Prospects for Machine Learning Activity within the United States National Weather Service

Abstract

Abstract The National Weather Service (NWS) Office of Science and Technology Integration commissioned a report to assess the status of artificial intelligence (AI) and machine learning (ML) activity within the agency with a view toward identifying existing obstacles and recommending future directions. The purpose of this essay is to communicate the steps that the NWS plans to take to realize the potential benefits of AI in operations. AI activities are growing rapidly within atmospheric sciences, and the NWS is part of this growth. However, the activity is fragmented and lacks the needed infrastructure for improved coordination of effort. Current obstacles to progress include insufficient workforce training in AI/ML, a lack of curated datasets and software that can be used for development and evaluation of these approaches, the absence of a centralized clearing house available to NWS personnel for technical expertise and consultation, limited operational compute resources, and a lack of a clear end-to-end project pathway that encompasses exploration, development, testbed/proving ground, and operational implementation. These limitations are addressable. Training materials specific to NWS interests can be developed through collaboration with existing NOAA centers. Establishing a reference library staffed with AI/ML consultants tasked with collaborating with operational units would reduce siloed efforts and enhance productivity. Establishing funding vehicles for theme-based projects with a sustainable pathway through operational implementation would help bridge the research-to-operations “valley of death.” Given the growth of AI/ML across the U.S. Weather Enterprise and the already substantial involvement of academic and private sector entities, these developments within the NWS will be of interest to the atmospheric science field.