Impact of Doppler weather radar data on numerical forecast of Indian monsoon depressions

Transcription

1 Quarterly Journal of the Royal Meteorological Society Q. J. R. Meteorol. Soc. (2010) Impact of Doppler weather radar data on numerical forecast of Indian monsoon depressions A. Routray, a U. C. Mohanty, a *S.R.H.Rizvi, b Dev Niyogi, c Krishna K. Osuri a and D. Pradhan d a Centre for Atmospheric Sciences, Indian Institute of Technology Delhi, India b National Centre for Atmospheric Research, MMM Division, Boulder, Colorado, USA c Purdue University, West Lafayette, Indiana, USA d Regional Meteorological Centre, IMD, Kolkata, India *Correspondence to: U. C. Mohanty, Centre for Atmospheric Sciences, Indian Institute of Technology Delhi, Hauz Khas, New Delhi , India. ucmohanty@gmail.com This work is a first assessment of utilizing Doppler Weather Radar (DWR) radial velocity and reflectivity in a mesoscale model for prediction of Bay of Bengal monsoon depressions (MDs). The Weather Research Forecasting (WRF) modelling system Advanced Research version (ARW) is customized and evaluated for the Indian monsoon region by generating domain-specific Background Error (BE) statistics and experiments involving two assimilation strategies (cold start and cycling). The monthly averaged 24 h forecast errors for wind, temperature and moisture profiles were analysed. From the statistical skill scores, it is concluded that the cycling mode assimilation enhanced the performance of the WRF threedimensional variational data assimilation (3DVAR) system over the Indian region using conventional and non-conventional observations. DWR data from a coastal site were assimilated for simulation of two different summer MDs over India using the WRF-3DVAR analysis system. Three numerical experiments (control without any Global Telecommunication System (GTS) data, with GTS, and GTS as well as DWR) were performed for simulating these extreme weather events to study the impact of DWR data. The results show that even though MDs are large synoptic systems, assimilation of DWR data has a positive impact on the prediction of the location, propagation and development of rain bands associated with the MDs. All aspects of the MD simulations such as mean-sea-level pressure, winds, vertical structure and the track are significantly improved due to the DWR assimilation. Study results provide a positive proof of concept that the assimilation of the Indian DWR data within WRF can help improvethe simulation of intense convective systems influencing the large-scale monsoonal flow. Copyright c 2010 Royal Meteorological Society Key Words: variational data assimilation; cycling mode; background error Received 22 December 2008; Revised 26 April 2010; Accepted 13 June 2010; Published online in Wiley Online Library Citation: Routray A, Mohanty UC, Rizvi SRH, Niyogi D, Osuri KK, Pradhan D Impact of Doppler weather radar data on numerical forecast of Indian monsoon depressions. Q. J. R. Meteorol. Soc. DOI: /qj Introduction Monsoon depressions (MDs) are some of the most important synoptic-scale disturbances occurring over the Indian region during the summer monsoon season. The number of MDs which occur over India, their strength and their longevity are the primary contributors to the quantity of Indian summer monsoon rainfall and to the Copyright c 2010 Royal Meteorological Society

2 A. Routray et al. frequency of flooding throughout India s central river basins (Rao, 2001). Improved forecasting of MDs and their inland evolution is one of the top priorities for the Indian monsoon operational and research community (Sikka and Rao, 2008; Chang et al., 2009). In response to this priority need, mesoscale assimilation studies have been under way to utilize different observational datasets to study the MD and low-pressure systems affecting India (Bhowmik and Prasad, 2001; Rajan et al., 2001; Das et al., 2003; Singh and Pal, 2003; Mukhopadhyay et al., 2004; Vaidya et al., 2004; Routray et al., 2005, 2010; Mandal et al., 2006; Sandeep et al., 2006; Vinodkumar et al., 2007, 2008, 2009a, 2009b; Govindankutty et al., 2008; Xavier et al., 2008). Doppler weather radar (DWR) observation is an important data source for mesoscale and microscale weather analysis and forecasting. Currently the variational techniques have received considerable attention for retrieval or assimilation of DWR observations (Wu et al., 2000). Studies such as Gao et al. (1999) and Xiao et al. (2005, 2007) used three-dimensional variational data assimilation (3DVAR) DWR analysis system in the Advanced Regional Prediction System (ARPS) and the Mesoscale Model version 5 (MM5) and showed improvements in the simulation of mesoscale events. The improved initial conditions using DWR enhanced the short-range prediction of extreme events. Building on the positive results obtained from past studies, a Weather Research Forecasting (WRF) 3DVAR system (hereafter WRF-Var) is being developed and is undergoing tests over different regions for assimilating the DWR data. Recently, four DWRs have been installed along the east coast of India for monitoring extreme weather conditions. Despite the importance of the DWR data for use in warning of the extreme weather events associated with convective systems, there has not been much effort made to include the DWR data in the assimilation cycle of the operational weather prediction models in India (Das et al., 2006; Abhilash et al., 2007). In this paper, we present the first results associated with DWR radial velocity and reflectivity data assimilation into the WRF-Var modelling system for prediction of Indian MDs. The main purpose of this study is to assess the impact of assimilating DWR radial velocity and reflectivity data on the simulation of precipitation and wind fields associated with the MDs using the WRF-Var modelling system. 2. Synoptic overview of the MDs We studied two MDs that occurred between 2 and 4 August 2006 (case-1) and 4 and 7 July 2007 (case-2) over the Bay of Bengal (BOB). These two cases were chosen primarily because of the availability of the DWR data along the storm track. Further, the two cases are typical of MDs making landfall over India and as such the results can be generalized for regional application of DWR data over the monsoon region. For the first MD, a low-pressure area formed over northwest BOB off the West Bengal and Orissa coasts on 1 August The system moved southwest and intensified into a depression at 0300 UTC 2 August near 20.5 N, 87.5 E. At 0900 UTC 2 August, it further intensified into a deep depression centred over 20.0 N, 87.0 E. On 3 August the deep depression crossed the south Orissa coast (19.16 N, E). At this time the tropical system came under the influence of upper-air southeasterly winds, which gave it northwestward movement. While moving northwestwards the system weakened into a depression by 0900 UTC 3 August (20.5 N, 82.0 E) and further weakened into a lowpressure area on 5 August over central India. The system produced heavy to very heavy rainfall along eastern and central India. Kalpana-1 satellite images (Figure 1(a) and (b)) show consistent cloud bands over eastern and central India. The second MD case being studied occurred in the first week of July A low-pressure area formed over the north BOB on 3 July It deepened into a depression over the Bangladesh coast at 0300 UTC 4 July near 22.0 N, 89.5 E. The system moved in a westerly direction and lay centred at 0300 UTC 5 July near 23.0 N, 88.0 E, and further intensified into a deep depression at 1200 UTC 5 July. The MD continued to move in a westerly direction and weakened at 0300 UTC 7 July near 23.5 N, 83.5 Eover centralindia.thissystemalsoresultedinwidespreadrainfall with scattered heavy to very heavy rainfall and extremely heavy rainfall over eastern and central India covering West Bengal, adjoining Orissa and Madhya Pradesh during the life span of the system. Kalpana-1 imagery at 0000 UTC for 6 and 7 July 2007 (Figure 1(c) and (d)) shows a dense cloud mass over the east coast and north-central part of India in association with the depression. 3. Modelling system 3.1. WRF-Var We used the Advanced Research version of the WRF (ARW) modelling system (Skamarock et al., 2005). As part of the WRF model, the 3DVAR system provides an analysis x a via the minimization of cost function J( x): J(x) = J b + J o = 1 2 (x xb ) T B 1 (x x b ) (y yo ) T (E + F) 1 (y y o ) (1) The VAR problem can be summarized as the iterative solutionofeq. (1) to findtheanalysis state x that minimizes J(x). In the equation, x b is the background (previous forecast) and y o is the observation. The fit to individual data points is weighted by estimates of their errors: B, E and F as the background, observation (instrumental) and representivity error covariance matrices, respectively. The representivity error is an estimate of inaccuracies introduced in the observation operator H used to transform the gridded analysis x to observation space y =Hx for comparison with the observations. This error is domain- and resolutiondependent and may also include a contribution from approximations in H. Details about the components and real-time applications of the 3DVAR system are reported in Barker et al. (2004) Methodology for Doppler radar data assimilation The DWR radial velocity contains information on vertical atmospheric motions which are important for convective The India Meteorological Department (IMD) classifies rainfall as heavy, very heavy, and extremely heavy for values greater than 70 mm/day, 100 mm/day and 120 mm/day, respectively.

