area of deep convection that persisted near a low-level circulation in the Bay of Bengal. With favorable outflow and low wind shear, the system slowly organized as a TC. On 1 May, TC Nargis began rapidly intensifying as a result of greatly improved outflow in association with an approaching upper-level trough. A well-defined eye with a diameter of 19 km developed early on 2 May. IMD estimated the peak speed of Nargis at 165 km h−1 as it approached Myanmar coast. On the same day, at around 1200 UTC it made landfall in the Ayeyarwady Division of Myanmar.
The numerical experiments were conducted from different initial conditions for the simulation of two tropical TCs using the WRF model. For each initialization time four sets of experiments were conducted: first a control experiment (Con), another experimental runs where Kalpana-1 water vapor (WvK) winds are assimilated for both TCs. To see the performance of Kalpana-1 winds, similar sets of experiments (WvM) were repeated using the Meteosat-7 water vapor winds derived at EUMETSAT and the fourth analysis experiments (Ana) where GFS analysis is used as initial conditions to see the impact of all other observations except the water vapor winds, in the National Centers for Environmental Prediction (NCEP) data assimilating system. Since the experiments were conducted on hindcast mode, the selection of initial conditions was based upon the simultaneous availability of Kalpana-1 and Meteosat-7 winds.
The different experiments performed for this study are given in Table 1. For the Con experiments 1° × 1° NCEP Global Forecast System (GFS) 6-h forecast (valid for a particular time when satellite winds were available) were used for the preparation of initial conditions, while in case of Ana NCEP GFS analysis were used. In case of WvK and WvM experiments 6-h GFS forecast interpolated to the model grid points were used as first guess (FG) and the satellite winds are assimilated one time in the outer domain for each experiment for preparing the analysis. However, the NCEP Global Data Assimilation System (GDAS) analysis with 1° × 1° resolution was used for the model boundary conditions. The 6-h forecast was used as the FG for a specific reason. As it is well known that Meteosat-7 water vapor winds (see online at http://www.emc.ncep.noaa.gov/mmb/data_processing/prepbufr.doc/table_2.htm) and Kalpana-1 winds do not used in preparing NCEP GDAS analysis via data assimilation system. Thus a 6-h GFS forecast creates an ideal test bed for assessing the impacts of satellite winds. In the case of Sidr, the numerical experiments were conducted from three different initial conditions (viz., 0600 UTC of 11, 12, and 14 November 2007, respectively, and as a total 12 experiments was conducted). While for Nargis six different initial conditions (viz., 0000 and 0600 UTC of 29, 30 April, and 1 May 2008, respectively, were used and a total of 24 experiments were conducted). During the period of two TCs, Kalpana-1 winds were estimated only 2 times a day (viz., 0000 and 0730 UTC). For the experiments with 0000 UTC analysis, the Kalpana-1 winds derived at 0000 UTC are assimilated, while for the experiments starting with 0600 UTC analysis, the nearest Kalpana-1 winds derived at 0730 UTC are utilized for assimilation. All the experiments were carried out over the identical regions and double-nested configurations, with 45-km outer and 15-km inner domain horizontal resolutions having 120 × 120 grids in the outer and 160 × 160 grids in the inner domains, covering the regions 8.7°S–36.3°N, 61.12°–108.9°E and 4.9°–25.4°N, 80.0°–101.3°E, respectively (Fig. 2). In the vertical, the model has 31 vertical layers with the top model layer at 50 hPa.
The 48-h simulations were performed for each experiment and the track positions (minimum MSLP point) were collected at every 6-h interval. The 6-h track positions and intensities (maximum surface wind speed and magnitude of MSLP) were then compared with the Joint Typhoon Warning Center’s (JTWC) best-track analysis for tracks and intensity for forecast verification. A typical examples of satellite-derived water vapor winds from Kalpana-1 and Meteosat-7 valid at 0600 UTC 14 November 2007 are shown in Fig. 3. As the satellite-derived water vapor winds are generally in the upper troposphere (i.e., in between 500 and 100 hPa), it is clearly visible from the figure that both satellites have produced wind vectors with uniform coverage, large-scale and synoptic-scale features are well captured, and vertical distribution of information is in the upper troposphere.
