1 Introduction

Tropical cyclones are among the most deadly natural hazards as they are associated with very strong winds, heavy rain, and storm surge. Due to the increasing human habitation near the coasts accurate and timely forecasting of tropical cyclone has posed a challenging task. During the recent years, significant improvement in accuracy and reliability of numerical weather prediction (NWP) methods has been made, primarily because of phenomenal increase in the use of remote sensing data over oceans to initialize tropical cyclone in numerical models (Krishnamurti et al. 1991; Harasti et al. 2004) and also due to the increase in computing power. However, limitations remain, particularly in the prediction of intensity and mesoscale features associated with the landfall, causing inland flooding.

Assimilation of high-resolution Doppler weather radar (DWR) observations (radial wind and reflectivity) have long been recognized (Xiao et al. 2005, 2007; Gao et al. 2004; Xue et al. 2003) as an efficient way to improve mesoscale and microscale weather analysis, and mesoscale forecasting. Xiao et al. (2005) carried out study on the assimilation of DWR radial wind and reflectivity into MM5 model using the 3 DVAR system for the heavy rainfall events over South Korea region. A number of case studies on the positive impact of DWR radial wind and reflectivity observations in the assimilation cycle of Advanced Regional Prediction System (ARPS) are documented by Xue et al. (2003).

With the installation of four GEMATRONIC METEOR 1500S model DWRs at Chennai (during the year 2002), Kolkata (2003), Machilipatnam (2004), and Vishakhapatnam (2006), it has heightened the prospects for the operational implementation of numerical model to explicitly predict the evolution of mesoscale phenomena. Figure 1 presents the location of these DWRs along the east coast of India. Radar observations from offshore radar sites provide useful information about the mesoscale structure of tropical cyclone around a landfall. Observational studies based on the Indian DWR data are reported by several authors (Sinha and Pradhan 2006; Bhatnagar et al. 2003). But, it remained a challenging task to ingest radial wind and reflectivity observations of Indian DWR in the assimilation cycle of a numerical model. The main interest in this study is to ingest radial wind and reflectivity observations from a offshore radar site of India (Chennai) in the assimilation cycle of ARPS (version 5.2.5) at 9-km horizontal resolution and demonstrate its impact on the analysis and forecast of mesoscale convective features associated with Bay of Bengal land-falling cyclone of 27–28 November 2008.

Fig. 1
figure 1

a Study zone for cyclonic storm NISHA and b Observed track of the cyclonic storm

Full size image

On 27 November 2008, Tamil Nadu coast of India was hit by a tropical cyclone named as NISHA. It caused very heavy rainfall and flood over coastal areas of Tamil Nadu and Pondicherry. This provided a unique opportunity to study the impact of DWR observations of Chennai (lat. 13.0°N and long. 80.0°E) with the use of ARPS model to simulate mesoscale convective features that led to inland flooding.

Experiments are carried out to study the impact of retrieved wind fields from single Doppler radar for the simulation of convective events over India by MM5 model using three-dimensional variational (3DVAR) data assimilation modeling (Abhilash et al. 2007). In these experiments, wind fields derived from DWR radial wind applying uniform wind technique (UWT) are used in the assimilation cycle of MM5 model. The limitation of the UWT technique is that, due to the simple spatial dependence assumed for the wind field, this technique fails when the wind field is complex (Waldteufel and Corbin 1979). Zhao Kun and Xue Ming (2009) studied the impact of radar data on the analysis and prediction of the structure, intensity, and track of land-falling Hurricane Ike-2008, at a cloud-resolving resolution. Rigo et al. (2010) has analyzed warm season thunderstorms using an object-oriented tracking method based on radar and total lightning data. Recently, Srivastava et al. (2010a, b) carried out study on the assimilation of Indian DWR radial wind and reflectivity observations into ARPS model for the thunder storm and cyclone cases over Indian region using ADAS and ARPS3DVAR techniques.

The positive impact of these data in the prediction has motivated us to evaluate the sensitivity of assimilation of Doppler weather radar data, i.e., reflectivity (Z), radial velocity (Vr) data individually and in combination, to simulate mesoscale characteristics associated with land-falling cyclone. The prime objective of this study is to evaluate the impact of DWR data on mesoscale characteristics of a land-falling cyclone.

Current features of ARPS model are described in Sect. 2. A brief description of the cyclone NISHA is given in Sect. 3. Section 4 deals with the data sources and design of experiments, results are discussed in Sect. 5 and finally conclusions are summarized in Sect. 6.