3 Impact of Doppler Radar on Forecast of Monsoon Depressions Figure 1. Kalpana-1 satellite imageries: (a) (b) for case-1; (c) (d) for case-2; (e) model domain and (f) coverage of DWR(Kolkata) radial velocity data after preprocessing compared with the Kolkata radiosonde wind at 1.5 km (around 850 hpa). This figure is available in colour online at wileyonlinelibrary.com/journal/qj initiation and forecasting. The methodology for assimilating DWR data reflectivity and vertical velocity component of radial velocity within the WRF-Var analysis system follows Xiao et al. (2005, 2007). The Doppler reflectivity is related to the amount of precipitation hydrometeors (such as rain, snow and ice). In the Doppler reflectivity assimilation, the total water mixing ratio q t is used as a control variable and introduced as a warm rain process to partition the moisture and water hydrometeor increments in the physical transformation. For the vertical velocity component of radial velocity, the Richardson (1922) balance equation is introduced into WRF-Var to produce the vertical velocity increments. Reflectivity assimilation has more impact on the moisture and hydrometeors. On the other hand, the assimilation of radial velocity has primary impact on the wind analysis and the effect on moisture and hydrometeor analysis is secondary. Assimilation of both radial velocity and reflectivity leads to adjustments in both the dynamical and thermodynamical fields (Xiao et al., 2007). We also conducted experiments to isolate the potential impact of the DWR radial velocity and the reflectivity individually

