Scholarlink

<p>Flooding is a devastating natural hazard; therefore, creating a highly accurate flood susceptibility map is a crucial tool for flood control and management. The goal of the present study is to create a flood potential map for the Borujerd-Dorud basin of Lorestan province, to evaluate the performance of the Deep Learning, CatBoost, XGB, RF, KNN, SVM models against one another, and to choose the model with the greatest accuracy employing machine learning methods in addition to the significant and influential elements on it. The first step in this study was to identify the relevant information layers on floods; next, the maps were specified as model inputs. Prepared in the Arc GIS software environment, the input layers—slope, aspect, distance from rivers, geology, precipitation, soil hydrological group, infiltration, and LU/LC. The models were trained then modeled for evaluation using 270 flood-prone locations set by the Regional Water Company of Lorestan. In the present study, spatial group k-fold cross-validation (5 folds)&#xa0;was employed&#xa0;to account for spatial autocorrelation. Geographical regions (1 × 1&#xa0;km grid cells) were used as groups to ensure that the training and test sets were spatially independent. The findings indicated that while all models are good at generating flood potential maps, the Random Forest model outperforms others with AUC = 0.71. To put this in simple terms, an AUC (Area Under the Curve) measures how well the model distinguishes between flood and non-flood locations. A score of 0.71 means the model has a 71% ability to correctly rank a random flood location as more susceptible than a random non-flood location. This level of accuracy provides planners with a reliable and actionable map to prioritize areas for flood mitigation efforts. Compared to other models, the KNN model is less accurate in generating flood susceptibility maps. All models showed that distance from rivers was a significant influence on flood vulnerability. The findings of the current study could be quite important for planning, management, control, and minimization of the harmful consequences of 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

Preparation of flood potential maps using machine learning and comparison of their performance

  • 期刊论文

  • 发表日期：

    2026年1月28日

查看全文
  • Ali Haghizadeh,
  • Tayebeh Sepahvand,
  • Leila Ghasemi,
  • Babak Shahinejad

摘要

Flooding is a devastating natural hazard; therefore, creating a highly accurate flood susceptibility map is a crucial tool for flood control and management. The goal of the present study is to create a flood potential map for the Borujerd-Dorud basin of Lorestan province, to evaluate the performance of the Deep Learning, CatBoost, XGB, RF, KNN, SVM models against one another, and to choose the model with the greatest accuracy employing machine learning methods in addition to the significant and influential elements on it. The first step in this study was to identify the relevant information layers on floods; next, the maps were specified as model inputs. Prepared in the Arc GIS software environment, the input layers—slope, aspect, distance from rivers, geology, precipitation, soil hydrological group, infiltration, and LU/LC. The models were trained then modeled for evaluation using 270 flood-prone locations set by the Regional Water Company of Lorestan. In the present study, spatial group k-fold cross-validation (5 folds) was employed to account for spatial autocorrelation. Geographical regions (1 × 1 km grid cells) were used as groups to ensure that the training and test sets were spatially independent. The findings indicated that while all models are good at generating flood potential maps, the Random Forest model outperforms others with AUC = 0.71. To put this in simple terms, an AUC (Area Under the Curve) measures how well the model distinguishes between flood and non-flood locations. A score of 0.71 means the model has a 71% ability to correctly rank a random flood location as more susceptible than a random non-flood location. This level of accuracy provides planners with a reliable and actionable map to prioritize areas for flood mitigation efforts. Compared to other models, the KNN model is less accurate in generating flood susceptibility maps. All models showed that distance from rivers was a significant influence on flood vulnerability. The findings of the current study could be quite important for planning, management, control, and minimization of the harmful consequences of floods.