Sbir phase i: a physically informed machine learning model for subseasonal forecasting of extreme temperatures

This National Science Foundation (NSF) Technology, Innovation, and Partnerships (CFDA 47.084) Project Grant award in the amount of $274,912 provides funding to Salient Predictions Inc., a for-profit small business, to develop a machine learning model that can forecast extreme heat or cold events up to one month in advance. The goal is to create a predictive model that combines weather forecasting technologies with machine learning to deliver more accurate sub-seasonal temperature forecasts, enabling industries like energy and agriculture to better prepare for and mitigate the impacts of extreme temperature events. This project aims to address current limitations in extreme weather forecasting, such as the lack of operational products and insufficient lead time for communities to take protective action. The award period runs from September 1, 2024, to August 31, 2025. No sub-awards are planned under this grant.

Abstract

The broader/commercial impact of this small business innovation research (sbir) phase i project is in helping industries such as energy and agriculture increase protective measures and improve resiliency in the face of extreme temperature events. Sub-seasonal predictions on extreme weather should help customers optimize performance, mitigate risk, improve resiliency, and plan initiatives up to a month in advance. Although instruments like satellite and radar are highly accurate in detecting these extreme events at short time horizons, not enough time is given for businesses and communities to take action to ensure their survival. For a community that might be impacted by extreme heat or cold, this is not enough time to take action to mitigate crop damage, protect power generation facilities, etc. For example, extreme cold in texas brought on by the ICE storm URI placed immense pressure on power grids, costing >$100 billion over several days. The company believes that the opportunity for increased knowledge and greater lead time to make strategic business decisions like such as advanced contracting with maintenance crews, foliage management, stockpiling replacement parts, insulating critical components, or installing DE-icing equipment will drive customer adoption of the sub-seasonal model over current models. The combination of a changing climate, a chaotic atmosphere and the relative rarity of extreme events makes forecasting extreme heat and cold events at a sub-seasonal timescale a particularly challenging problem. This project focuses on developing a machine learning model to improve forecasting extreme heat or cold 1-month out. There are three significant problems with current forecasting systems: 1) historical analogs that are used to develop these models are becoming increasingly irrelevant, (2) day to one-week timeframes do not allow enough time for communities to prepare for extreme events, and (3) lack of operational products. This project aims to build a proprietary model that combines the technologies underlying current weather forecasting tools with improved machine-learning powered models and the ocean, land, and atmospheric data that highly influence extreme heat or cold. Such a model may enable more accurate weather predictions 1-month out. A predictive model capable of delivering precise forecasts over a sub-seasonal timeframe, integrating the constantly shifting dynamics of the atmosphere attributed to climate change, would empower industry (namely energy and agriculture) to effectively prepare for extreme temperature events to best serve and protect citizens. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.