4. Results
In section 4a, we provide an analysis of the impact of the assimilation of WV winds on the model initial conditions. The impact of these assimilated initial conditions on the forecast of track and intensity from the inner domain (15 km) of the two TCs are presented in sections 4b,c.
a. Analysis of assimilation of WV winds on model initial conditions
The impact of the satellite winds on the model forecasts is determined by the analyzed initial conditions (viz., maximum surface winds, magnitude of MSLP for initial intensity error, and central position of MSLP for initial position error) from the different experiments were compared with observed values. Since there are a total of 36 experiments (including TCs Sidr and Nargis), each with differing synoptic conditions, it is difficult to analyze each case individually and identify consistent features. Instead initial conditions are analyzed by calculating the averages of initial TC position error, the average of MSLP differences, and the surface wind differences for each TC.
The initial vortex position error, surface wind difference, and MSLP difference from the observed values for the different experiments for TC Sidr are shown in Fig. 4a. Here averages are computed based on three different initial conditions. It is clearly visible from the figure that initial position errors have improved after assimilating WV winds in WvK and WvM when compared with Con experiment. The initial position error in Con is 62.3 km, while for Ana, WvK, and WvM these are 44.3, 40.1, and 51.9 km, respectively. This shows the assimilation of observations, which goes to NCEP GFS assimilating system for making GFS analysis, (in case of Ana) and the assimilation of WV winds only (both WvK and WvM experiments) has made positive impacts on the initial position of vortex when compared with Con experiments. The impact of WvM is smaller on vortex positioning compared to WvK experiments. However, the assimilation of satellite winds (both Kalpana-1 and Meteosat-7) has not shown any impact on initial intensities (i.e., in the initial surface wind and MSLP).
The initial position error, surface wind differences and MSLP differences from the observed values for different experiments for Nargis are shown in Fig. 4b. It is clearly visible from the figure that positions of vortices have improved after the assimilation of satellite winds. The initial position error in the Con is 100.5 km, while for Ana, WvK, and WvM these are 98.3, 74.3, and 76.5 km, respectively. The assimilation of WV winds has made positive impact in reducing the initial position error when compared with corresponding Con experiments for both satellites. However, the impacts of both satellites are the same for intersatellite comparisons. It is also observed that assimilation of satellite winds did not show any impact on initial intensities (i.e., in the initial surface wind and MSLP). The average of initial maximum surface wind difference for Con experiment is 21.8 m s−1 while in cases of Ana, WvK, and WvM these are 20.2, 19.7, and 19.8 m s−1, respectively.
A typical example of the spatial pattern of initial sea level pressure for the different experiments valid at 0600 UTC 14 November 2007 for TC Sidr is shown in Fig. 5. It clearly shows the impact of satellite winds and the minimum sea level pressure contour in Con, WvK, and WvM is 1002 hPa, while for Ana it is 1001 hPa. Similarly, the Fig. 6 shows a typical example of surface wind from the initial condition of the Con experiment and differences from Con for the other experiments valid at 0600 UTC 14 November 2007 for TC Sidr. The use of GFS analysis as the initial condition (Fig. 6b) has strengthened the cyclonic flow around the center, while this is not happened in the case of WvK (Fig. 6c) and WvM (Fig. 6d) experiments.
b. Track forecasts
The simulated tracks for both TCs were determined from the center of MSLP contours and compared with best-track analysis from JTWC. The observed and predicted tracks for TC Sidr starting from 0600 UTC 14 November 2007 from the Con, Ana, WvK, and WvM are shown in Fig. 7a. It is noticed from the figure that, the forecasts for 12- to 42-h from WvK have improved compared to Con and Ana when Kalpana-1 WV winds were assimilated. However beyond 42-h WV winds have no positive impact. In the case of WvM experiments, the forecast shows improvement over Con only up to 24 h. The experiment Ana has also not shown any improvement in the track forecast as well. The similar results were also noticed in the predicted tracks from the other initial conditions (not shown here) as well.
