An Artificial Neural Network Downscaling of MERRA Mountain Gap Wind Events
Whereas numerical models capture the MGW precursor synoptic scale patterns, they often fail to adequately resolve the interactions of large scale flow with small scale terrain features. This study evaluates the utility of an Artificial Neural Network (ANN) for the downscaling of MGW events simulated by models with coarse grid spacing. The inputs to the ANN include MERRA surface fields. The Cross-Calibrated, Multi-Platform (CCMP) ocean surface wind product is used as ground truth to train the ANN. Preliminary results show that the ANN is able to generate wind fields that are very similar to corresponding CCMP observations. The technique is being extended by using 1 kilometer grid spacing numerical model outputs to train the ANN and conduct downscaling to finer spatial scales. Since the ANN is computationally efficient, it can be deployed in an operational setting to downscale low-resolution numerical model output. The training and performance evaluation of the ANN will be presented.