4 A. Routray et al. and collectively along with Global Telecommunication System (GTS) observations over the Indian monsoon region (Routray, 2007). Four numerical experiments were conducted with assimilation of (1) only GTS observations, (2) GTS along with both radar radial velocity and reflectivity (RVRF), (3) GTS along with radial velocity (RV), and (4) GTS along with only reflectivity (RF). The numerical experiments were performed for 15 days continuously with a 6-hourly assimilation cycle. The mean root-mean-square error (RMSE) and bias for wind components, temperature, and humidity were computed over the experimental domain at different pressure levels (figures not included but available in Routray (2007)). The RMSEs of wind components were reduced in the RV experiment at the surface as well as at 850 hpa and 500 hpa, as compared to the RF simulations. Similarly, the RMSE of thermodynamics parameters were smaller in the RF simulation. The RMSE and bias of all the variables are significantly reduced at different pressure levels in the RVRF assimilation experiment as compared to the other experiments. 4. Numerical experiments Numerical experiments were conducted with WRF and WRF-Var modelling systems for calculating the background error (BE), and for the verification of different WRF-Var experiments against analyses and observations. Results were also analysed to assess the impact of the assimilation of the Indian DWR data on the MD forecasts. Description of each numerical experiment is given in the following subsections Background error covariances and assimilation approaches with WRF-Var The National Meteorological Center (NMC) method (Parrish and Derber, 1992) was used to calculate the BE over the Indian region. The WRF model (30 km horizontal grid spacing) with Final National Centers for Environmental Prediction (NCEP) analyses (FNL; 1 1 ) provided as model initial and boundary conditions was run for the month of August 2005 to produce 12 h and 24 h forecasts at 0000 UTC and 1200 UTC. This default numerical experiment is referred to as No-Assim. Figure 1(e) shows the model domain configured for the present study. A summary of the relevant model configuration, physics and optionsusedinthepresentstudyisshownintablei. Different single observation tests at various locations (latitude, longitude and sigma level) were performed over theindiandomain.theseweredonetoinvestigatethe response of BE in the WRF-Var system over the Indian domain. Following the calculation of BE statistics and singleobservation tests, we conducted two additional experiments to identify the optimal procedure for the WRF-Var analysis system. The first experiment, named cold start, was the same as No-Assim but with data assimilation performed at each initial time (0000 UTC and 1200 UTC) to produce 12 h and 24 h forecasts. The second experiment, called cycling, was the same as the cold start experiment but with the FNL analyses providing the first guess and boundary conditions for the first cycle. In the 12-hourly cycling update, the WRF model forecasts were used as first guess for subsequent cycles. The computed BE statistics were used in both the experiments (cold start and cycling). The mean vertical Table I. WRF model configuration used in this study. Horizontal grid distance 30 km Integration time step 90 s Number of grid points x-direction 151 (61 E, 102 E) points y-direction 151 (1 N, 38 N) points Vertical levels in First guess 27 analysis Vertical coordinate Terrain-following hydrostatic-pressure coordinate (51 levels) Model top 10 hpa Microphysics Lin et al. (1983) scheme Radiation scheme (long RRTM scheme wave) Radiation scheme (short wave) Dudhia s short-wave radiation Surface layer physics Monin Obukhov scheme Surface layer parametrization Thermal diffusion scheme PBL parametrization YSU scheme (Hong et al., 2004) Cumulus parametrization schemes Betts Miller Janjić scheme (Janjić, 1994) Time integration 3 rd -order Runge Kutta Spatial differencing scheme 6 th -order centred differencing Map projection Mercator Horizontal grid distribution Arakawa C-grid Main prognostic variables u, v, w, p /, θ /, / Number of domains Single domain Central point of the Central lat. and lon.: domain 20.0 N and 80.5 E Initial and boundary conditions 3-dimensional real data (FNL: 1 1 ) profiles of RMSE were computed over the experimental domain at different pressure levels against available GTS observations (upper-air, SYNOP, buoy, ship, SATOB etc.) and independent analyses (FNL) for the three experiments (cold start, cycling and No-Assim) Assimilation of Indian DWR data Data quality control is an important step in data assimilation procedures. A preprocessor module and standalone software were developed for processing and quality controlling the voluminous Indian DWR data. From the DWR data, radial velocity, reflectivity and other information such as radar latitude, radar longitude, number of elevation and azimuth steps were retrieved. The data quality-control procedures followed the typical practices employed at the radar data processing centres such as the Korean Meteorological Administration (KMA), South Korea, and the Center for the Analysis and Prediction of Storms (CAPS), Oklahoma, USA. These procedures include tests for range folding, and discarding the radial velocities with absolute values outside the range of 2 to 30 m/s and also discarding the reflectivity

5 Impact of Doppler Radar on Forecast of Monsoon Depressions Table II. RMSE of DWR observations before (O B) and after (O A) 3DVAR analysis at model initial time for case-1 and case-2. Time (yyyymmdd hh) Radial velocity (m/s) Reflectivity (dbz) O B O A O B O A UTC (case-1) UTC (case-2) outsidetherange10to60dbz.dwrdataarearchived on spherical coordinates with range: km, azimuth: 360 degrees (1 beam width), elevation: degrees and resolution: km. These datasets are voluminous and not compatible for ingestion within the analysis system. For the Indian domain, the analysis is performed at 30 km resolution, therefore it is not desirable to include each and every radar datum in this dataset. The preprocessor software was therefore designed to thin the DWR dataset by (1) building a desired thinning grid (two-dimensional), (2) collecting the maximum specified number of radar observations in each thinning grid square, and (3), for each grid square, identifying the nearest data location with maximum number of levels from the left-lower corner. After thinning, two sets of data were produced: (1) input for the WRF-Var assimilation system, and (2) data suitable for plotting the processed radar files. Figure 1(f) shows the data coverage (after quality control and thinning) of radar radial velocity data at 1.5 km (approximately 850 hpa). The radiosonde wind observations from Kolkata radiosonde/radar wind (RS/RW) station valid at 0000 UTC 5 July 2007 are also shown. The radial velocity is consistent with the observed RS/RW winds. For the two Bay of Bengal (BOB) MDs being studied in this paper, Kolkata DWR (22.5 N, 88.4 E) data were processed and assimilated within WRF-Var. This is because both the depressions formed within the radius of the Kolkata DWR observations. For each MD, three numerical experiments (named as CNTL; 3DV GTS and 3DV DWR) were performed. These were designed to assess the effect of DWR assimilation relative to GTS data assimilation. In the CNTL experiment, the model was integrated without any data assimilation. In the second experiment (3DV GTS), a 6-hourly assimilation cycle was performed using the conventional and nonconventional data obtained through GTS/Internet. In the 6 h 3DVAR update cycle, the 6 h forecast from the previous cycle served as the background for the next cycle (Abhilash et al., 2007; Xiao et al., 2007). A total of four cycles were performed to initialize the model prior to the start of the actual model integration. The 3DV DWR experiment is similar to 3DV GTS, but the Kolkata DWR radial velocity and reflectivity observations are used along with the other conventional and non-conventional data in the assimilation system. Table II shows the RMSE of the DWR data assimilation before (O-B) and after (O-A) 3DVAR analysis at model initial time for both the cases. The reflectivity and radial velocity simulation is also improved for both cases. The model was integrated for 54 h in all the experiments The IMD is currently adopting the warning decision support system (WDSS-II) software from University of Oklahoma, USA (Lakshmanan et al., 2007) for DWR data quality control and thinning purposes. from the initial time 0000 UTC 2 August 2006 for case-1, and from 0000 UTC 5 July 2007 for case-2, respectively. 5. Results and discussion Various aspects of the computed BE statistics were studied with different single-observation tests. Results of these single-observation tests with temperature and wind components at different locations and at different sigma levels are briefly discussed in section 5.1. The objective verification assessed the efficacy of the WRF-Var modelling system with different assimilation modes (cold start and cycling). These results are presented in section 5.2. After identifying the optimal data assimilation mode within WRF- Var, the focus is then on the discussion of the impact of assimilation of DWR radial velocity and reflectivity data into the WRF-Var system on the simulation of the two MDs. A discussion of the model simulation results for the MDs is provided in section Single-observation test The single-observation test is often used as a proof-ofconcept test to determine how the observed entity spreads to its vicinity via the established correlations among 3DVAR variables (Wu et al., 2002). A single temperature and u- wind observation perturbation test was applied at the middle of the domain (20.1 N, 80.6 E; sigma level 25). The innovation (observation minus background; O-B) of the single temperature and single u-wind was assigned as 1Kand1ms 1 respectively. Results indicate that the shape of the horizontal temperature increment (Figure 2(c)) is confined to about 150 km. The magnitude and spread of the temperature increment (Figure 2(c)) is small. The magnitude of the wind (Figure 2(a) and (b)) response from the temperature increment is also small; however, the spatial impact is relatively more as compared to the temperature increments (Figure 2(c)). Figure 2(d) (f) show the analysis increments in response to the single u-wind component (1 m s 1 )observation applied at the middle of the domain. A spread of u- increment (Figure 2(e)) of 0.7 m s 1 is observed. The response of v-increments (Figure 2(f)) supports cyclonic and anticyclonic circulation to the north and south of the u-increment location, respectively. Both cold and warm temperature increments (Figure 2(d)) are found. As seen in Figure 2(c), the scale of temperature response on wind (< T, u >) was large and the same is noted for wind response on temperature (< u, T >) infigure2(e)and (f). The temperature increment is created via the hydrostatic balance, mass-wind balance and Richardson equation. The temperature increment (Figure 2(d)) response is small, but its length scale is larger than that of the wind increments.

