Bare Earth DEM Generation for Large Floodplains Using Image Classification in High-Resolution Single-Pass InSAR

Abstract

The increasing frequency and severity of flood events demand improved accuracy of hydrodynamic models to better mitigate the societal and economic consequences of these disasters. Ideally, these models derive elevation data from high-accuracy LiDAR, which often captures the subtle variations in elevation and microtopography—elements that are critical to high-resolution hydrodynamic modeling. Due largely to the cost of acquisitions, these data, however, are not widely available at high spatial resolutions for large-scale areas, i.e. floodplains and coastal regions, especially outside of the U.S. and Western Europe, where flood-prone communities already lack the infrastructure and resources to manage these disasters. High-resolution interferometric synthetic aperture radar (InSAR) systems may offer a viable and cost-effective alternative to LiDAR, with comparable spatial resolutions and vertical accuracies. Like any swath mapping sensor, systematic errors in calibration knowledge increase as the lever-arm increases when viewing further off vertical. InSAR-derived digital elevation models (DEMs) are often corrected using LiDAR or ground control points, although this limits the application of InSAR to those areas where these data exist. In this paper, we present a novel approach for using near-range InSAR data to correct inherent systematic bias that propagates across adjacent, overlapping swaths for the same airborne acquisition. This eliminates the need for LiDAR in generating InSAR DEMs, while maintaining a vertical error of approximately 0.26 m relative to LiDAR, at a spatial resolution of 30 m. Data was acquired using the NASA/JPL airborne, single-pass, Ka-band GLISTIN-A (Glacier and Ice Surface Topography Interferometer) InSAR over the Red River Basin (RRB), ND. The accuracy of the final DEM is evaluated using National Geodetic Survey (NGS) GPS and a high-resolution LiDAR DEM.