Learning Architecture for Short-Term Forecasts of Cloud Characteristics Using Satellites (LAST FOCUS)

The Learning Architecture for Short-Term Forecasts of Cloud Characteristics Using Satellites (LAST FOCUS) introduces a novel approach and state-of-the-art ML methodologies to forecast cloud properties including, but not limited to, ceiling, optical depth, phases, and reflectance.Ā Using a suite of cloud-observing geostationary and Low-Earth Orbiting (LEO) satellites, atmospheric profiles, and an extensible ML ecosystem for weather modeling, LAST FOCUS produces outputs for aviation operations.Ā LAST FOCUS is interoperable with and extends existing capabilities including Navy Research LaboratoryÆs (NRL) Geo-located Information Processing System (GeoIPS) and the Weather Machine Learning Platform (WxMLP) Forecasting cloud characteristics is difficult.Ā Multiple small-scale physical processes and thermodynamics must be precisely represented in models in order to accurately depict cloud development in time and space.Ā For enterprise weather forecasting and modeling, precipitation is of much higher interest than clouds, and validating and improving the model characterization of forecasted precipitation has been quite rigorous compared to that of modeling non-precipitating clouds.Ā The goal of LAST FOCUS is to overcome the numerous challenges associated with reliably predicting cloud characteristics for aviation by using observational cloud characteristic data and time series deep learning convolutional neural networks as a starting point and moving toward a multivariate CNN approach using forecasted environmental variables in the CNN to constrain unrealistic forecasts.