6 A. Routray et al. Figure 2. (a) (c) Response of the analysis increments to a single temperature observation 1 K at (20.1 N; 80.6 E; 25-sigma level). Similarly, (d) (f) are the responseof the analysis incrementsto a single u-wind perturbation of 1 m/s Verifications against observations and analyses Figure 3(a) (h) show the 30-day (2 31 August 2005) domain-averaged RMSE vertical profile for 24 h forecasts from different experiments (cold start, cycling and No- Assim) for wind, temperature and moisture. The RMSEs are calculated against the available observations and the FNL analyses, averaged over the experimental domain at each pressure level. In the cycling experiment, the RMSE of wind and temperature (Figure 3(a) (c)) gradually improved above the 850 hpa pressure levels and significant improvement can be found for the upper atmosphere. The calculated RMSE profiles of 24 h forecasts against analyses are shown in Figure 3(e) (h). Again, in the cycling experiment, the RMSE values are significantly lowered compared to the other two experiments for all parameters for the different pressure levels. Also the RMSEs are smaller in the cold-start experiment as compared to the No-Assim experiment except for temperature. Overall, the cycling mode assimilation enhanced the performance of the WRF-Var over the Indian region using conventional and non-conventional observations Impact of DWR data on model simulations The impact of assimilating DWR fields on the model performance is examined in this section Mean-sea-level pressure (MSLP) The subjectively analysed MSLP by IMD is shown in Figure 4(a) and (d) for both the cases. The corresponding 24 h forecast of MSLP from three numerical experiments for both the MDs are illustrated in Figure 5. The MSLP patterns over the experimental domain are well simulated in all the simulations and they reasonably reproduce the observations. In particular, the 3DVAR assimilation results show that the location of the centre of the depression is predicted reasonably well during day-1 in both the cases as compared to the CNTL and 3DV GTS simulations. For both the MD cases, the position and the intensity of the depression are better simulated by the 3DV DWR compared to the CNTL and 3DV GTS Wind Figure 4 (b) (c) and (e) (f) shows the subjectively analysed IMD wind fields at 850 hpa and 500 hpa for the two cases. Figures 6 and 7 show the day-1 wind forecasts at 850 hpa and 500 hpa from the three experiments for case-1 and 2, respectively. As seen in the figures, for both the cases, the circulation around the depression is simulated well. The winds were stronger and closer to the observation in the assimilation experiments as compared to the CNTL simulation. The location and northwest movement of the MDs is also slightly better simulated in the 3DV experiments. For case-1, the maximum wind speed at 850 hpa simulated by the 3DV DWR experiment (Figure 6(c)) ranges between 20 and 30 m s 1, spread over a large area on the southeastern and northwest part of the vortex. The wind speeds reasonably match with the corresponding radiosonde observations over the region, shown in Figure 4(b). The position of the MDs is corrected in the 3DV DWR simulation after assimilation of DWR data as compared to the CNTL and

