Forecast Beyond Accuracy

As a research team, we build scalable, uncertainty-aware algorithms for high-impact weather events: extreme precipitation and atmospheric rivers, hail, and damaging winds. We focus on forecasting not just whether an event will occur, but its location, timing, and severity, how risk is changing, and how much confidence each prediction deserves. Accuracy tells how a model performed; uncertainty tells how far to trust it.

Specifically, our methods of interest span analog forecasting, deep learning, and generative models for applications including model downscaling and calibration. Our research and analysis rely heavily on geospatial data like satellites, radars, and gauges, and crowd-sourced observations, as well as numerical simulations like MPAS. This suite of data and the capability to process them, thanks to UGA's supercomputer Sapelo2 from GACRC, enable our research on extreme events, risk management, and water resources. To integrate Geoinformatics and AI is not just to interpret the Earth as it is, but to anticipate how it will evolve in the future, both near and far.