The observed and predicted tracks for TC Nargis starting from 0600 UTC 30 April 2008 for Con, Ana, WvK, and WvM are shown in Fig. 7b. It is observed that, the forecasts from 12 to 36 h have improved in WvK experiments when compared to Con and Ana. However, beyond that, WV winds have no positive impact for 42- and 48-h forecasts. In most of the cases the forecasts have improved by assimilation of Meteosat-7-derived WV winds compared to Con and Ana experiments. Experiment Ana has also not shown any improvement in the track forecast as well. It is also observed that most of the time the tracks from Con and Ana are following each other. The similar results were also noticed in the predicted tracks from the other initial conditions (not shown here) as well.
The mean track errors for every 6-h forecast from all four sets of experiments combining both TCs together are shown in Fig. 8. The mean track error for the 24-h forecast in WvK and WvM are 94.6 and 68.7 km, respectively; while for Con and Ana, the errors are 136.8 and 104.7 km, respectively. In 48-h forecast, the mean track errors in WvK and WvM are 139.4 and 173.8 km, respectively; while for Con and Ana, the errors are 245.7 and 169.3 km, respectively. The Student’s t test is also calculated to find the confidence intervals on whether assimilating the satellite winds has an impact on the TC forecasts. However, because of a relatively small number of cases after combining both TCs (i.e., total of 9 samples: 3 for Sidr and 6 for Nargis); the level of statistical confidence in the results is low. The improvement in 18-h forecasts for both WvK and WvM, 24-h forecast in WvM, and 48-h forecast in WvK are statistically significant at the 90% confidence level, while all other forecasts are statistically significant at the level of the 75% confidence level. Thus, the assimilation of winds, either from the Kalpana-1 or Meteosat-7, has lead to some improvement in the track forecasts for both TCs, even with this very limited number of samples.
A quantitative evaluation of the impact of satellite winds on WRF model forecast separately for TCs Sidr and Nargis is presented in Table 2. The relative forecast errors were computed for Ana, WvK, and WvM with respect to Con for both TCs are shown in the bracket. For TC Sidr, the assimilation of satellite winds reduces the mean track errors at all forecast time, with the largest relative reductions occurring for the 48-h (54%) forecast for Ana, the 48-h (51%) forecast for WvK, and the 24-h (59%) forecast for WvM experiments. Similarly, for TC Nargis, the largest relative reductions occurring for the 18-h (48%) forecast for Ana, the 48-h (35.9%) forecast for WvK, and the 24-h (42.6%) forecast for WvM experiments.
c. Intensity forecast
The temporal variation of the observed central MSLP and maximum surface winds and those from the simulations are compared for analyzing the intensity forecast. The temporal distribution of the observed and simulated MSLP from the three different initial conditions for all experiments for TC Sidr are shown in Fig. 9. The observed MSLP was 998 hPa at the model initial time (0600 UTC 11 November 2007) which dropped to a lowest value of 918 hPa at 1800 UTC 14 November 2007. This lowest value was sustained until 0600 UTC 15 November 2007 and subsequently increased to 978 hPa at 0000 UTC 16 November 2007. The initial intensity for all the experiments (viz., Con, Ana, WvK, and WvM experiments) starting at 0600 UTC 11 November 2007 for both satellites (Kalpana-1 and Meteosat-7) are slightly weaker than the observed intensity. It is found from the figure that the MSLP for the experiments were reduced to 963 hPa for WvK and 965 hPa for WvM after 48 h of integration at 0600 UTC 13 November 2007, which are slightly smaller as compared Con experiments. The MSLP for Ana is 970 hPa after 48-h integration, which is slightly higher than Con. However, the observed MSLP at that time was 937 hPa. Similar biases are also noticed for all the experiments stating at 0600 UTC 12 November 2007 for both satellites. The MSLP for Ana, WvK, and WvM winds experiments are reduced to 970, 953, and 965 hPa at 0600 UTC 14 November 2007 after 48 h of integration, while the observed MSLP was 933 hPa. The initial intensities for all the experiments starting at 0600 UTC 14 November 2007 were weaker as compared to the observed value.
