An Innovative DEM Improvement Technique for Highly Dense Urban Cities

Abstract

This paper presents an innovative approach to derive an improved Digital Elevation Model (DEM) using multispectral imagery and Artificial Neural Network (ANN). The DEM is crucial in land and water management which reflects the actual topographic characteristic on earth surface. However, a high accuracy DEM is very difficult to acquire because it is often very costly and is treated as confidential.

DEM from Shuttle Radar Topography Mission (SRTM) has been improved using multispectral imagery of Sentinel 2 and the ANN with its strength of pattern recognition in big data processing. SRTM is widely used in the area where the high accuracy DEM is not available as it is easily accessible to the public with no cost. However, its accuracy is limited due to its coarse resolution (≈30 m) and sensor limitations. Sentinel 2 provides the 13 spectral band spans from the visible and the near infrared to the short wave infrared at different resolutions ranging from 10 to 60 m. Sentinel 2 produces different reflectance values in different land-uses. These two remote sensing data are used in ANN as input data. The ANN is trained with reference DEM which has a high accuracy level and different weights are calculated to reduce the error between the elevation of SRTM and reference DEM.

The trained ANN is applied to a different place to evaluate the performance. The improved SRTM presents clearer images with higher resolution than the original SRTM with 6 to 26% lower Root Mean Square Error (RMSE). The paper should be of interest to readers in the areas of remote sensing, artificial intelligence and land/water management, especially for the policymakers who require land surface simulation with higher accuracy of topography.