7 Impact of Doppler Radar on Forecast of Monsoon Depressions (a) (b) (c) (d) 300 Pressure (hpa) Zonal wind (m/s) Meridional wind (m/s) Temperature (k) Q (g/kg) (e) (f) (g) (h) 300 Pressure (hpa) Zonal wind (m/s) Meridional wind (m/s) Temperature (k) Q (g/kg) cycling cold start No-Assm Figure 3. Vertical profiles of RMSE from 24 h forecast (2 31 August 2005) calculated against observations for (a) zonal wind (m/s), (b) meridional wind (m/s), (c) temperature (K), and (d) specific humidity (g/kg). Similarly, (e) (h) are the same as (a) (d) respectively but for RMSE calculated against analysis. 3DV GTS simulations. Over the oceanic regions as well as the peninsular Indian region, the assimilation simulations during day-1 at the 850 hpa level are able to accurately predict the winds. Similarly, in the 3DV DWR simulations (Figure 6(f)) at the 500 hpa pressure level, the maximum wind speed has the magnitude of m s 1 located over a large area on the north and northeast sector of the depression and the oceanic region. These features are relatively poorly simulatedinthecntland3dvgts experiments. The presence of strong winds over the northern quadrant is one of the characteristic features of monsoon depressions (Rao, 1976). This circulation pattern is noted in both the assimilation experiments. However, the magnitude of wind overtheregionisimprovedinthe3dvdwr simulations at the 500 hpa level when DWR data are assimilated. Similar results are found for case-2, where the location of the depression on the day-1 forecast at different pressure levels (Figure 7(a) (f)) is well simulated in the 3DV DWR simulations. At 850 hpa, the strong magnitude of winds on the southwestern side of the vortex as well as the peninsular India region is better reproduced in the 3DV experiments (Figure 7(b) and (c)) as compared to the CNTL (Figure 7(a)). The position and magnitude of wind (22 26 m s 1 ) in the northern part of the depression is well simulated in 3DV DWR (Figure 7(c)) and closer to the observations (Figure 4(e)). At the 500 hpa level, the 3DV DWR (Figure 7(f)) experiment simulated strong winds (22 26 m s 1 ) in the northern quadrant not captured in the CNTL (Figure 7(d)) and 3DV GTS (Figure 7(e)) experiments. A noteworthy feature as the MD intensifies is the high winds (25 30 m s 1 ) in the lower tropospheric westerlies at hpa (Krishnamurti et al., 1975; Daggupaty and Sikka, 1977). The 3DV DWR simulations for both the depression cases reproduce this feature well. Overall, the assimilation of DWR radial velocity and reflectivity data has resulted in a positive impact on simulation of wind structures associated with MDs. Results from Xiao et al. (2005) suggest that the vertical velocity component in the Doppler radial velocity observation not only adjusts the vertical velocity in the analysis, but also influences the horizontal wind analysis. The vertical velocity is important for simulating the convective systems, while the Doppler reflectivity is important for correcting the horizontal winds. In both MD cases, the convection, winds and the northward propagation of the MDs are relatively better initiated and represented in the model initial condition after assimilating the DWR data. The initial position errors of the MDs are also minimized after assimilation of DWR data. In the east west cross-section of vertical velocity (figure not provided) for both the cases, the day-1 simulations show the maximum updraughts in the upper atmosphere. Both the assimilation experiments produced strong vertical motion as compared to the CNTL experiment. The magnitude of the vertical velocity is also more in the 3DV DWR experiment as compared to the 3DV-GTS experiments.

8 A. Routray et al. Figure 4. Observed mean-sea-level pressure (MSLP) and streamline analyses at different pressure levels for (a) MSLP; (b) 850 hpa; (c) 500 hpa, valid at 0000 UTC 3 August 2006 (case-1). (d) (f) are the same as (a) (c) respectively but for case-2, valid at 0000 UTC 6 July Source: IMD. In case-1, strong positive vertical velocity cores are found at the upper pressure level (around 300 hpa). Similarly, the 3DV DWR experiment in case-2 predicts two strong magnitudes of vertical velocity cores (0.8 m s 1 ) at upper levels as compared to the two other experiments. Also in the 3DV DWR simulation, upward motions are found in the west side of the MD and subsidence to the east. The features are consistent with the findings of Krishnamurti et al. (1975, 1976) Simulated rainfall and composite reflectivity One of the objectives for assimilating the high-resolution DWR data is to help improve the rainfall forecast associated with the meso-convective systems which accompany the MDs. DWR data contain precipitation hydrometeor information, and direct assimilation into a 3DVAR system is a challenge. Xiao et al. (2007) used total water content as a control variable and incorporated a warm-rain partitioning scheme in the 3DVAR. The 24 h accumulated precipitation for day-1 as obtained from CNTL, 3DV GTS and 3DV DWR simulations and the corresponding Tropical Rainfall Measuring Mission (TRMM-3B42V6: Huffman et al., 2003) satellite estimates are shown in Figure 8 (for case-1 and case-2). Comparisons of the 24 h accumulated model-simulated rainfall and observed precipitation of case- 1 and case-2 are given in Tables III and IV. Another typical characteristic of the MD is that heavier rainfall is chiefly

9 Impact of Doppler Radar on Forecast of Monsoon Depressions Figure h forecast of MSLP (hpa) with contour interval 1 hpa (day-1) for (a) CNTL, (b) 3DV GTS, and (c) 3DV DWR, valid at 0300 UTC 3 August (d) (f) are the same as (a) (c) but for case-2, valid at 0300 UTC 6 July Figure 6. Simulated wind speed and magnitude (shaded, m/s) at 850 hpa from (a) CNTL; (b) 3DV GTS, and (c) 3DV DWR experiments for day-1 valid at 0000 UTC 3 August (d) (f) are the same as (a) (c) but at 500 hpa.

