Scholarlink

<p>Flooding is the most prevalent monsoon calamity in Kerala (a state in SW India), with the 2018 and 2019 events being the most catastrophic. Many tropical river basins in Kerala were severely battered by flooding during these two&#xa0;years. Thus, this study aims to create a&#xa0;flood risk map of the tropical Keecheri-Puzhakkal river basin in Kerala, which is&#xa0;usually&#xa0;flooded every monsoon, employing the Analytic Hierarchy Process (AHP), Fuzzy-AHP (F-AHP), Support Vector Machine (SVM), and SVM-Naïve Bayes (SVM-NB) stacking models to identify the model with better performance and to list the important conditioning factors (CFs). A total of nine CFs, such as slope, soil, stream density, aspect, land use/land cover (LULC), normalized difference water index (NDWI), stream power index (SPI), sediment transport index (STI), and topographic wetness index (TWI) have been selected for hazard modelling. The Area Under the Curve (AUC) values of the four hazard maps confirmed an acceptable performance (AUC ≥ 0.70) for the knowledge-driven AHP and F-AHP models, and an excellent performance (AUC ≥ 0.80) for the data-driven SVM and SVM-NB models. However, the SVM-NB model (AUC: 0.831) accomplished the highest performance, followed by the SVM model (AUC: 0.829), the F-AHP model (AUC: 0.769), and the AHP model (AUC: 0.768). The validation employing other metrics also supported this, substantiating our findings that data-driven models outperform the knowledge-driven models. Furthermore, the ensemble models—F-AHP and SVM-NB—exhibited slight enhancements over their respective standalone counterparts, AHP and SVM, underscoring the advantages of integrating models. Finally, exposure and vulnerability layers have been integrated with the hazard layers to produce risk maps. This study found that slope, LULC, SPI, TWI, and stream density are the top five important CFs. The outcomes of this modelling will help land use planners and policymakers identify suitable models for flood risk modelling in the future, so that the government will be equipped to deal with extreme events like the 2018 floods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative Flood Risk Mapping Using Knowledge-Driven, Data-Driven, and Ensemble Models in a Humid Tropical River Basin in India

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Chandini Padmanabha Panicker Chandra Senan,
  • Rajendran Shobha Ajin,
  • B. R. Anaswara Devi,
  • A. Rajaneesh,
  • Jitendra K. Nagar,
  • K. S. Sajinkumar

摘要

Flooding is the most prevalent monsoon calamity in Kerala (a state in SW India), with the 2018 and 2019 events being the most catastrophic. Many tropical river basins in Kerala were severely battered by flooding during these two years. Thus, this study aims to create a flood risk map of the tropical Keecheri-Puzhakkal river basin in Kerala, which is usually flooded every monsoon, employing the Analytic Hierarchy Process (AHP), Fuzzy-AHP (F-AHP), Support Vector Machine (SVM), and SVM-Naïve Bayes (SVM-NB) stacking models to identify the model with better performance and to list the important conditioning factors (CFs). A total of nine CFs, such as slope, soil, stream density, aspect, land use/land cover (LULC), normalized difference water index (NDWI), stream power index (SPI), sediment transport index (STI), and topographic wetness index (TWI) have been selected for hazard modelling. The Area Under the Curve (AUC) values of the four hazard maps confirmed an acceptable performance (AUC ≥ 0.70) for the knowledge-driven AHP and F-AHP models, and an excellent performance (AUC ≥ 0.80) for the data-driven SVM and SVM-NB models. However, the SVM-NB model (AUC: 0.831) accomplished the highest performance, followed by the SVM model (AUC: 0.829), the F-AHP model (AUC: 0.769), and the AHP model (AUC: 0.768). The validation employing other metrics also supported this, substantiating our findings that data-driven models outperform the knowledge-driven models. Furthermore, the ensemble models—F-AHP and SVM-NB—exhibited slight enhancements over their respective standalone counterparts, AHP and SVM, underscoring the advantages of integrating models. Finally, exposure and vulnerability layers have been integrated with the hazard layers to produce risk maps. This study found that slope, LULC, SPI, TWI, and stream density are the top five important CFs. The outcomes of this modelling will help land use planners and policymakers identify suitable models for flood risk modelling in the future, so that the government will be equipped to deal with extreme events like the 2018 floods.