The temporal distribution of the observed and simulated maximum surface winds from the three different initial conditions for all experiments for TC Sidr are shown in Fig. 10. The observed maximum surface wind was 17 m s−1 at the model initial time (0600 UTC 11 November 2007) and peaked to its highest value 70 m s−1 at 1800 UTC 14 November 2007. This highest value was sustained until 0600 UTC 15 November 2007 and subsequently decreased to 30 m s−1 at 0000 UTC 16 November 2007. All the experiments starting at 0600 UTC 11 November 2007 attained more than 52 m s−1 maximum surface winds at 0600 UTC 13 November 2007. The observed value at that time was 58 m s−1. Though the initial maximum surface wind was weak as compared to the observed value for all the experiments starting at 0600 UTC 12 November 2007, however after 48 h of integration it peaked at around 50–55 m s−1. At that time the observed maximum surface wind was 60 m s−1. The initial maximum surface winds for all the experiments starting at 0600 UTC 14 November 2007 were very weak as compared to the observed value and the impact of assimilation of satellite WV winds in these experiments was not there when compared with other initial conditions experiments. The Student’s t test is also calculated to find the confidence intervals on whether assimilating the satellite winds has any positive impact on the TC intensity forecasts. It is found that the intensity forecast improvement is statistically not significant, with this very limited number of samples.
The temporal distribution of the observed and simulated MSLP from the six different initial conditions for all the experiments for TC Nargis are shown in Fig. 11. The top panel shows the experiments for which 0000 UTC is the initial conditions, while the bottom panel shows the experiments for 0600 UTC. The observed MSLP was 979 hPa at the model initial time (0000 UTC 29 April 2008) and dropped to the lowest value of 929 hPa at 1200 UTC 2 May 2008. This lowest value was sustained until 1800 UTC 2 May 2008 and subsequently increased to 955 hPa at 0600 UTC 3 May 2008. The initial intensity for all the experiments (viz., Con, Ana, WvK, and WvM experiments) starting at 0000 UTC 29 April 2008 were weaker than the observed intensity (Fig. 11a). It is observed from the figures that MSLP for WvK, Ana, and WvM experiments were reduced to 980, 962, and 989 hPa, respectively, after 48 h of integration at 0000 UTC 1 May 2008, which were less compared to Con experiments. However, the observed MSLP at that time was 984 hPa. Similar observations were also noticed for all the experiments stating at 0600 UTC 29 April 2008 (Fig. 11b). The MSLP for Ana, WvK, and WvM experiments were reduced to 967, 968, and 974 hPa at 0600 UTC 1 May 2008 after 48 h of integration, while the observed MSLP was 979 hPa. The initial intensity for all the experiments starting at 0000 and 0600 UTC 30 April 2008 were weak as compared to observed value; however, the reduction of MSLP during 48 h of integration was more in the simulation compared to the observed value. It is also noticed that assimilation of WV winds lead to higher intensification of the TC compared to Con experiments. Similar observations were also noticed for all the experiments starting at 0000 and 0600 UTC 1 May 2008.
The temporal distribution of the observed and simulated maximum surface winds from the six different initial conditions for all experiments for TC Nargis are shown in Fig. 12. The top panel shows the experiments for which 0000 UTC is the initial conditions, while the bottom panel shows the experiments for 0600 UTC. The observed maximum surface wind was 43 m s−1 at the model initial time (0000 UTC 29 April 2008) and peaked to its highest value 58 m s−1 at 0600 UTC 2 May 2008 and subsequently decreased to 25 m s−1 at 0600 UTC 3 May 2008. All experiments starting at 0000 and 0600 UTC 29 April 2008 attained more than 38–45 m s−1 maximum surface winds at 0000 and 0600 UTC 1 May 2008, respectively. The observed values at these times were 35 and 38 m s−1, respectively. Though the initial maximum surface wind was weak as compared to observed for all the experiments starting at 0000 and 0600 UTC 30 April 2008, after 48 h of integration it peaked at around 55–60 m s−1. While the observed values these times were 50 and 58 m s−1, respectively. The initial maximum surface winds for all the experiments starting at 0000 and 0600 UTC 1 May 2008 were very weak as compared to observed value; however, in most cases, it has peaked up winds about 55–60 m s−1 after 48 h of integration. The observed values at these times were 35 and 25 m s−1, respectively. Thus most of the experiments have simulated the observed features quite well.
