Weather forecast using WRFg at 2 km resolution over two provinces in northwestern Panama

Abstract

In this paper, the 24- and 48-hour precipitation and temperature forecasts generated by dynamic downscaling (DDS) using the WRF-ARW (running on CPU) and AceCAST (running on GPU) models are analyzed for the western region of Panama (particularly in the provinces of Bocas del Toro and Chiriquí) during hurricanes ETA and IOTA. Various microphysics and cumulus parameterization schemes are used to generate rainfall forecasts with 11 and 2 km resolution, then these forecasts are compared against measurements from weather stations. The results indicate that under the ETA and IOTA events, the rainfall generated by BMJ microphysics scheme and Kain-Fritsch cumulus parameterization was the most similar to the observed rainfall data. On the other hand, it was found that the computation time of the forecasts obtained by AceCAST (GPU) was at least two times shorter than the WRF-ARW (CPU) model, thus using the computing power of GPUs to generate weather forecasts significantly reduces the issuance time of forecasts.