Cloud Characterization with a Convolutional Neural Network Using Ground Weather Radar Scans

Accurate cloud characterization is essential for reducing risk in government and industrial activities such as rocket launches. Current ground weather radar scan strategies use volume coverage patterns (VCPs) to characterize both nonprecipitating clouds and severe weather thunderstorms for use in forecasting. Clear-air mode VCPs leave unscanned gaps in elevation bands, which are linearly interpolated to characterize the clouds. Current interpolation methods overestimate the vertical gradient and report additional risk, resulting in fewer launch opportunities. This research shows that a convolutional neural network improves cloud characterization accuracy over the current interpolation results leading to more accurate forecasting.