The mean difference of simulated MSLP and maximum surface winds from the observed values are shown in Figs. 13a,b. The differences are calculated for all the experiments based on 6-h model outputs combining both TCs together. It is clearly noticed that the forecast improvement in the simulations of MSLP and surface winds in WvK and WvM (when WV winds were used for assimilation) are statistically not significant when compared with Con experiments. However, in case of intersatellite comparison WvK has slightly better impact as compared to WvM experiment. The evaluations demonstrate that the assimilation of satellite winds does not lead to the improvement in intensity forecast. Given the distribution of WV winds (Fig. 3), it is not surprising that assimilating these winds has any impact on the intensity. Most of these winds are near the perimeter of the storm, which is not likely to have any information on the intensity of the storm.
The deep layer mean (DLM) wind can also be analyzed to understand the impact of satellite winds on the large-scale flow fields. The DLM is defined as the vertical pressure–weighted average of the initial condition. The DLM wind from a particular experiment valid at 0600 UTC 12 November 2007 from the Con, Ana, WvK, and WvM experiments are examined. Figure 14a shows the DLM wind from the Con; the differences in DLM wind in the other experiments from the Con due to assimilation of satellite winds are in other figures (Figs. 14b–d). The DLM winds are represented as vectors and contour gives the magnitude of flow, while vorticity of DLM winds is represented by shading. The vorticity fields provide a reliable interpretation of the vortex–environment interactions. In the Con experiment DLM flow shows a cyclonic circulation (Fig. 14a) around the TC center. A dipole structure is seen both in wind and vorticity field when difference figure (Ana − Con) is calculated (Fig. 14b), with anticyclonic flow to the west and cyclonic flow to the east of the center. The vorticity field has also shown a dipole structure. Similarly, a northwest–southeast dipole is seen in the vorticity field (Fig. 14c), when the difference between WvK and Con is calculated. The DLM flow in the WvK experiment reveals weak cyclonic flow around the TC center. This shows that assimilation Kalpana-1 WV wind reduces the strength of overanalyzed flow present in the first-guess field around the TC center. However, no dipole is seen in the WvM experiment (Fig. 14d) both in the flow as well as in the vorticity field. The assimilation of Meteosat-7 WV wind decreases the strength of the DLM flow in the north, while it increases to the south of the storm center. It may also be noticed that assimilating winds from Kalpana-1 has the net effect of adding a northeasterly wind component to the analysis near the TC, while the Meteosat-7 winds lead to a southwesterly wind on top of the TC; this could be due to the difference of magnitude of the retrieved winds from Kalpana-1 and Meteosat-7, when compared with the first-guess field. It is visible in the distribution of WV wind (Fig. 3) that the strength of the vortex around the TC is higher in the retrieved winds from Kalpana-1 than the Meteosat-7-derived winds. This may be due to different methods of target selection and tracking procedure adopted for retrieving winds from Kalpana-1 and Meteosat-7.
5. Summary and conclusions
In this study an attempt has been made to demonstrate the impact of satellite-derived water vapor winds from the Indian geostationary satellite Kalpana-1 for the track and intensity prediction of two Bay of Bengal TCs, Sidr and Nargis, using the WRF model. To test the performance of Kalpana-1 winds, the same sets of experiments were repeated using Meteosat-7 winds and compared with observed data. Similarly another experiment was conducted using GFS analysis as initial conditions to see the impact of assimilation of all observations. It was found that initial position errors were less in WvK and WvM experiments when compared with the Con experiment for both TCs. The assimilation of WV winds has made positive impact in the initial position error when compared with Con for both satellites. However, in case of intersatellite comparisons Meteosat-7 have less impact on the initial position error than Kalpana-1 winds for the both TCs, while the assimilation of satellite winds (both Kalpana-1 and Meteosat-7) has not shown any large changes in the initial intensities (i.e., initial surface wind and MSLP).