2 Current features of the ARPS

Currently, a number of non-hydrostatic research and operational models (such as ARPS, WRF, MM5, RAMS) are available to the mesoscale modeling community. For the mesoscale modeling, the Advanced Regional Prediction System (ARPS) is an entirely new three-dimensional non-hydrostatic model known to handle storm-scale weather phenomena. It is a comprehensive multi-scale prediction system formulated in generalized terrain following coordinates that include data ingests quality control, objective analysis package, DWR parameter retrieval, and data assimilation procedure (Xue et al. 2003). The ARPS consists of four principal components: (a) programs to remap and super-ob the radar and satellite data to the analysis grid, (b) ARPS Data Assimilation System (ADAS) for analyzing all kinds of meteorological observations, (c) a cloud and hydrometeor analysis that applies diabatic adjustments to the temperature field, and (d) a non-hydrostatic forecast model. The assimilation can be performed as a sequence of intermittent cycles. An incremental analysis update procedure (IAU) can be employed in which the analysis increments are applied to the model state gradually over a period of time. The salient features of the ARPS version 5.2.5 are documented by Xue et al. (2003).

For assimilation of DWR radial velocity, ADAS uses a successive correction scheme, known as the Bratseth method (Bratseth 1986; Brewster 1996) and includes radial velocity adjustment. As in the optimum interpolation (OI) method, here also, correlations among the data must be specified. Typically, the correlations are a function of spatial distance. The total correlation is also affected by separation in the vertical. ADAS allows the vertical correlation to be specified as a function of height separation or a function of potential temperature separation. This ability is important when analyzing single-level data such as surface observations, which should influence a deep layer at times when the well-mixed boundary layer is deep and only a shallow layer when there is a nocturnal inversion.

With the assimilation of reflectivity observations, the cloud analysis procedure in the ADAS creates three-dimensional fields of cloud water, rainwater and improved fields of water vapor and temperature (Zhang et al. 1998). Identification of precipitation area is made from the use of model analysis filed of temperature, and then hydrometeor mixing ratio is determined utilizing radar reflectivity observations. Direct replacement of the background hydrometeors is done in areas where observed reflectivity is greater than a prescribed threshold (typically 10 dBZ). Precipitation is removed from the background in areas having reflectivity less than the precipitation threshold (within the radar volume coverage). An important aspect of building and maintaining thunderstorm updrafts in a non-hydrostatic model is the inclusion of the effect of latent heat release due to condensation processes in the updraft regions. A moist adiabatic ascent from the analyzed cloud base (with entrainment) is calculated and any excess in this temperature over the analyzed temperature is then added to the analyzed value. In order to take into account features like thunderstorm updrafts, the system incorporates latent heat adjustment in areas that have analyzed clouds and positive vertical motion. A moist diabatic ascent from the analyzed cloud base with entrainment is calculated and any excess temperature over the analyzed temperature is then added to the analyzed value. It is worth to be mentioned that recently 3DVAR analysis method is also incorporated for the ARPS (Gao et al. 2004). But this is not applied in this study.

3 Cyclone NISHA

Tamil Nadu coast of India was hit by a tropical cyclone named NISHA around 0000 UTC of 27 November 2008. It caused widespread rainfall with scattered heavy to very heavy and isolated extremely heavy category over coastal Tamil Nadu and Pondicherry. Exceptionally, heavy rainfall occurred over districts of Cauvery delta. It led to flood over coastal areas of Tamil Nadu and Pondicherry. There was loss of 78 human lives and huge crop loss. The area of influence of the cyclone and the corresponding observed track are shown in Fig. 1a, b.

The cyclonic storm was initially located as a low-pressure area over Sri Lanka and neighborhood on 24 November, concentrated into a depression at 0900 UTC of 25th and intensified into a cyclonic storm “NISHA” (T.No. 2.5) at 0300 UTC of 26th. It is then moved very slowly north-northwestwards and crossed Tamil Nadu and Pondicherry coast between 0000 UTC and 0100 UTC of 27 November. The three hourly observed track positions, estimated central pressure, maximum sustained wind, and pressure drop, etc., as reported by India Meteorological Department (IMD 2009) are given in Table 1.

