An Empirical Approach to Monitor the Flood-Prone Regions of North India Using Sentinel-1 Images

Floods in India is among the perilous natural disasters with a high impact on its economic sectors. One of the critical factors to handle such hazardous events is monitoring the affected areas and changes in flood patterns. Flood management is a very complex issue, largely owing to the growing population and investments in flood-affected regions. Satellite images especially Synthetic Aperture Radar (SAR) images are very useful and effective because SAR images are acquired day and night in all types of weather conditions. This research analyzes a combination of machine learning algorithms implemented on Sentinel-1A (SAR) data using supervised classification techniques to monitor the flooded areas in the North Indian region. Random Forest (RF) and the K-nearest neighbour (KNN) classification is applied to classify the different land covers such as water bodies, land, vegetation, and bare soil land covers. The outcomes of the presented work depict that the SAR data provides efficient information that helps in monitoring the flooded extents and the analysis shows that Sentinel-1 images are quite effective to detect changes in flood patterns in urban, vegetation, and regular water areas of the selected regions. The distribution of flooded areas was 16.6% and 16.8% in the respective region which is consistent with the resultant images of the proposed approach using RF and KNN classifiers. The obtained results indicate that both classifiers used in the work generate higher classification accuracy. These classifiers define the potential of multi-polarimetric SAR data in the classification of flood-affected areas. For a thorough evaluation and comparison, the RF and KNN are utilized as benchmarked classifiers. The classification accuracies based on the investigated results from the three SAR images can be improved by incorporating spatial and polarimetric features. In the future, the deep-learning classification techniques using ensemble strategies are expected to achieve an increased accuracy level with an overall classification strategy of urban and vegetation mapping.