10 A. Routray et al. Figure 7. Same as Figure 6 but for case-2, valid at 0000 UTC 6 July concentrated in the western and southwestern sectors of the depression (Sikka, 1977). These heavy rain episodes can typically reach 300 mm. These features are better simulated in the 3DV DWR runs for both the cases. During day-1, the model simulated a well-established closed circulation system only in the 3DV DWR simulations. For case-1, the 3DV DWR simulation (Figure 8(d)) shows heavy rainfall ( mm) over the land masses and widespread rainfall over oceanic regions. The TRMM estimates (Figure 8(a)), however, show maximum rainfall only over the oceanic regions. Therefore, the simulated rainfall was compared with the surface rain-gauge observations (Table III). The locales of heavy rainfall (230 mm) recorded at the Jeypore station (18.51 N; E); Koraput (18.49 N; E) and 200 mm rainfall over Bhanupratappur (20.19 N; E) as well as Pottangi (18.34 N; E) during day-1 are well simulated in the 3DV DWR experiment as compared to the CNTL and 3DV GTS simulations (Table III). Similar results are seen for case-2 (Figure 8(h)). The CNTL experiment underestimates the rainfall (by mm). The position of maximum precipitation (270 mm) recorded at the Jamsolaghat station (22.13 N; E); Kharagpur (25.07 N; E) and the second maximum rainfall (260 mm) over Mohanpur (21.52 N; E) during day-1 are well captured in the 3DV DWR model simulation as compared to the CNTL and 3DV GTS simulations (Table IV). Additionally, the spatial coherence of the simulated rainfall is also better in the 3DV GTS simulation as compared to the CNTL simulation. A noteworthy feature from the 3DV DWR experiment is the simulation of intense convective rain-bearing clouds and their northwesterly movement with the MD. Thus, ingesting the radar reflectivity information into the 3DVAR analysis yields positive impacts on the rainfall forecast skill from the MDs. The spatial (15 25 N; E) correlation coefficient (CC) and RMSE of rainfall between TRMM and the model outputs are calculated with the day-1 forecast over the land (masking the ocean part) for case-1 and 2. The RMSE values are smaller in the 3DV DWR simulation (24.3 mm in case-1 and 20.0 mm in case-2) as compared with the 3DV GTS (34.9 mm in case-1 and 26.3 mm in case-2) and CNTL (40.5 mm in case-1, 32.0 mm in case-2) simulations. Similarly, the CC is improved in the 3DV DWR (0.74 in case-1 and 0.62 in case-2) as compared to 3DV GTS (0.56 in case-1 and 0.47 in case-2) and CNTL (0.37 in case-1 and 0.25 in case-2) experiments. The Equitable Threat Score (ETS) was also calculated. Consistent with the results summarized above, the scores are considerably better in the 3DV DWR experiment throughout the period, as well as for different rainfall threshold values as compared with the CNTL and 3DV GTS experiments. The observed reflectivity for case-1 and case-2 from the Kolkata DWR is shown in Figure 9(a) and (b) valid at 0000 UTC of 3 August 2006 and 6 July 2007 respectively. Figure 10 shows the 24 h simulated composite radar reflectivity for case-1 and case-2 from three experiments. The modelderived echoes from the 3DV DWR simulations for both the cases are distributed over a large area with a maximum reflectivity range from 25 to 35 dbz as compared to CNTL and 3DV GTS simulations which had reflectivity values from 15 to 25 dbz. The maximum observed reflectivity (Figure 9(a) and (b)) is around dbz in both the cases. All the model simulations underestimated the intensity as compared to that observed; even then the results from DWR

11 Impact of Doppler Radar on Forecast of Monsoon Depressions Table III. Comparison between station-wise observed and model simulated rainfall in cm for day-1 valid at 0300 UTC 3 August Station names Lat (deg) Lon (deg) Observed (cm) CNTL 3DV GTS 3DV DWR Jeypore Koraput Bhanupratappur Pottangi R. Udaigiri Nawarangpur Palakonda Malkangiri Kashinagar Gariabund Bobbili Narayanpur Dantewara Visakhapatnam Table IV. Same as Table III but valid at 0300 UTC 6 July Station Names Lat (deg) Lon (deg) OBS (cm) CNTL 3DV GTS 3DV DWR Jamsolaghat Kharagpur Mohanpur Midnapore Kalaikunda Joshipur Rairangpur Diamond Harbour Udala Haldia Contai Jaipur Bankura Uluberia assimilations are considerably closer to the observations over the region. The spatial distribution of the reflectivity simulated by the 3DV DWR experiment is relatively closer to the observed reflectivity for both the cases as compared to other experiments. In the 3DV DWR experiments, the convective rain-band clouds embedded within the MDs are also better simulated as compared to the other simulations. The convective clouds quickly dissipate in the 3DV GTS and CNTL simulations as the MDs moved northwestwards during day-1 day-2 (figure not provided for day-2). The rain bands behind the convective clouds are also well simulated in the 3DV DWR experiments Track predictions The IMD-observed MD tracks and 54 h forecasts of the CNTL, 3DV GTS and 3DV DWR tracks (location of minimum MSLP centre) for the two MDs are shown in Figure 11(a) and (c). Corresponding track errors for the two cases are shown in Figure 11(b) and (d). Within the uncertainty of observations, the3dv DWR simulated track is better than the CNTL and 3DV GTS tracks for both the cases. The initial position errors in both the depression cases are also less in the 3DV DWR experiment. The tracks are also improved in the 3DV GTS simulation as compared to the CNTL simulated track. For case-1, the maximum track error is about 325 km and 250 km for the CNTL and 3DV GTS simulations at 0000 UTC 4 August 2006 respectively. Corresponding track errors in the 3DV DWR experiment are about 70 km. The maximum track error for the 3DV DWR run is about 170 km at 0600 UTC 3 August For case-2, the model-simulated speed of the depression is slightly higher than the observation. However, the 3DV DWR simulated track matches the observed track up to 0600 UTC 6 July Track errors increased with the forecast period. However, the errors in 3DV DWR simulation are comparatively less than that in the CNTL and 3DV GTS simulations. Overall the results show that the track errors are reduced significantly in assimilation experiments and particularly in the DWR assimilation simulations as compared with other experiments.

12 A. Routray et al. Accumulated Precipitation (a) 03 August 2006 (e) 06 July 2007 (b) (f) (c) (g) (d) (h) Figure h accumulated precipitation (cm) for day-1: (a) TRMM, (b) CNTL, (c) 3DV GTS, and (d) 3DV DWR, valid at 0300 UTC 3 August (e) (h) are the same as (a) (d) respectively but for case-2, valid at 0300 UTC 6 July Conclusions The WRF-3DVAR system was customized and evaluated for the Indian region. Domain-specific background error (BE) and two different assimilation strategies (cold start and cycling) were investigated using the WRF-Var analysis scheme. The WRF-ARW modelling system with 30 km horizontal grid spacing was used to calculate the BE. A 30- day simulation with 24 h forecasts at 0000 UTC and 1200 UTC initial conditions was analysed. Single-observation tests for a single temperature (1 K) and wind component (u-wind; 1 m s 1 ) perturbations were also performed. The synthetic tests indicate that the analysis increments respond well to the assimilation of the temperature and wind over the monsoon domain. A series of cycling- and cold-start-based assimilation experiments with the WRF- Var modelling system were conducted using conventional and non-conventional observations for a 30 day period.

