Deep Learning Techniques for Flash Flood Management

This research uses deep learning methods, along with geospatial data from the USGS National Map and other public geospatial data sources, to develop forecasting tools capable of assessing water level rate of change in high risk flood areas. These tools build on existing models developed by the USGS, FEMA, and others, and are used to determine evacuation routing and detours to mitigate the potential for loss of life during flash floods. The project scope includes analysis of publically available flood data along a river basin as part of a pilot project in Missouri. These data are then used to determine the rate of rise based on projected rainfall totals. This rate of rise is used to model evacuation or detour planning modules that can be implemented to assure the safety of the community and highway personnel, as well as the safe and secure transport of goods along public roadways. These modules can be linked to existing real-time rainfall gauges and weather forecasts for improved accuracy and usability. The transportation safety or disaster planner can use these results to produce planning documents based on geospatial data and information to develop region-specific tools and methods.

The research team will disseminate research results through a variety of methods including MATC newsletters, seminars, and through the Missouri Local Technical Assistance Program (LTAP), headquartered on the S&T campus. The team will also promote technology transfer by presenting research results in transportation conferences (e.g., TRB) and publishing the results widely in high quality journal articles and conference proceedings. The research team will further disseminate results, as appropriate, through contributions to geospatial tools and products hosted by collaborative partners.

This proposed project will result in a series of tools and protocols based on deep learning methods. These materials will promote safety and economic viability of the surface roadways by providing real-time processes for flash flood control and alerts as well as driver rerouting schemas. These processes are data driven and use geospatial data and information to fully model areas categorized as high risk for flooding.

Download the Final Report