Scholarlink

<p>Climate change is becoming a major concern as it affects hydrological processes and causes glacier melt. These changes have a significant impact on runoff patterns in glacier-fed basins. Accurate runoff prediction is essential for effective planning and sustainable water management. The study evaluates and compares the performance of six models for runoff prediction, including five machine learning (ML) models Artificial Neural Network (ANN), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Long Short-Term Memory (LSTM), and Extreme Gradient Boosting (XGBoost) and the Soil and Water Assessment Tool (SWAT) model in the Hunza River Basin (HRB). These models were selected to represent classical, deep learning, and ensemble approaches. This study fills a gap by applying climatic factors such as precipitation, temperature, relative humidity, wind speed, and snow cover as ML inputs for runoff prediction from 2007 to 2022. It employs multiple modeling approaches, utilizing key climatic parameters to enhance runoff prediction in glacier-fed basins. This integration is crucial for an accurate assessment of temporal changes and for predicting runoff. The performance metrics include the Coefficient of Determination (R²), Nash Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Relative Squared Error (RSE), and Percent Bias (PBIAS). XGBoost achieved an R² of 0.889 and an RMSE of 54.56, indicating high predictive accuracy with low error bias. Following XGBoost, the LSTM, ANN, SVR, and KNN models produced satisfactory results, with R² values of 0.86, 0.858, 0.83, and 0.84, respectively. In comparison, the SWAT model showed the lowest predictive performance, with an R² of 0.776 and a high RMSE of 77.52. The years 2020, 2021, and 2022 were identified as extreme runoff years, highlighting a limitation of the models, which are less accurate at predicting high flow variations. Statistical tests, performance metrics, and residual analysis confirmed that XGBoost performed best, while the SWAT model showed the lowest predictive accuracy compared with the other models. In summary, the ML models proved highly effective in predicting runoff. These results are crucial for optimizing water resource management, improving flood control, and supporting strategic planning in regions affected by climate change.</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

Runoff prediction under climatic variability using SWAT and machine learning models: a case study of the Hunza River basin

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • Muhammad Ghawas Kareem,
  • Deshan Tang,
  • Muhammad Farhan,
  • Anis Ur Rehman Khalil,
  • Hafiz Ahmad Hammad Abid

摘要

Climate change is becoming a major concern as it affects hydrological processes and causes glacier melt. These changes have a significant impact on runoff patterns in glacier-fed basins. Accurate runoff prediction is essential for effective planning and sustainable water management. The study evaluates and compares the performance of six models for runoff prediction, including five machine learning (ML) models Artificial Neural Network (ANN), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Long Short-Term Memory (LSTM), and Extreme Gradient Boosting (XGBoost) and the Soil and Water Assessment Tool (SWAT) model in the Hunza River Basin (HRB). These models were selected to represent classical, deep learning, and ensemble approaches. This study fills a gap by applying climatic factors such as precipitation, temperature, relative humidity, wind speed, and snow cover as ML inputs for runoff prediction from 2007 to 2022. It employs multiple modeling approaches, utilizing key climatic parameters to enhance runoff prediction in glacier-fed basins. This integration is crucial for an accurate assessment of temporal changes and for predicting runoff. The performance metrics include the Coefficient of Determination (R²), Nash Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Relative Squared Error (RSE), and Percent Bias (PBIAS). XGBoost achieved an R² of 0.889 and an RMSE of 54.56, indicating high predictive accuracy with low error bias. Following XGBoost, the LSTM, ANN, SVR, and KNN models produced satisfactory results, with R² values of 0.86, 0.858, 0.83, and 0.84, respectively. In comparison, the SWAT model showed the lowest predictive performance, with an R² of 0.776 and a high RMSE of 77.52. The years 2020, 2021, and 2022 were identified as extreme runoff years, highlighting a limitation of the models, which are less accurate at predicting high flow variations. Statistical tests, performance metrics, and residual analysis confirmed that XGBoost performed best, while the SWAT model showed the lowest predictive accuracy compared with the other models. In summary, the ML models proved highly effective in predicting runoff. These results are crucial for optimizing water resource management, improving flood control, and supporting strategic planning in regions affected by climate change.