Simulation of DEM based on ICESat-2 data using openly accessible topographic datasets

Digital Elevation Model (DEM) is a 3-dimensional digital representation of the terrain or the Earth’s surface. It is the ideal and most widely used method for determining topography with (i.e. Digital Surface Model) or without the objects (i.e. Digital Terrain Model). DEMs are generated from various techniques such as traditional Surveying, Photogrammetry, InSAR, LiDAR, Clinometry and Radargrammetry. It has been observed that mostly LiDAR-generated DEMs provide the best accuracy except for the VHR datasets acquired from UAVs having spatial resolution of few centimeters. The unavailability of LiDAR data in most of the region restricts global researchers from high-resolution and accurate DEMs. The recent launch of ICESat-2 with a 13m beam footprint and 0.7m pulse interval, promises elevations at high orbital precision. Its accuracy is of the order of few centimeters in complex topography, because of this ICESat-2 proves to be a good source to generate high-accuracy DEMs. ICESat-2 provides discrete photon data with elevations of points on the Earth’s surface. Traditional interpolation techniques tend to over-smooth the estimated space and still are unable to justify the complicated continuity in the topographical data. Machine learning algorithms are widely being used to extract patterns and spatial extent in geographic data. Machine learning regression algorithms are implemented in this study to estimate a DEM from ICESat-2 LiDAR point data using CartoDEM V3 R1. This study was conducted over a hilly terrain of Dehradun region in the foothills of Himalayas in India. The robustness of these algorithms has been tested for a plain region of Ghaziabad, Uttar Pradesh, India in an earlier study. Various regression-based machine-learning techniques were compared to interpolate DEM from ICESat-2 data. The RMSE of the interpolated DEM resulted from the Gradient Boosting Regressor, Random Forest Regressor, Decision Tree Regressor, and Multi-Layer Perceptron (MLP) Regressor was 7.13m, 7.01m, 7.15m, and 3.76m, respectively when evaluated against the TANDEM-X DEM of the same region. The MLP Regressor is found to perform the best among the four algorithms tested. The accuracy of the simulated ICESat-2 DEM using MLP Regressor was assessed using the DGPS points collected over the area and the RMSE was of the order of 6.58m.