Table 1 Best track positions and other parameters for cyclonic storm NISHA
Full size table

INSAT imageries of the system based on 0000 UTC observations of 27 November 2008 when the system lay close to the coast are shown in Fig. 2. The organization of mesoscale convection over the region is clearly depicted. After crossing the coast, the system came closer to DWR, Chennai, features grew prominent and center could be fixed, based on a few spiral bands and in some cases with partial eye wall. Center was fixed during 0500 to 1300 UTC of 27 November. According to the radar report of IMD (2009), maximum velocity observed in the cyclone field was not associated with the eye-wall region, but mostly associated with strong echoes in spiral bands. Maximum observed radial wind was around 55 knots. According to Tamil Nadu, government reports 8 lakhs acres of paddy crop in Nagapattinam, Thanjavur, and Tiruvarur (delta) districts and 55,250 hectares of paddy crop in Cuddalore district were submerged due to heavy rain. The areas affected by flood and wind of Tamil Nadu state as documented by India Meteorological Department (IMD 2009) are shown in Fig. 3a, b. In Fig. 3a, the area affected due to the very heavy rainfall is shaded by slant lines and flooded area by crossed lines. In Fig. 3b, the area affected by strong wind is shaded by slant lines.

Fig. 2
figure 2

Cyclone NISHA as observed by INSAT KALPANA-1 at 0000 UTC of 27 November 2008

Full size image
Fig. 3
figure 3

a The area affected by flood and b the area affected by strong winds due the cyclonic storm

Full size image

4 Data sources and design of experiments

The scan strategy was optimized to obtain DWR observations (Sen Roy et al. 2010). It takes about 10 min to complete one scan, detail descriptions of the scan strategy are given in Table 2. For quality control of these raw data, before ingesting into the rapid updates assimilation cycle of ADAS, application software (Lakshmanan et al. 2006) called “Warning Decision Software Integrated Information (WDSSII)” is used. This removes noise and ground/sea clutter and unfolds velocity.

Table 2 Ten minutes scan strategy being followed by DWR sites (Range step = 0.5 km, Clutter filter width = 1 mps)
Full size table

The size of the model domain selected for this study is 901 × 901 km with 9-km grid spacing, and Chennai (latitude 13°N & longitude 80°E) is kept at the center of the domain. The vertical grid stretched from 20 m at the surface to about 400 m at the model top, which is located at about 15 km height. The detailed design of the experiment is shown in Table 3. For the experiments, Kain–Fritsch cumulus parameterization scheme, Kessler warm rain microphysics and 1.5 TKE turbulent mixing option are used. Also, all moist processes are activated for the experiments. The model ARPS (version 5.2.5) is run with background and boundary conditions from the National Center for Environmental Prediction Global Forecast System (NCEP GFS) freely available in the Internet at the resolution of 1° × 1° lat/long. NCEP GFS analysis field interpolated at 9-km resolution is considered as initial condition in which radar data are incorporated. No other observations are included with the assumption that NCEP analysis includes all other observations. The NCEP GFS fields are interpolated using EXT2ARPS module available with ARPS model for ADAS grid at 9-km resolution. Half-hour intermittent assimilation cycles are performed within 3 hour assimilation window from 0000 UTC to 0300 UTC. The life cycle used for the IAU assimilation window is 30 min. Four sets of numerical experiments are conducted. Four sets of forecasting experiments are as follows: (a) using NCEP model analysis to initialize the ARPS model (arps_con); (b) using initial condition produced from the ADAS with radial velocity data (arps_vel); (c) using initial condition produced from the NCEP analysis fields and cloud analysis with reflectivity data (arps_ref); and (d) using initial condition produced from the ADAS with radial velocity and cloud analysis with reflectivity data (arps_both).

Table 3 Model configuration
Full size table

The DWR observations are first preprocessed for the quality control and converted into the desired format of model assimilation. Then, radial velocity data are analyzed using ADAS, while reflectivity data are used through the cloud analysis procedure. The temperature adjustment scheme based on the moist adiabatic temperature profile is used in the cloud analysis scheme. Figure 4 shows the 3 hour assimilation window with radar data assimilation at every 30-min interval for three hours (during the hour 0000 UTC–0300 UTC of 27 November 2008) and then 21 hour forward forecasts (valid up to 0000 UTC of 28 November 2008) with no additional data input.

