Assessing the Accuracy of Digital Elevation Models for Darjeeling‐Sikkim Himalayas

Abstract

A high-quality Digital Elevation Model (DEM) is essential to understand the Earth's surface processes. The quality of DEM is a measure of how accurately each pixel represents the elevation of the Earth's surface. This present study has assessed the quality of three different types of DEMs by comparing the vertical and spatial (horizontal) accuracy with reference to the 22 Ground Control Points (GCPs) measured by high precision Differential Global Positioning System (DGPS) and stream networks, respectively. For this purpose, 10 m resolution DEM has been prepared from the Cartosat-1 stereoscopic satellite images by automatic image-matching techniques. The remaining two, 12 m resolution TanDEM-X and 12.5 m resolution ALOS PALSAR, are open-source data. The accuracy of these DEMs has been tested in a part of the Darjeeling-Sikkim Himalayas. The results show that vertical accuracy of Cartosat-1 DEM and TanDEM-X are almost similar in areas covered with open forest. However, TanDEM-X represents the elevation of terrain at a satisfactory level, except for areas covered with dense forest. Stream networks digitized from a 2.5 m resolution ortho-rectified Cartosat-1 satellite image have been compared with the stream networks automatically extracted from Cartosat-1 DEM, TanDEM-X, and ALOS PALSAR for spatial accuracy assessment. Stream networks extracted from Cartosat-1 DEM and TanDEM-X closely match the reference network in the north-south oriented valleys. In contrast, a higher mismatch was noticed between the reference and ALOS PALSAR extracted stream networks in northwest-southeast oriented valleys.