The mean track error forecast from WvK and WvM has shown improvement compared to Con or Ana for all forecasts. In case of WvK and WvM (i.e., when WV winds were used for assimilation), the intensity forecast improvement are statistically not significant when compared with Con or Ana experiments for both TCs. In the simulation of maximum surface winds, WvK performed slightly better compared to the corresponding Con experiment. Thus, the analysis has demonstrated that assimilation of Kalpana-1 or Meteosat-7 WV winds has lead to some improvement in the track forecast; however, there was not any improvement in intensity forecasts for TCs Sidr and Nargis, with some exceptions. It is not surprising that assimilating WV winds has any impact on the intensity, as most of these winds are near the perimeter of the storm, which is not likely to have any information on the intensity of the storm. As the operational derivational of Kalpana-1 winds will evolve over time, assessing the impact of these winds will require continuous evaluation. The present results are very preliminary and are limited to only two TCs. Innovative upgradation of wind retrieval techniques, new approaches of data assimilation, and experimentation with large number of test cases are some of the ways by which the impact of the assimilation of satellite data on model forecast can be improved further. However, the present study can provide some insight to the WRF model users over the Indian Ocean region. More work is also needed to better understand the height assignments of satellite-derived winds as well.
Acknowledgments
The authors thank the two anonymous reviewers for their critical and insightful comments/valuable suggestions, which were helpful in substantially improving the content and quality of this manuscript. The authors acknowledge the use of WRF model, which is available online at the National Center for Atmospheric Research (NCAR), and the WRF User Support Group sites for useful suggestions during the model installation. The authors are also thankful to EUMETSAT for providing the derived wind products during TCs Sidr and Nargis from Meteosat-7 for assimilation. The encouragement and help from the director and the deputy director of RESA and the head of ASD/MOG/RESA of the Space Applications Centre, ISRO, Ahmedabad, India, are also gratefully acknowledged.
REFERENCES
Barker, D. M. , W. Huang , Y-R. Guo , A. Bourgeois , and X. N. Xiao , 2004: A three-dimensional variational data assimilation system for MM5: Implementation and initial results. Mon. Wea. Rev., 132 , 897–914.
Deb, S. K. , C. M. Kishtawal , P. K. Pal , and P. C. Joshi , 2008: A modified tracers selection and tracking procedure to derive winds using water vapor imagers. J. Appl. Meteor. Climatol., 47 , 3252–3263.
Dudhia, J. , 1989: Numerical study of convection observed during the winter monsoon experiment using a meso-scale two-dimensional model. J. Atmos. Sci., 46 , 3077–3107.
Goerss, J. S. , C. S. Velden , and J. D. Hawkins , 1998: The impact of multispectral GOES-8 wind information on Atlantic tropical cyclone forecasts in 1995. Part II: NOGAPS forecasts. Mon. Wea. Rev., 126 , 1219–1227.
Holmlund, K. , 1995: Half hourly wind data from satellite derived water vapour measurements. Adv. Space Res., 16 , 59–68.
Holmlund, K. , 1998: The utilization of statistical properties of satellite-derived atmospheric motion vectors to derive quality indicators. Wea. Forecasting, 13 , 1093–1104.
Hong, S. Y. , and H. L. Pan , 1996: Non-local boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124 , 2322–2339.
Kain, J. S. , and J. M. Fritsch , 1990: A one dimensional entraining/detraining plume model and its application in convective parameterization. J. Atmos. Sci., 47 , 2784–2802.
Kain, J. S. , and J. M. Fritsch , 1993: Convective parameterization of mesoscale models: The Kain–Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 24, Amer. Meteor. Soc., 165–170.
Kishtawal, C. M. , S. K. Deb , P. K. Pal , and P. C. Joshi , 2009: Estimation of atmospheric motion vectors from Kalpana-1 imagers. J. Appl. Meteor. Climatol., 48 , 2410–2421.
Kurihara, Y. , R. E. Tuleya , and M. A. Bender , 1998: The GFDL hurricane prediction system and its performance in the 1995 hurricane season. Mon. Wea. Rev., 126 , 1306–1322.
Leslie, L. M. , J. F. LeMarshall , R. P. Morison , C. Spinoso , R. J. Purser , N. Pescod , and R. Seecamp , 1998: Improved hurricane track forecasting from the continuous assimilation of high-quality satellite wind data. Mon. Wea. Rev., 126 , 1248–1257.
Nash, J. E. , and J. V. Sutcliffe , 1970: River flow forecasting through conceptual models. Part I: A discussion of principles. J. Hydrol., 10 , 282–290.
Parrish, D. F. , and J. C. Derber , 1992: The National Meteorological Center’s Spectral Statistical Interpolation analysis system. Mon. Wea. Rev., 120 , 1747–1763.
