Toward Rapid Flood Mapping Using Modeled Inundation Libraries

Abstract

New methods are needed for mapping floods in near real-time that leverage the increasing availability of remotely sensed data during flood disasters, availability of improved elevation data, and improvements in web-based mapping technology. There are important, ongoing improvements in elevation data production and availability that support new methods of flood disaster mapping. Concurrently, there is a rapid increase in the temporal frequency of high resolution remote sensing data that is being acquired that can also support novel application development. This study focuses on the use of prebuilt, modeled inundation libraries capable of using traditional and novel inputs as proxies for water surface elevation to produce near real-time estimates of flood inundation. Thus, the research explores potential synergies between inundation libraries and ancillary datasets with the goal of expediting the timeline for information extraction from remotely sensed data and the improving flood inundation map accuracy. It also profiles the computational cost of the modeling algorithm used. The study presents strategies for production of wide area, modeled flood inundation libraries. Gage-based interpolation methods using the FLDPLN model showed little difference in flood extent estimation accuracy between horizontal and vertical interpolation methods for FLDPLN model depth-to-flood (DTF) values. Conditioning of DTF profiles using HEC-RAS modeled water surface elevations (WSE) showed sensitivity to reference flood levels, while conditioning with two HEC-RAS model WSE profiles showed the best results. Simulation of imagery-derived flood boundary points as inputs to flood extent estimation using interpolated DTF profiles showed very good results with a very limited number of input points. The results showed improved, asymptotic behavior when correspondence was measured with an increasing number of input points when compared to reference floods. Larger magnitude floods showed better correspondence relative to moderate magnitude floods. Baseline computational performance measures for inundation library generation with the FLDPLN models showed that longer stream segments show better overall computational efficiency. Some landscape factors can influence overall computational runtime, including proximity to reservoirs and lakes, wide floodplains, and complex tributary geometries.