The blending is done by a weighted average of DARTS nowcast and WRF forecast model. The weights are a function of lead time. The DARTS nowcast is highly weighted in first 30 minutes, and WRF model is highly weighted after 2 hour lead time. The hyperbolic tangent function is used to estimate the weights of WRF model between 30 minutes and 2 hours. The blended nowcast is mainly based on DARTS nowcast in first 30 minutes, and the weights of the WRF model increases gradually from 30 minutes to 2 hours. After 2 hours the blended nowcast is mainly based on WRF model. The blending parameters are investigated in this paper to provide the best nowcast.
This study also investigates the performance of using high-resolution WRF model against HRRR model forecast which is 3 km grid resolution. The HRRR model uses the Three-Dimensional Variational (3DVAR) data assimilation techniques to assimilate the radar data in every 15 minutes interval. In this study, the NEXRAD and CASA radar data are assimilated every 5 minutes through 3DVAR, and the forecast is compared with HRRR model. Evaluating the complexity and benefits of assimilating high resolution model is the focus of this part of the research.
Thirdly, this study also compares the results from direct (reflectivity) and indirect (hydrometer) radar data assimilation techniques. The Model Evaluation Tools (MET) verification package is used to estimate the statistical skill scores of nowcasting and Method for Object-Based Diagnostic Evaluation (MODE) is used to analyze the predicted and observed storm location and structure.