Scholarlink

<p>Floods are a major threat to life, infrastructure and livelihoods in Bangladesh, especially in low-lying and rapidly urbanising coastal areas. This study assesses the effectiveness of Random Forest (RF) and Maximum Entropy (MaxEnt) models in mapping flood susceptibility in the Fens district in southeastern Bangladesh, using 14 flood control factors, including topographic, hydroclimatic, soil and spectral variables, embedded in Google Earth. Flood and non-flood inventory points were randomly allocated to the training (80%) and validation (20) datasets. All predictor variables had a tolerance value above 0.20, which indicates a low multilinearity. The performance of the model was evaluated by using accuracy, precision, recall, F1 score, and metrics ROC-AUC. Both models had strong predictive power: MaxEnt had a superior discriminative capability (AUC = 0.98) and a higher recall, while RF had a higher overall accuracy (0.93), a precision (0.92), and a F1 score (0.95). The models differed in the distribution of sensitivity classes: RF predicted the highest suitability in medium (329.08&#xa0;km²) and low (324.35&#xa0;km²), while MaxEnt predicted the lowest suitability in medium (365.86&#xa0;km²) and low (337.0 km2), with a minor (17.11 km2) and very high suitability in the middle class. These findings demonstrate the value of machine learning-based flood susceptibility mapping in data-poor environments, providing critical information for disaster preparedness, spatial planning and early warning systems, but underline that areas close to the threshold should be interpreted with caution because of possible uncertainties in classifications.</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

Geospatial driven machine learning approach for flood susceptibility mapping in southeastern region of Bangladesh

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月2日

查看全文
  • Mohammad Ismail Hossain,
  • Swarnali Bhattacharjee,
  • Md Nahid Ferdous,
  • Mashiyat Raunaq Preetom,
  • Kimihiko Hyakumura,
  • Anjum Tasnuva,
  • Md Refath Hossan

摘要

Floods are a major threat to life, infrastructure and livelihoods in Bangladesh, especially in low-lying and rapidly urbanising coastal areas. This study assesses the effectiveness of Random Forest (RF) and Maximum Entropy (MaxEnt) models in mapping flood susceptibility in the Fens district in southeastern Bangladesh, using 14 flood control factors, including topographic, hydroclimatic, soil and spectral variables, embedded in Google Earth. Flood and non-flood inventory points were randomly allocated to the training (80%) and validation (20) datasets. All predictor variables had a tolerance value above 0.20, which indicates a low multilinearity. The performance of the model was evaluated by using accuracy, precision, recall, F1 score, and metrics ROC-AUC. Both models had strong predictive power: MaxEnt had a superior discriminative capability (AUC = 0.98) and a higher recall, while RF had a higher overall accuracy (0.93), a precision (0.92), and a F1 score (0.95). The models differed in the distribution of sensitivity classes: RF predicted the highest suitability in medium (329.08 km²) and low (324.35 km²), while MaxEnt predicted the lowest suitability in medium (365.86 km²) and low (337.0 km2), with a minor (17.11 km2) and very high suitability in the middle class. These findings demonstrate the value of machine learning-based flood susceptibility mapping in data-poor environments, providing critical information for disaster preparedness, spatial planning and early warning systems, but underline that areas close to the threshold should be interpreted with caution because of possible uncertainties in classifications.