Grand Challenges: Effective and Equitable Weather Forecasting in a Changing Climate with Machine Learning

Artificial intelligence (AI) and machine learning (ML) are assuming a growing role in all the physical sciences. However, naively applied, ML may replicate and even amplify structural inequities that exist in environmental science data and weather observation networks. Environmental science is already plagued by systemic inequities. For instance, black populations are underserved by existing weather radar systems and may therefore receive less accurate forecasts. Many of these same marginalized communities already struggle to recover from extreme weather events and are also expected to experience disproportionate impacts from climate change. Thus, it is critical for society that applications of ML in weather forecasting are ethical, fair, and trustworthy. This project seeks to characterize and improve ML-based weather forecasting, with an emphasis on designing systems that are both accurate and equitable. This goal will be achieved across four research thrusts: (1) quantification and estimation of uncertainty, (2) quantification and estimation of inequity, (3) development of accurate and equitable algorithms, and (4) development of an accurate and equitable sensor placement strategy. These four research thrusts will significantly and meaningfully advance the science of weather forecasting by expanding beyond the traditional foci on just physical processes to include uncertainty and societal considerations in the forecasting workflow.