Accurate and reliable forecasts of precipitation are critical to providing preparation time for water resource management officials, energy and agriculture sectors, and citizens

Recent work by arcodia, barnes, ET AL. (2023) found that neural networks are capable of identifying predictable states of the climate system within climate models that are useful for making confident, accurate subseasonal precipitation predictions in the real world. The aim of the proposed work is to extend this methodology to improve real-time U.S. Week 3-4 precipitation forecasts with uncertainty quantification. We will leverage explainable machine learning approaches to identify forecasts-of-opportunity (e.g. regions and climate states that are more predictable) and large-scale sources of predictability. The lead PI and CO-PIS research has designed and implemented explainable artificial intelligence (XAI) approaches for advancing data-driven subseasonal-to-seasonal climate predictions, including assessing prediction uncertainty. Of note is the focus on leveraging machine learning to identify forecasts-of-opportunity and applying XAI techniques to understand the physical sources of predictability. This approach will be applied to produce a real-time forecasting tool to identify when, where, and why a week 3-4 precipitation forecast can be considered a forecast-of-opportunity, ultimately assessing skill and confidence in forecasts. The goals of the proposed work are threefold: (1) refine and extend the methodology of arcodia, barnes, ET AL. (2023) to quantify skill and evaluate predictability of week 3-4 U.S. Precipitation forecasts by applying machine learning approaches to identify forecasts-of-opportunity in model forecasts, (2) identify regional state-dependence of predictability and its sources across climate models and datasets, and (3) develop a machine learning-based tool to be used by CPC week 3-4 forecasters to identify forecasts-of-opportunity and climate state-dependent predictability sources in real-time. The activities proposed will be performed by scientists at the climate prediction center (CPC) and colorado state university (CSU), who have extensive experience working together and transitioning similar empirical tools into CPC operations.