Machine Learning in Earth System Science Applications Using Satellite Data

Machine learning (ML) methods are enabling usage of the large amounts of satellite data available for Earth system science applications. Traditional ML and newer deep learning methods have proven their value for applications such as predicting surface irradiance for solar power grid integration planning, providing better data for modeling wildland fires, predicting convection that leads to severe weather, and predicting lightning, among many other applications. This papers surveys some of these applications, the ML methods enabling them, how those methods are becoming more trustworthy, and discusses prospects for future advances.