Sudoor Manthan- IIRS Science Portal
Digital Elevation Models (DEMs) are crucial for studying natural phenomena. Recent research at IIRS has focused on enhancing DEMs through advanced techniques like combining data from Remote Sensing, Satellite Photogrammetry, and SAR Interferometry (InSAR/IfSAR), as well as Spaceborne LiDAR. Notably, studies in the Ghaziabad and Dehradun regions have shown significant improvements in DEMs through fusion methods.
Analyzing fused DEMs from Sentinel-1A and 1B InSAR pairs, along with TanDEM-X 90m DEM, reveals better accuracy compared to individual input DEMs. Fusion methods using Artificial Neural Networks (ANN) show remarkable reductions in Root Mean Square Error (RMSE), reaching 3.46m (for the ANN model in Keras, Google Colab) and 4.34m (from the ANN model in MATLAB NN-Toolbox) in plain areas, with a 94.65% improvement over input DEMs.
The ANN-based fusion approach efficiently models relationships between input features and elevation values, especially evident in challenging terrains like Dehradun. Here, the fused DEMs achieve an RMSE of 10.95m, significantly lower than individual InSAR DEMs, with an 82.62% improvement. Successful implementation in such terrains requires meticulous data preparation and iterative model refinement.
In summary, these findings highlight the efficacy of ANN-based fusion methods in enhancing DEMs, particularly in complex terrains like hilly regions with dense forests and variable slopes.
References