THE INITIAL CONDITION AND EXPLICIT PREDICTION OF CONVECTION USING ARPS ADJOINT AND OTHER RETRIEVAL METHODS WITH WSR-88D DATA

Jidong Gao¹, Ming Xue¹, Zhi Wang¹ and Kelvin K. Droegemeier^1,2

Center for Analysis and Prediction of Storms¹ and School of Meteorology²

University of Oklahoma, Norman, OK 73019

I. INTRODUCTION

The WSR-88D Doppler radar has the ability to scan the atmosphere at high spatial and temporal resolutions, but direct measurements are limited to reflectivity and radial velocity. There is no direct measurement of either the complete wind field or thermodynamic (T and P) variables, both of which are required to initialize storm-scale NWP models.

For the reason, the analysis and retrieval of the complete 3-D wind-field, temperature, and cloud microphysical information from radar data are attracting a great deal of attention. Several methods for retrieving unobserved meteorological variables from single-Doppler radar data have been developed in recent years at the Center for Analysis and Prediction of Storms (CAPS) (Zhang and Cal-Chen 1996; Xu et al. 1994; Qiu and Xu 1996; Shapiro et al. 1995a, 1996; Sun et al. 1991). They can be divided roughly into two categories: those in which three-dimensional winds are determined first, then the pressure and buoyancy retrieved using the winds; and those that attempt to retrieve winds and other state variables simultaneously from single-Doppler and other observations.

In order to assess the influence of WSR-88D Doppler radar data on a deep-convective storm, experiments have been performed using both methods for a VORTEX-95 squall line case. These experiments focus on the impact of the KTLX (Twin Lakes, Oklahoma) WSR-88D level II data on the initial fields and on subsequent forecasts.

II. DATA ASSIMILATION METHODS

2.1. Adjoint Method

The ARPS (Xue et al. 1995) is a general-purpose, nonhydrostatic, compressible model designed for storm- and mesoscale atmospheric simulation and real-time numerical weather prediction. The complete model solves equations for momentum, potential temperature, pressure, water substances, and subgrid-scale turbulent kinetic energy. It also includes various physical processes. In this study, the dry version of the forward model and it's adjoint are used in the assimilation and retrieval while the full physics version is used for the prediction once the initial condition is obtained.

Sun et al. (1991) successfully retrieved the three-dimensional wind and thermodynamic variables from model-generated observations of a single velocity component and reflectivity. This technique has since been applied successfully to observed single-Doppler data (Sun et al. 1997). In the past a few years, the ARPS adjoint has been developed and tested (Wang et al. 1997). It, together with the forward model, forms a 4D-VAR system that can be used to retrieve 3-D wind and thermodynamic fields needed for model initialization. The cost function used in this study is defined to measure the distance between the radial velocity derived from model variable and corresponding observations, and the distance between initial analysis and the background fields (obtained from ARPS Data Analysis System - ADAS). The minimization algorithm used here is Liu and Nocedal's limited memory, quasi-Newton conjugate gradient method (Liu 1989).

2.2. SDVR-Thermodynamic Retrieval Method

This method involves using a sequence of radar volume scans and a Single-Doppler Velocity Retrieval (SDVR) technique followed by variational wind adjustment and thermodynamic retrieval/recovery so as to obtain an estimate of the 3-D wind and thermodynamic fields. Several SDVR techniques developed at CAPS can be used here, including Shapiro's two-scalar algorithm, Xu's least-square and simple adjoint methods, and Gal-Chen and Zhang (1996) retrieval method. In this study, we choose the simple least-square method (Qiu and Xu 1996) to retrieve the complete 3-D wind. The other parts of the procedure are the same as described in Shapiro et al. (1996), except that the retrieved winds, temperature and pressure are used directly as the ARPS initial condition instead of going through ADAS again.

III. NUMERICAL EXPERIMENTS

Experiments have been performed using above methods for a VORTEX-95 May, 7 squall line case. The radar reflectivity fields from the KTLX radar at 2000 UTC (Fig.1) show a squall line passing through central Oklahoma.

Using analyzed data from NCEP RUC (Benjamin et al. 1994) as a background, we performed three sets of experiments. They include (1) a control run from an initial condition created by ADAS without radar data; (2) runs with initial fields obtained from the SDVR-thermodynamic retrieval method (part of the ARPS forward data assimilation procedure) using 5 volume scans of radial velocity spanning a time period of 24 min; and (3) runs with initial fields retrieved using 4D-VAR with radial velocity over a period of 6 min. and two radar scans. All model runs start at 1800 UTC. We run the ARPS for 3 hours on a 5 km grid covering 285 x 285 km². This domain is nested one-way within a 9 km coarse grid covering 900x900 km². For the coarse grid, the RUC analysis was used as the initial condition and the RUC forecast fields as boundary conditions. The results from the coarse grid run were used to force the fine grid boundaries. Our comparison experiments were performed on the fine grid only.

