Project titlerefining ensemble climate model projections for the hawaiian islands:highresolution spatial downscaling of precipitation for flood risk withgenerative machine learningproject DATES8 1…

This Cooperative Agreement awarded by the U.S. Geological Survey (USGS) under the National and Regional Climate Adaptation Science Centers (CFDA 15.820) program will provide $475,793 to the University of Hawaii to develop new machine learning methods for downscaling global climate model outputs to generate high-resolution (1 km) daily and sub-daily precipitation projections for the Hawaiian Islands. The project aims to improve modeling of future flood risk to support climate adaptation planning by government agencies, environmental nonprofits, and other stakeholders in Hawaii. Key deliverables include a research paper on the machine learning techniques, 20-year projections of future rainfall at 1 km resolution, a paper on diffusion models for estimating flood risk, and 1 km resolution flood risk maps for the Hawaiian Islands. The award period is from Aug 1, 2024 to Jul 31, 2027.

Project titlerefining ensemble climate model projections for the hawaiian islands:highresolution spatial downscaling of precipitation for flood risk withgenerative machine learningproject DATES8 1 2024 through 7 31 2027AWARD purposeas the impact of climate change grows affecting ecosystems economies and communities worldwide the need for effective mitigation and adaptation strategies becomes critical. A key element in these efforts is the ability to forecast the future impacts of climate change accurately. While global circulation models offer climate projections their coarse spatial resolution falls short in capturing regional characteristics accurately which are crucial for managing risks related to climate hazards like floods droughts and heatwaves. Dynamic downscaling which includes regional climate models offers more detailed projections but is computationally intensive thus limiting its usefulness for risk assessment. Alternatively statistical downscaling provides a computationally efficient way to achieve highresolutionclimate projections. Recent breakthroughs in generative artificial intelligence AI offer new methods for statistical downscaling. This project will leverage advancements in generative AI to generate highresolution 1KM precipitation maps at daily and subdaily scales for the hawaiian islands and use these to quantify the risk of flooding events. Improving these risk estimates will significantly improve the ability of local decisionmakers to plan for climate change impacts.activities to be performedthe project will develop a new machine learning approach to statistical downscaling that uses diffusion models to quantify uncertainty. The new statistical downscaling model will be trained and evaluated using twenty years of historical daily rainfall maps for hawaii then used to downscale global circulation model outputs to project future precipitation and flood risk.deliverables and expected outcomesthis project will produce four research PRODUCTS1 a paper describing the ML methods developed 2 projected future rainfall maps for all the hawaiian islands at 1KM resolution 3 a paper describing the ability of diffusion models to estimate flood risk 4 projected flood risk maps for all the hawaiian islands at 1KM resolution. Future projections will be made for 20 years.intended beneficiaryiesmultiple stakeholders have an interest in seasonal rainfall forecasting and the risk of floods. These include federal state and county resource managers agencies and environmental nonprofits as well as the public. Stakeholders include haleakala national park hawaii volcanoes national park USFWS USGS USFS hawaii commission on water resource management hawaii division of forestry and wildlife honolulu board of water supply maui department of water supply kauai department of water hawaii country department of water the nature conservancy leeward haleakala watershed restoration partnership three mountain alliance and the public. Subrecipient activitiesn a