Application of Machine Learning Methods to improve vertical accuracy of
CARTOSAT DEM
Abstract
In recent decades, the application of Digital Elevation Models (DEMs)
has been widely used in various aspects such as land management and
flood planning since it reflects the actual topographic characteristic
on the Earth’s surface. However, obtaining a high-quality DEM is often
quite challenging because it is time-consuming, costly, and often
confidential. This study presents an innovative approach to derive an
improved vertical accuracy of CARTOSAT 10m DEM by blending it with
publicly available SRTM (Shuttle Radar Topography Mission) DEM using
machine learning methods such as Genetic Programming (GP) and Artificial
Neural Networks (ANN). SRTM-1 DEM and CARTOSAT DEM in India are applied
to GP and ANN to generate improved vertical accuracy high-quality DEM.
The results revealed that the proposed approach improves the vertical
accuracy by considering the reference as Ground control Points (GCPs)
elevation from Differential Global Positioning System (DGPS) survey
data. A significant improvement of 47 and 35% generated DEMs in RMSE
compared to the SRTM-1 and CARTOSAT, respectively.