Fig 1. 2000 UTC reflectivity field at 1.5deg. elevation from the KTLX WSR-88D Doppler radar on May7 1995

IV. RESULTS

We now compare the initial condition and predictions created from these experiments. Fig 2 depicts the initial analyzed wind vectors and perturbation potential temperature fields at z = 2 km. The basic structure of the initial wind fields (Fig. 2c) created by the ARPS adjoint agree qualitatively with the dual-Doppler analyzed fields (not shown). The retrieved perturbation potential temperature ([theta]') and vertical velocity fields are in good agreement with what is expected from conceptual models, although the retrieved [theta]' is rather noisy.

The most notable differences in predicted reflectivities can be found in Fig.3. The areal coverage and intensity of precipitation in northern Oklahoma is reasonably predicted in Fig 3c by using the initial fields created by ARPS adjoint. The structure of the initial wind fields created by ADAS, without radar data, appears uniform (Fig 2a). Temperature disturbances and vertical velocities are very weak. It is obvious from Fig 3a that the location of precipitation is not well predicted compared to Fig 1, especially in northern Oklahoma. The initial fields obtained from the ARPS SDVR-thermodynamic retrieval method is not as good as the ARPS adjoint, but is better than that obtained from ADAS without radar data (Fig 2b). The amplitudes of temperature perturbations and vertical velocities in the initial condition are weaker than those of adjoint case but are larger than when using ADAS analysis. The forecasted convective elements in northern Oklahoma are also relatively weak (Fig 3b), although they were stronger at earlier times but subsequently moved out of the north boundary. It should be pointed out that our retrieval and prediction using the SDVR-retrieval method is not as good as that of Shapiro (1996). The difference appears to be mainly because that we did not perform re-analysis using the retrieved wind and thermodynamic fields. Instead, these fields were interpolated to the ARPS grid and used as the initial condition directly, thus limiting the area of influence by the retieved data to the vicinity of observations.

V. CONCLUSION AND FUTURE DIRECTION

Although the above real data tests using the ARPS adjoint and SDVR-thermodynamic retrieval schemes are still preliminary, they suggest the importance of radar data in storm-scale NWP. the injection of radar data is found to be beneficial for the spinup of convection no matter which method is used.

For the results presented here, only the dry version of ARPS adjoint was used. We have finished the adjoint of moist processes in the ARPS and plan to assimilate the observation of reflectivity into the model with moist processes. We will also test the impact of other data, such as mesonet and aircraft data.

VI. ACKNOWLEDGMENTS

Limin Zhao, Alan Shapiro and Stephen Weygandt provided us the WSR-88D radar data and the program to process these data. This research was supported by the Center for Analysis and Prediction of Storms (CAPS) at the university of Oklahoma under grant ATM91-20009 from the National Science Foundation (NSF).

REFERENCE

Benjamin, et al., 1994: The Rapid Update Cycle at NMC. Preprints, 10th Conf. on Num. Wea. Pred., Portland, OR, Amer. Meteor. Soc., 566-568.

Brewster, K., 1996: Application of a Bratseth analysis scheme including Doppler radar data. Preprints, 15th conf. on Wea. Analysis and Forecasting, Norfolk, VA, Amer. Meteor. Soc. Companion Volume.

Liu, D. C., 1989: On the limited memory BFGS method for large scale optimization. Math Programming, 45, 503-528.

Qiu, C. J., and Q. Xu, 1996: Least squares retrieval of Micro burst winds from single-Doppler radar data, Mon. Wea. Rev. 124, 1132-1144.

Shapiro, A., et al., 1995a: Single-Doppler velocity retrievals with Phoenix II data: clear air and microburst wind retrievals in the planetary boundary layer. J. Atmos. Sci., 52, 1265-1287.

Shapiro, A., et al., 1996: Initial forecast fields created from single-Doppler wind retrieval, thermodynamic retrieval and ADAS. Preprints, 11th conference on Numerical Weather Prediction, 19-23 Feb., Amer. Meteor. Soc., San Francisco, CA., 551-554.

Sun, J., et al., 1991, Recovery of three-dimensional wind and temperature fields from simulated single-Doppler data, J. Atmos. Sci., 48, 876-890.

Sun, J., et al., 1997: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part I: Model development and simulated data experiments. J. Atmos. Sic., 54, 1642-1661.

Wang, Z., et al., 1997: Application of a new adjoint algorithm to the 3D ARPS storm-scale model using simulated data. Mon. Wea. Rev., 125, in press.

Xu, Q., et al., 1994: Adjoint-method retrievals of low-altitude wind fields from single-Doppler reflectivity measured during Phoenix II. J. Atoms. Oceanic. Tech., 11, 275-288.

Xue, M, et al., 1995: ARPS Version 4.0 User's Guide, Available from CAPS, University of Oklahoma.

Zhang, J., and Gal-Chen, T. 1996: Single Doppler wind retrieval in the moving frame of reference, J. Atmos. Sci., 53, 2609-2623.