Flood Inundation Mapping of the 2018 Kerala Floods: A ComparativeStudy of Traditional Remote Sensing, Machine Learning, and Deep Learning Methods.

The exponential increase in flood intensity that causes loss of life and economic and structuraldamage to their connected environment calls for strategic rescue and response solutions forrisk mitigation. This study focuses on flood mapping using satellite imagery combined withmachine learning (ML) and deep learning (DL) techniques. Remote sensing and GeographicInformation Systems (GIS) serve as vital tools in this process, enabling the effective utilizationof satellite data.While academics consistently contribute to novel flood mapping approaches, a research gapthat requires a discussion about the comparative performances of these ML and DLtechniques exists, which this paper aims to address. This comparison is crucial as it highlightsthe strengths and limitations of each method, contributing valuable insights to the literature onflood risk management. The study focuses on the Ernakulam District of Kerala, chosen due toits frequent flooding and the availability of diverse datasets.The methodology involves the use of satellite imagery for flood analysis, employing an arrayof techniques: a thresholding method recommended by the UN-SPIDER Office for OuterSpace Affairs, and statistical ML methods including Random Forest, Support VectorClassification (SVC), Real AdaBoost, alongside a deep learning semantic segmentationmethod, UNet. Modelled using JavaScript and Python languages, the models and thepackages are completely reusable. The dataset comprises two before and after floods satelliteimages: the thresholding method uses Sentinel-1 SAR images, and the ML and DL methoduses Sentinel-2 MSI Level 1C, a digital elevation model image from SRTM for featureengineering, processed to identify flood-affected areas. The data is normalized and cleanedto account for cloud and missing data before the analysis. Alongside, we sourced the labelledflood data from the Kerala State Disaster Management Authority (KSDMA) and filtered andrasterized it on QGIS.The results emphasize the varied effectiveness of these methods, with Random Forestoutperforming others with a 96.61% accuracy rate. At the same time, the UNet-Linear Modellags at 75% accuracy, indicating the significant impact of hyperparameter tuning and datasetsize on model performance. This comparative analysis not only delineates the strengths andweaknesses of traditional and advanced techniques but also sets a precedent for futurestudies to build upon an understanding of flood risk management and rapid responsestrategies.