Floods cause more damage than any other disaster. The economic impacts of floods push people into poverty and cause setbacks to development as government budgets are stretched and people without…

The National Aeronautics and Space Administration (NASA) awarded a $377,733 grant to the University of Arizona, a public research university and Hispanic Serving Institution, to develop algorithms that improve satellite-based flood inundation mapping and quantify flood risk. The goal is to support flood mitigation efforts by leveraging satellite observations to estimate extreme flood risk in urban and agricultural areas, complementing traditional modeling approaches. The research team will use deep learning, Bayesian statistics, and data from optical and radar satellites to map inundated areas, calculate flood return periods at watershed and household scales, and compare results to existing flood forecasting models in Bangladesh. This work aligns with NASA's focus on assessing and responding to extreme events and using Earth science research to inform decisions and societal benefits. The University holds other federal grants and contracts, including an Other Transaction Agreement with the Defense Advanced Research Projects Agency (DARPA) for research on quantum nanostructures.

Floods cause more damage than any other disaster. The economic impacts of floods push people into poverty and cause setbacks to development as government budgets are stretched and people without financial protection are forced to sell assets. Floods are increasing in urban areas, where most of the world's population and economic activity are located. Mapping floods from satellites improves response, relief, preparedness, planning, and mitigation efforts to save money, lives, and property. This proposal will develop algorithms to improve inundation mapping with deep learning and quantify flood risk to support flood mitigation efforts. Most approaches to estimate flood risk do not incorporate satellite observations of inundation but instead rely on physically based hydraulic flood models or discharge measurements. These approaches have limitations. Discharge for extreme (infrequent, large impact) flood events, currently measured by satellite, models, or gauges, can be highly uncertain. Hydraulic flood models rely on high-resolution elevation and infrastructure data to accurately predict inundation. Flood infrastructure is rapidly changing in the anthropocene and can render model predictions inaccurate. A more direct approach to understand extreme floods and support financial instruments to mitigate their impact is to measure the spatial extent of inundation directly from space. This approach captures inundation wherever IT occurs on the landscape. Surface water detection algorithms are routinely used to map individual flood events, but with limited ability to capture fast moving urban floods. Flood detection algorithms are rarely used to continuously estimate inundation extent over time and space to estimate flood frequency and risk. This proposal leverages inundation observations to understand extreme flood risk in human-modified landscapes, specifically in urban and agricultural areas, contributing to two of NASA s key science goals, to improve the capability to assess and respond to extreme events and use earth system science research to inform decisions and provide benefits to society. These objectives will be achieved by i) using deep learning approaches with high-resolution imagery to map inundation and damage, II) estimating return periods for fractional inundated area measured by satellite, generalizable to any watershed or region, III) estimating per-pixel return periods accounting for spatial dependence using satellite observations and topography with bayesian hierarchical models. In both urban and agricultural regions, this proposed work will use commercial and public sensors to i) map maximum inundated area for extreme flood events, II) quantify flood risk at watershed scales, and III) estimate of flood return periods at household scales (30M resolution). Optical and radar satellites, including modis, sentinel-1, planetscope, and the HLS (harmonized landsat sentinel-2 dataset (claverie ET AL. 2018)) will be used. Satellite-based flood maps and return period predictions for bangladesh will be compared to those produced by the bangladesh flood forecasting and warning centre (from a physics-based model). Given the growing length of the satellite record, time series of inundated area could be used for exeedence probability estimation. Improving flood risk estimates in cities and agricultural areas from watershed to pixel spatial scales could improve flood resilient land use and infrastructure planning and underpin innovative index-based insurance mechanisms. Applications of satellite-based inundation estimates to support insurance and climate risk financing will be discussed with collaborators in bangladesh. The methods developed in this proposal will open a new avenue for flood risk analysis and its mitigation based on satellite observation, to complement traditional discharge and flood model-based approaches, especially in human-modified landscapes where traditional approaches face substantial challenges.