The specific objectives of this project are to: (1) quantify the skills, uncertainties, and biases in predicting the one to ten year forecast window at time scales in different regions of the world, and for different atmospheric models, through comparison with high-quality precipitation observations; (2) develop a machine-learning-based methodology that utilizes precipitation observations for downscaling and improving the accuracy of one to ten year precipitation predictions, and validate the results through comparison with high-quality precipitation observations; and (3) develop a toolkit that automates the (i) computation of skills, uncertainties and biases in one to ten year projections for any region of the world or atmospheric model specified by the user, and (ii) downscaling of one to ten year predictions for any region of the world.

The project will use state-of-the-art statistical evaluation and downscaling methods to evaluate and downscale one to ten year models based on high-quality precipitation observation datasets as reference. It will consider all publicly available global one to ten year models, including currently active models from the sixth Coupled Model Intercomparison Project. The research will be conducted across the continental United States, as well as globally. The reference precipitation observation datasets consist of a gridded rain-gauge product across the United States, and a satellite-based product across the globe. This will be the first time these two relatively high-quality precipitation observations will be used for evaluating one to ten year precipitation predictions. The project will employ comprehensive metrics (both traditional and emerging) for evaluation and new machine learning (ML) approaches for downscaling and calibration.

The Department of Defense (DoD) relies on a variety of public forecasting approaches for planning and preparedness. By quantifying uncertainties associated with current one to ten year atmospheric models and by developing new ML approaches for downscaling and improving the accuracy of one to ten year models, the project will provide best available and actionable precipitation predictions (along with estimation uncertainty) that meet DoD needs for weather and natural hazards exposure assessments. (Anticipated Project Completion - 2026)