Answers are grounded in this paper and cite their sources.
Floods are complex hydrological phenomena influenced by precipitation patterns and basin morphology. Although optical remote sensing is widely used for flood mapping, its effectiveness is often limited by cloud cover, which leads to data gaps and misclassifications. This study presents a novel method for detecting flooded areas in cloud-contaminated optical imagery by integrating hydrological and morphometric attributes derived from Digital Elevation Models (DEMs). The approach includes a customized cloud and shadow mask generated using the Rpart classifier, designed to reduce false positives while preserving surface water signals. The method employs a two-step Random Forest (RF) classification process. The first step (RF1) integrates optical imagery with DEM attributes to identify visible surface water. The second step (RF2) uses only DEM-derived features to replicate RF1 and extend flood mapping beneath cloud-covered regions. The method was validated using synthetic cloud data, achieving an Out-of-Bag Global Accuracy (GA) of 94% and a Jaccard Index of 0.94 for the RF2 replication task. In independent validation, the model showed strong performance: for randomly sampled test data, GA reached 94.06% with a Matthews Correlation Coefficient (MCC) of 72%; for stratified sampling, GA reached 88% with an MCC of 78%. These results highlight the model's robustness across diverse conditions. The proposed cloud-aware flood detection method provides a fast and accurate tool for post-disaster mapping, supporting emergency response efforts and enhancing flood risk management by mitigating the limitations of optical-only approaches.
No takes yet. Share an insight, caveat, or question.
Maia et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: