A-RAFT
Accessible Real-time AI-based Flood Tracking.
How can hydrology, precipitation, elevation, infrastructure, and other data be combined into a single accessible platform that uses ML/AI to help local emergency managers and planners better predict floods and mitigate flood impacts, especially for vulnerable residents?
The team will develop an AI-based emulator for real-time and high-resolution prediction of inundation. For a flood event, the emulator will provide maximum flood depths at 1 m resolution. Powered by AI, the emulator can capture both precipitation and storm-surge caused floods and will reduce the running time from weeks to seconds or hours, while maintaining the accuracy of high-spatiotemporal hydrodynamic modeling or remotely sensed products.
A-RAFT will make use of a large language model (LLM) to query geospatial datasets on flood depth and other related information on flood vulnerability. Their approach involves adapting pre-trained language models, prompts, and fine-tuning techniques from and on their AI-based flood model.
A Convergence Research Project
The A-RAFT team includes a range of expertise in flooding and familiarity with national-scale in-situ and remote sensing hydrological observations. Co-PIs from Florida International University and Rutgers University have extensive experience in hydraulic/hydrological flood modeling for two pilot cities, Miami, FL and Newark, NJ, and bring valuable knowledge of local areas and connections to local stakeholders. Co-PIs and researchers at the Centre of Machine Learning for Social Good of the University of Auckland bring scientific expertise in AI, especially LLM, as well as connections to a broader community of AI scientists that will help guide their work. The team also includes social scientists, public policy, and technology professionals with experience in user centered design, co-creation, and building partnerships with community organizations and private companies.
How has the CORE Institute helped the team?
"The CORE Institute has been a wonderful opportunity to connect with a talented and diverse set of fellows that share a passion for research that helps the many challenges associated with our changing climate. The fellowship bootcamp and incubator has helped our team ask relevant questions, refine our ideas for convergence research, and think about what it truly means to design research and technology that can be implemented and used effectively to address concrete challenges. Connecting with mentors who have engaged in convergence research projects has been very valuable in developing our own proposal and charting a course for the future."