13 Impact of Doppler Radar on Forecast of Monsoon Depressions Figure 9. Observed reflectivity (dbz) from Kolkata DWR station valid at 0000 UTC for (a) 3 August 2006, and (b) 6 July This figure is available in colour online at wileyonlinelibrary.com/journal/qj Figure 10. Composite reflectivity (dbz) for case-1: (a) CNTL, (b) 3DV GTS, and (c) 3DV DWR, valid at 0000 UTC 3 August 2006 (day-1). (d) (f) are the same as (a) (c) respectively but for case-2, valid at 0000 UTC 6 July 2007 (day-1). From the statistical skill score, it is concluded that the cycling mode assimilation enhanced the performance of the WRF-Var over the Indian region. The study also focused on assessing the impact of assimilating DWR radial velocity and reflectivity in the WRF-Var data assimilation system for the simulation of two typical Bay of Bengal MDs. For each case, three numerical experiments assimilating GTS and DWR (3DV DWR), assimilating only GTS (3DV GTS), and without assimilation of either GTS or DWR (CNTL) were performed. For both the MDs, the 3DV DWR experiment could simulate the large-scale fields relatively better as compared to the CNTL and 3DV GTS simulation. Table II summarizes that the 3DV DWR experiment properly assimilated the Doppler radial velocity and reflectivity. Assimilation of Doppler data with a cycling mode extracted useful information from Doppler data and improved the forecast skill. The MSLP was better simulated in the 3DV DWR as compared to CNTL and 3DV GTS simulations during day-1. Assimilation of radial velocities helped with the triggering and modulation of convection as well as precipitation in the model results. The assimilation of Doppler data in the 3DV DWR assimilation experiment produced several important features and hydrometeor distributions embedded within the MDs. The movement and location of the depression were well represented in the

14 A. Routray et al. Case-1(2-4 August 2006) (a) (b) Track Vector displacement errors (VDEs in km) CNTL 3DV_GTS 3DV_DWR 250 VDEs Forecast hours(ddhh) Case-2 (5-7 July 2007) (c) Track (d) Vector displacement errors (VDEs in km) 450 VDEs (km) CNTL 3DV_GTS 3DV_DWR Forecast hours (ddhh) Figure hourly track (observed and simulated) and VDEs (km) from CNTL; 3DV GTS and 3DV DWR. (a) Track, and (b) VDEs for case-1 (initial time 0600 UTC 2 August 2006). (c) (d) are the same as (a) (b) respectively, but for case-2 (initial time 0000 UTC 5 July 2007). assimilation experiment (3DV DWR) as compared to the IMD observations. The model precipitation and convection fields also compared well with the TRMM rainfall estimates and the Kalpana-1 visible satellite convection imagery. Additionally, several typical features of an MD such as precipitation maximum over the south-west quadrant have been correctly simulated by the 3DV DWR experiment for both the cases. The monsoon depression tracks were also well simulated in the DWR assimilation experiment. For both the cases, the track errors are significantly reduced in 3DV DWR simulation as compared to the CNTL and 3DV GTS simulation. The intensity and structure of the MDs is thus better simulated with the DWR assimilation. This work is a first assessment of utilizing IMD DWR radial velocity and reflectivity in a mesoscale model for prediction of Bay of Bengal MDs. Study results provide a positive proof of concept that the assimilation of the Indian DWR data within WRF can help improve the simulation of intense convective systems influencing the large-scale monsoonal flow. Additional studies are being conducted to assimilate the extensive quality-controlled multi-radar radial velocities and reflectivities into the assimilation system and decide on optimal model horizontal resolutions, and will be reported in a follow-up paper. Acknowledgements The authors gratefully acknowledge the NCEP for providing the FNL analyses and GTS observations. The authors thank the Kolkata Regional Meteorological Centre for providing DWR radial velocity and reflectivity observations which were used in the WRF-Var data assimilation system in this study. Thanks also to the IMD for providing the synoptic upper-air charts and observed tracks of the monsoon depressions that were used to validate the model results of this experiment. Dale Barker, NCAR, is gratefully acknowledged for his immense assistance with this study. DN was supported by NSF CAREER (ATM Liming Zhou and Jay Fein), NASA-THP (Jared Entin), and a NOAA JCSDA grant. We express our sincere thanks to anonymous reviewers and the Associate Editor for their valuable comments and suggestions for improvement of the paper. References Abhilash S, Das S, Kalsi SR, Das Gupta M, Mohankumar K, George JP, Banerjee SK, Thampi SB, Pradhan D Impact of Doppler radar wind in simulating the intensity and propagation of rainbands associated with mesoscale convective complexes using MM5-3DVAR system. Pure Appl. Geophys. 164:

