Two-Stage Artificial Intelligence Algorithm for Calculating Atmospheric Motion Vectors

Abstract:

Wind velocity observations are important for understanding the atmosphere and weather forecasting. Geostationary and polar-orbiting satellite imaging can provide wind velocity fields through feature tracking, producing Atmospheric Motion Vectors (AMVs). However traditional feature tracking, which merely looks at the evolution of pixel values, without incorporating any physical insight, produces relatively large errors. Here we propose a two-stage Artificial Intelligence algorithm that corrects the noisy AMVs from feature tracking using structures learned from Numerical Weather Prediction (NWP) fields. In other words, we use the physical insight from NWP to automatically correct AMVs. Our two-stage approach is found to perform significantly better than a traditional algorithm.