Rattenborg, M. , 2000: Operational Meteosat wind products towards MSG. EUM P28, Proc. Fifth Int. Winds Workshop, Lorne, Australia, EUMETSAT, 37–46.
Rattenborg, M. , and K. Holmlund , 1996: Operational wind products from new Meteosat Ground Segment. Proc. Third Int. Winds Workshop, Ascona, Switzerland, EUMETSAT, 53–59.
Skamarock, W. C. , J. B. Klemp , J. Dudhia , D. O. Gill , D. M. Barker , W. Wang , and J. G. Powers , 2005: A description of the Advanced Research WRF version 2. NCAR Tech Note NCAR/TN-468+STR, 88 pp. [Available from UCAR Communications, P.O. Box 3000, Boulder, CO 80307].
Soden, J. B. , C. S. Velden , and R. E. Tuleya , 2001: The impact of Satellite Winds on Experimental GFDL hurricane model forecasts. Mon. Wea. Rev., 129 , 835–852.
Velden, C. S. , C. M. Hayden , W. P. Menzel , J. L. Franklin , and J. S. Lynch , 1992: The impact of satellite-derived winds on numerical hurricane track forecasting. Wea. Forecasting, 7 , 107–118.
Velden, C. S. , T. L. Olander , and S. Wazong , 1998: The impact of multispectral GOES-8 wind information on Atlantic tropical cyclone track forecasts in 1995. Part I: Dataset methodology, description, and case analysis. Mon. Wea. Rev., 126 , 1202–1218.
Wu, W-S. , R. J. Purser , and D. F. Parrish , 2002: Three-dimensional variational analysis with spatially inhomogeneous covariances. Mon. Wea. Rev., 130 , 2905–2916.
Best tracks analyzed by JTWC for TCs Sidr and Nargis.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
Two nested domains used in WRF model simulations. Spatial resolutions are 45 and 15 km for domains 1 and 2, respectively.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
A typical example of satellite-derived water vapor winds from (a) Kalpana-1 valid at 0730 UTC and (b) Meteosat-7 valid at 0600 UTC 12 Nov 2007. The height of these vectors are between 500 and 100 hPa.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The average initial position error, maximum surface wind and MSLP difference from the observed value for the different experiments in TCs (a) Sidr and (b) Nargis.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The typical examples of sea level pressure from the initial conditions of different experiments: (a) Con, (b) Ana, (c) WvK, and (d) WvM valid at 0600 UTC 14 Nov 2007 for TC Sidr.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The typical example of surface wind from the initial condition of the experiments: (a) Con, and the difference of winds from the other experiments from the control run: (b) Con − Ana, (c) Con − WvK, and (d) Con − WvM valid at 0600 UTC 14 Nov 2007 for TC Sidr.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The observed and predicted tracks for (a) TC Sidr starting at 0600 UTC 14 Nov 2007 and (b) TC Nargis at 0600 UTC 30 Apr 2008.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The mean track errors for every 6 h from all the experiments combining TCs Sidr and Nargis together.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The temporal distribution of the observed and simulated MSLP from the three different initial conditions for TC Sidr from all experiments.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The temporal distribution of the observed and simulated maximum surface winds from the three different initial conditions for TC Sidr from all experiments.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The temporal distribution of the observed and simulated MSLP from the different experiments: (a) 0000 UTC and (b) 0600 UTC as the initial times for TC Nargis.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The temporal distribution of the observed and simulated maximum surface winds from the different experiments: (a) 0000 UTC and (b) 0600 UTC as the initial times for TC Nargis.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
(a) The mean difference of MSLP from the observed for all experiments combining TCs Sidr and Nargis. (b) As in (a), but for maximum surface winds.
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
The large-scale steering flow from (a) the Con experiments and difference due to assimilations, (b) Ana − Con, (c) WvK − Con, and (d) WvM − Con. Vectors denote the deep layer mean winds (m s−1) and shading denotes the corresponding vorticity field (10−6 s−1).
Citation: Monthly Weather Review 138, 3; 10.1175/2009MWR3041.1
Design of experiments.
Comparison of average track errors (km) from Con, Ana, WvK, and WvM forecasts for TCs Sidr and Nargis. The relative forecast errors with respect to Con are shown in the brackets.
