Integrating Analytical Hierarchy Process and Machine Learning for Enhancing Flood Hazard Mapping

Floods, intensified by climate change and anthropogenic activities, pose significant global threats to human life and infrastructure. Although several studies have demonstrated the efficacy of the multicriteria decision-making (MCDM) methods involving human expertise and machine learning (ML) methods utilizing machine intelligence for developing flood hazard maps, there is a need for an integrated approach that combines the advantages of both methods and addresses their respective shortcomings. This study proposes an integrated MCDM and ML framework using the analytical hierarchy process and the random forest algorithm (AHP-RF). The upper Assam region of India, highly susceptible to floods, was selected as a case study. Factors influencing flooding, such as topography, land cover, embankment breach, river confluence points, normalized difference moisture, vegetation and water index, and rainfall-runoff characteristics, were utilized for developing a high-resolution flood hazard map using the proposed AHP-RF approach. The predicted maps were compared with historical flood inventory utilizing the area under the curve (AUC) and accuracy metrics for validation. Results indicate that the AHP-RF approach provides more accurate representations of flood hazard zones, with an improved AUC score of 0.74 and accuracy of 0.792 compared to the standalone AHP (AUC score of 0.654 and accuracy of 0.595). The derived high-resolution flood hazard map highlights that the built-up and agricultural lands of the Dhemaji and Lakhimpur districts are highly susceptible, while Dibrugarh and Tinsukia are moderate and low susceptible, respectively. Overall, the proposed AHP-RF approach can offer a reliable flood hazard map, aiding authorities and stakeholders in flood mitigation.