Scholarlink

<p>This study investigates drainage density (D<sub>d</sub>) characteristics and predictive modeling across climatic zones in Kurdistan Province, Iran, a tectonically active mountainous region in the Zagros fold belt exhibiting a pronounced west-to-east climatic gradient from humid to semi-arid conditions. D<sub>d</sub> values averaged 1.47 Km/Km² province-wide, with highest means in Mediterranean (1.71 Km/Km²) and semi-arid (1.65 Km/Km²) zones, contrary to classic arid-maximum paradigms, attributed to seasonal rainfall intensity, reduced vegetation cover, and enhanced erosional efficiency. Humid zones displayed the lowest D<sub>d</sub> (1.12 Km/Km²) due to greater infiltration promoted by dense vegetation. Six machine learning algorithms—Multiple Linear Regression (MLR), Random Forest (RF), Gradient Boosting (GB), Multilayer Perceptron (MLP), Decision Tree (DT), and Support Vector Machine (SVM)—were employed to predict D<sub>d</sub>, with models trained on 80% and tested on 20% of basin-scale data. Performance was evaluated using MSE, RMSE, MAE, R², MAPE, and MedAE. In training, ensemble methods (particularly GB) achieved near-perfect fits (R² &gt; 0.99), but severe overfitting emerged in testing, yielding negative R² in many cases. MLR demonstrated superior generalization (province-wide testing R² = 0.851), outperforming complex models. Climate stratification offered limited benefits, improving performance only in the homogeneous semi-humid zone, while degrading it in variable drier zones due to reduced sample sizes. Province-wide modeling, especially via parsimonious MLR, proved more robust. These findings highlight the risks of overfitting in data-limited hydrological and geomorphological applications and recommend simpler models for reliable D<sub>d</sub> prediction in climatically diverse regions, with implications for hydrological management and erosion assessment.</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

Machine learning-based modeling of watershed drainage density in different climates

  • 期刊论文

  • 发表日期：

    2026年5月15日

查看全文
  • Kamran Chapi,
  • Mahdi Sarbazi,
  • Zaniar Fatehi,
  • Shamim Faizi

摘要

This study investigates drainage density (Dd) characteristics and predictive modeling across climatic zones in Kurdistan Province, Iran, a tectonically active mountainous region in the Zagros fold belt exhibiting a pronounced west-to-east climatic gradient from humid to semi-arid conditions. Dd values averaged 1.47 Km/Km² province-wide, with highest means in Mediterranean (1.71 Km/Km²) and semi-arid (1.65 Km/Km²) zones, contrary to classic arid-maximum paradigms, attributed to seasonal rainfall intensity, reduced vegetation cover, and enhanced erosional efficiency. Humid zones displayed the lowest Dd (1.12 Km/Km²) due to greater infiltration promoted by dense vegetation. Six machine learning algorithms—Multiple Linear Regression (MLR), Random Forest (RF), Gradient Boosting (GB), Multilayer Perceptron (MLP), Decision Tree (DT), and Support Vector Machine (SVM)—were employed to predict Dd, with models trained on 80% and tested on 20% of basin-scale data. Performance was evaluated using MSE, RMSE, MAE, R², MAPE, and MedAE. In training, ensemble methods (particularly GB) achieved near-perfect fits (R² > 0.99), but severe overfitting emerged in testing, yielding negative R² in many cases. MLR demonstrated superior generalization (province-wide testing R² = 0.851), outperforming complex models. Climate stratification offered limited benefits, improving performance only in the homogeneous semi-humid zone, while degrading it in variable drier zones due to reduced sample sizes. Province-wide modeling, especially via parsimonious MLR, proved more robust. These findings highlight the risks of overfitting in data-limited hydrological and geomorphological applications and recommend simpler models for reliable Dd prediction in climatically diverse regions, with implications for hydrological management and erosion assessment.