Fig. 4
figure 4

Assimilation cycle for DWR data into ARPS model

Full size image

5 Results and discussions

5.1 Impact of DWR data assimilation on reflectivity forecast

The model simulated reflectivity fields by various experiments against the observed reflectivity valid at 0600 UTC of 27 November 2008 are presented in Fig. 5a–e. Three main cells are detected in each experiment, and these are marked as A, B and C, respectively. An intercomparison of these results reveals that they appeared as mesocells in panel (c) and (d). Spiral rain band structure is clearly seen in (c) and (d). The pattern in the panel (c) and (d) is close to observed reflectivity as produced by DWR, Chennai, in the panel (e). However, intensity of A is found to be slightly overestimated in the panel (c). Panel (a) and (b) failed to capture these mesoscale features. Due to assimilation of both radial wind and reflectivity data, vertical wind, specific water vapor, and cloud water content are adjusted in such a way that the model could produce more realistic mesoscale convective structure. This explanation is also suggested by Haase et al. (2000).

Fig. 5
figure 5

Intercomparison of reflectivity fields (dBZ) of different simulation experiments against the observed field valid at 0600 UTC of 27 November 2008: a reflectivity by arps_con experiment, b reflectivity by arps_vel experiment, c reflectivity by arps_ref experiment, d reflectivity by arps_both experiment, e observed reflectivity

Full size image

The model simulated reflectivity fields by various experiments against the observed reflectivity valid at 0900 UTC of 27 November 2008 are shown in Fig. 6a–e. The results of arps_both experiment show that cells A, B, and C continued to persist along with two other newly formed cells. Spiral rain band structure is maintained, and maximum intensity corresponding to the cell A and B is found to be similar to that of the observed field at 0600 UTC. The pattern is found more organized The arp_ref experiment also could retain three cells and rain band structure at 0900 UTC. In the numerical experiments of control (panel a) and arp_vel (panel b), the model failed to capture organized mesoscale rain band structure. When both reflectivity and radial velocity are assimilated in the model (arps_both experiment), features like number of cells, spiral rain band, and intensity are captured.

Fig. 6
figure 6

Intercomparison of reflectivity fields (dBZ) of various simulation experiments against the observed field valid at 0900 UTC of 27 November 2008: a reflectivity by arps_con experiment, b reflectivity by arps_vel experiment, c reflectivity by arps_ref experiment, d reflectivity by arps_both experiment, e observed reflectivity

Full size image

The patterns in the forecast charts show a large convection area with cyclonic curvature and some mesocells embedded in it. However, the locations of cell B in both (c) & (d) panel of Figs. 5 and 6 are far apart from each other by more than a degree latitude. Also, location of cell B in both the forecast differs widely from the observed location that is very close to the coast, while in the forecast these are very much in the interior. These differences may be the result of usual forecast limitations.

5.2 Impact of DWR data assimilation on wind and vorticity forecast

Figures 7a–d and 8a–d illustrate model predicted wind fields superimposed vorticity fields of 850 hPa by various experiments at 0600 UTC and 0900 UTC, respectively. Red circle shows the observed position of the system in each diagram. The system lay near lat. 11.75°N/long. 79.25°E at 0600 UTC and near lat. 12°N/long. 80°E at 0900 UTC. In Fig. 7, it is noticed that the center of the system in all the experiments has been close to the observed position, except that of arps_vel experiment. At 0600 UTC, system lay as a cyclonic storm with surface winds of the order 40 knots. The strength of 850 hPa wind field in the arps_both experiment is of the order 40–45 knots. In the experiment of arps_ref, wind speed is found to be marginally stronger as compared to all other experiments.

Fig. 7
figure 7

Intercomparison of the wind field (knot) superimposed vorticity field (×10−5 s−1). of various simulation experiments valid at 0600 UTC of 27 November 2008: a arps_con experiment, b arps_vel experiment, c arps_ref experiment, and d arps_both experiment

Full size image
Fig. 8
figure 8

Intercomparison of the wind field (knot) superimposed vorticity field (×10−5 s−1). of various simulation experiments valid at 0900 UTC of 27 November 2008: a arps_con experiment, b arps_vel experiment, c arps_ref experiment, and d arps_both experiment

Full size image

The center of the system at 0900 UTC in arps_ref experiment is found to be closer to the observed position (Fig. 8). At 0900 UTC, when the system lay as a deep depression, the arps_both experiment showed speed of 850 hPa winds of the order of 50 knots, which is consistent with the DWR reported maximum wind speed of 55 knots. In the arps_vel, the corresponding wind speed has been of the order 30–40 knots. Thus, the intercomparison of these results shows that when both reflectivity and radial velocity are assimilated in the model (arps_both experiment), then the features like center of the system, wind strength, and overlaid vorticity are simulated more accurately as compared to other experiments.