15 Impact of Doppler Radar on Forecast of Monsoon Depressions Barker DM, Huang W, Guo Y-R, Bourgeois AJ, Xiao QN A threedimensional variational data assimilation system for use with MM5: Implementation and initial results. Mon. Weather Rev. 132: Bhowmik SKR, Prasad K Some characteristics of limited-area model-precipitation forecast of Indian monsoon and evaluation of associated flow features. Meteorol. Atmos. Phys. 76: Chang H-I, Niyogi D, Kumar A, Kishtawal CM, Dudhia J, Chen F, Mohanty UC, Shepherd M Possible relation between land surface feedback and the post-landfall structure of monsoon depressions. Geophys. Res. Lett. 36: L15826, DOI: /2009GL Daggupaty SM, Sikka DR On the vorticity budget and vertical velocity distribution associated with the life cycle of a monsoon depression.j. Atmos. Sci. 34: Das AK, Mohanty UC, Das S, Mandal M, Kalsi SR Circulation characteristics of a monsoon depression during BOBMEX-99 using high-resolution analysis. Proc. Indian Acad. Sci. 112: Das S, Abhilash S, Das Gupta M, George JP, Kalsi SR, Banerjee SK, Thampi SB, Pradhan D, Mohankumar K Assimilation of Doppler weather radar wind in a mesoscale model and their impact on simulation of thunderstorms and severe weather systems. Report no.nmrf/rr/01/2006, 120 pp. Published by NCMRWF, A-50, Sector-62, NOIDA, India Gao J, Xue M, Shapiro A, Droegemeier KK A variational method for the analysis of three-dimensional wind fields from two Doppler radars.mon. Weather Rev. 127: Govindankutty M, Chandrasekar A, Bohra AK, George JP, Das Gupta M The impact of assimilation of MODIS observations using WRF- VAR for the prediction of a monsoon depression during September Open Atmos. Sci. J. 2: Hong S-Y, Dudhia J, Chen S-H A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Weather Rev. 132: Huffman GJ, Adler RF, Stocker EF, Bolvin DT, Nelkin EJ Analysis of TRMM 3-hourly multi-satellite precipitation estimates computed in both real and post-real time. Combined Preprints CD-ROM, 83rd AMS annual meeting. Paper P4.11 in 12th Conference on satellite meteorology and oceanography, 9 13 Feb 2003, Long Beach, California, 6 pp. Janjić ZI The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer and turbulence closure schemes. Mon. Weather Rev. 122: Krishnamurti TN, Kanamitsu M, Godbole R, Chang C-B, Carr F, Chow JH Study of a monsoon depression. I: Synoptic structure. J. Meteorol. Soc. Jpn 53: Krishnamurti TN, Kanamitsu M, Godbole R, Chang C-B, Carr F, Chow JH Study of a monsoon depression. II: Dynamical structure. J. Meteorol. Soc. Jpn 54: Lakshmanan V, Smith T, Stumpf GJ, Hondl K The warning decision support system integrated information. Weather and Forecasting 22: Lin Y-L, Farley RD, Orville HD Bulk parameterization of the snow field in a cloud model.j. Clim. Appl. Meteorol. 22: Mandal M, Mohanty UC, Das AK Impact of satellite derived wind in mesoscale simulation of Orissa super cyclone. Ind. J. Mar. Sci. 35: Mukhopadhyay P, Sanjay J, Cotton WR, Singh SS Impact of surface meteorological observations on RAMS forecast of monsoon weather systems over the Indian region. Meteorol. Atmos. Phys. 90: Parrish DF, Derber JC The National Meteorological Center s spectral statistical-interpolation analysis system. Mon. Weather Rev. 120: Rajan D, Mitra AK, Rizvi SRH, Paliwal RK, Bohra AK, Bhatia VB Impact of satellite derived moisture in a global numerical weather prediction model. Atmósfera. 14: Rao GN Occurrence of heavy rainfall around the confluence line in monsoon disturbances and its importance in causing floods. J. Earth Sys. Sci. 110: Rao YP Southwest monsoon. Meteorological Monograph, Synoptic Meteorology. No. 1/1976. India Meteorological Department: India. Richardson LF Weather prediction by numerical processes. Cambridge University Press. Routray A Radar data assimilation over Indian region using WRF-VAR system. Scientific Report, February 2007, MMM Division, NCAR, Boulder, CO 80301, USA, 27 pp. Routray A, Mohanty UC, Das AK, Sam NV Study of heavy rainfall event over west-coast of India using analysis nudging in MM5 during ARMEX-I. Mausam 56: Routray A, Mohanty UC, Niyogi D, Rizvi SRH, Osuri KK Simulation of heavy rainfall events over Indian monsoon region using WRF-3DVAR data assimilation system. Meteorol. Atmos. Phys. 106: Sandeep S, Chandrasekar A, Singh D The impact of assimilation of AMSU data for the prediction of a tropical cyclone over India using a mesoscale model.int. J. Remote Sensing 27: Sikka DR Some aspects of the life history, structure and movement of monsoon depressions. Pure Appl. Geophys. 115: Sikka DR, Rao PS The use and performance of mesoscale models over the Indian region for two high-impact events. Natural Hazards 44: Singh R, Pal PK Impact of QuikSCAT derived wind fields in a mesoscale non-hydrostatic model. Mausam 54: Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Wang W, Powers JG A description of the Advanced Research WRF Version 2. NCAR Technical Note NCAR/TN-408+STR. Available from Vaidya SS, Mukhopadhyay P, Trivedi DK, Sanjay J, Singh SS Prediction of tropical systems over Indian region using mesoscale model. Meteorol. Atmos. Phys. 86: Vinodkumar, Chandrasekar A, Alapaty K, Niyogi DS The effect of a surface data assimilation technique and the traditional fourdimensional data assimilation on the simulation of a monsoon depression over India using a mesoscale model. Natural Hazards 42: Vinodkumar, Chandrasekar A, Alapaty K, Niyogi D The impacts of indirect soil moisture assimilation and direct surface temperature and humidity assimilation on a mesoscale model simulation of an Indian monsoon depression. J. Appl. Meteorol. Clim. 47: Vinodkumar, Chandrasekar A, Alapaty K, Niyogi D. 2009a. Assessment of data assimilation approaches for the simulation of a monsoon depression over the Indian monsoon region, Boundary-Layer Meteorol. 133: Vinodkumar, Chandrasekar A, Niyogi D, Alapaty K. 2009b. Impact of land surface representation and surface data assimilation on the simulation of an off-shore trough over the Arabian Sea. Global and Planetary Change 67: Wu B, Verlinde J, Sun J Dynamical and microphysical retrievals from Doppler radar observations of a deep convective cloud. J. Atmos. Sci. 57: Wu W-S, Purser RJ, Parrish DF Three-dimensional variational analysis with spatially inhomogeneous covariances. Mon. Weather Rev. 130: Xavier VF, Chandrasekar A, Rahman H, Niyogi D, Alapaty K The effect of satellite and conventional meteorological data assimilation on the mesoscale modeling of monsoon depressions over India. Meteorol. Atmos. Phys. 101: Xiao Q, Kuo Y-H, Sun J, Lee W-C, Lim E, Guo Y-R, Barker DM Assimilation of Doppler radar observations with a regional 3DVAR system: Impact of Doppler velocities on forecasts of a heavy rainfall case. J. Appl. Meteorol. 44: Xiao Q, Kuo Y-H, Sun J, Lee W-C, Barker DM, Lim E An approach of radar reflectivity data assimilation and its assessment with the inland QPF of Typhoon Rusa (2002) at landfall. J. Appl. Meteorol. Clim. 46: