This study presents the development of interpretable flash flood susceptibility mapping for the Yarlung Tsangpo River Basin using H2O Auto-ML techniques.
Flash floods have become increasingly relevant due to their rapid onset and devastating impacts. This research evaluates flash flood susceptibility using machine learning models, which are integrated with interpretable methods to provide insights for policymakers.
The Yarlung Tsangpo River Basin, located on the Tibetan Plateau, is frequently prone to flash floods due to its diverse terrain and the effects of heavy rainfall. The study employed advanced machine learning techniques, particularly the automated ML platform H2O, to analyze and predict susceptibility to flash floods. This method allows for greater accuracy and interpretability, enabling decision-makers to understand the underlying factors influencing flood risk.
Using historical data spanning from 1980 to 2019, the research categorized the basin's susceptibility based on various topographic and environmental factors. The findings revealed startling effects with approximately 74.9% of recorded flash floods occurring within areas identified as having moderate to very high susceptibility. This key insight highlights the effectiveness of the methodology used.
The SHAP (Shapley Additive Explanations) approach offered insights not only on the susceptibility but also the significant contributions of topographic features such as elevation and terrain ruggedness to flash flood occurrences. This model's interpretability is integral, as it aids local authorities to prioritize areas for enhanced preventive measures.
This research advocates for continuous efforts to utilize data-driven methodologies for flood susceptibility mapping, emphasizing the amalgamation of machine learning and environmental analysis as pivotal for regions severely impacted by flash floods. By embedding such techniques within planning initiatives, resilience to natural disasters can be significantly bolstered.
Future research should focus on refining these predictive models and integrating them with changing climate variables to adapt to the region's dynamic environmental conditions.