5.3 Impact of DWR data assimilation on rainfall forecast

Figure 9a–d shows model predicted (24 h) rainfall by various experiments. Figure 10 depicts the corresponding observed rainfall reported by rain gauge stations in the area. It is clear from Fig. 9 that when only radial velocity is assimilated into the model, the east west rain band between Chennai and Nellore is better simulated. The maximum predicted rainfall amount in this band is found to be 32 cm. With the assimilation of only reflectivity, the model could capture east west oriented rain band between Chennai and Nellore with rainfall maximum of order 54 cm. When both radial velocity and reflectivity are assimilated into the model, then east west oriented rain band between Chennai and Nellore becomes more marked with rainfall maxima of order 81 cm. Apart from this, one more meso-rain cell, with maximum rainfall amount of 40 cm is noticed between Trichy and Salem. The rainfall pattern between Selam and Trichy (Fig. 10) is found to be well matched with that of observed rainfall pattern. The control experiment failed to capture this rain band pattern associated with the cyclonic system.

Fig. 9
figure 9

Model predicted 24-h accumulated rainfall in cm (valid at 0000 UTC of 28 November 2009 based on various experiments: a arps_con experiment, b arps_vel experiment, c arps_ref experiment, and d arps_both experiment

Full size image
Fig. 10
figure 10

Observed 24-h cumulative rainfall in cm (valid at 0300 UTC of 28 November 2009

Full size image

Table 4 displays an intercomparison of predicted rainfall in various experiments against the observed rainfall. The result shows that the predicted rainfall by arps_both and arps_ref experiments has been very close to the observed rainfall. Very heavy rainfall over Chennai (28 cm) is well captured in arps_both experiment in which the predicted rainfall has been of the order of 30 cm. For Chennai and Nellore stations rainfall is well comparable with the observed fields, for Vellore and Salem it is overestimated and for Cuddalore it is underestimated. Though rainfall over Trichy is not captured, meso-rain cell has been well simulated in arps_both experiment.

Table 4 Observed and predicted rainfall in various experiments
Full size table

The results show that the arps_both experiment (when both reflectivity and radial velocity are assimilated) could capture rainfall distribution as well the rainfall intensity reasonably well.

6 Concluding remarks

Installation of four GEMATRONIC METEOR 1500S model DWRs (during 2002–2006) has heightened the prospects to ingest the high dense radial wind and reflectivity observations of Indian DWR in the assimilation cycle of a numerical model. But, it remained a challenging task over the years due to number of issues such as optimization of scan strategy, networking for real-time data reception, conversion of data from VOL format to nonproprietary open source NetCDF format and the quality control of these data. Recently, these data in the required format are made available for NWP applications. Thus, the main objective of this study has been to ingest the radial velocity and reflectivity observations of DWR, Chennai, in the assimilation cycle of the mesoscale NWP system (ARPS) and to examine the sensitivity of assimilation of Doppler weather radar data, i.e., reflectivity (Z), radial velocity (Vr) data individually, and in combination, to simulate mesoscale characteristics associated with land-falling cyclone.

The intercomparison of the results of these experiments reveals that mesoscale characteristics of cyclone NISHA are well simulated when both reflectivity and radial velocity observations are used for the assimilation. The results demonstrate that the model with the assimilation of DWR observations could capture features like number of cells, spiral rain band structure, and intensity of each cell in arps_both experiment. The evolution of vorticity overlaid on horizontal wind at 850 hPa shows that the forecast is improved in terms of the magnitude and direction of lower tropospheric wind, time, and location of cyclone in the experiment arps_both. With the assimilation of both radial wind and reflectivity fields, vertical wind, specific water vapor, and cloud water content are adjusted in such a way that the model could produce better mesoscale convective features. The results of this study are very encouraging, particularly in the very short-range forecast scale. However, these results are preliminary, further investigation is required for a variety of different storm types and with single Doppler retrieval as well as dual Doppler retrievals algorithms utilizing observations from an adequate network of DWR. In the near future, India Meteorological Department (IMD) is going to install a good network of DWRs. The present work demonstrates important steps of transforming research to operations for improving mesoscale NWP capabilities over Indian region.