Scholarlink

<p>Accurate prediction of monthly streamflow (<i>Q</i>) is essential for effective water resources and watershed management, including drought preparedness and mitigation, climate impact assessment, and watershed-scale hydrological budgeting. However, selecting the optimal model for reliable <i>Q</i> prediction in data-scare regions remains a significant challenge. This study compared four model types applied to Alingar watershed in eastern Afghanistan: (1) a conceptual model, IHACRES (Identification of unit Hydrographs And Component flows from Rainfall, Evapotranspiration and Streamflow), (2) a semi-distributed physically-based model, SWAT (Soil and Water Assessment Tool), (3) an advanced Machine Learning (ML) model, Dual Perturb and Combine Tree (DPC Tree), and (4) its ensemble with Bootstrap Aggregation (BA-DPC Tree). To enhance model performance, three data sources were compared: ground truth data (GTD), satellite data (SD), and bias-corrected satellite data (BCSD). Additionally, effective rainfall estimated from the IHACRES model, along with rainfall, temperature, and time-lagged streamflow, were incorporated with ML models to boost predictive accuracy and enhance understanding of the physics behind the rainfall-runoff process. The results revealed that BCSD inputs produced the most accurate model output, followed by SD and GTD. The ensemble BA-DPC Tree (NSE = 0.89) outperformed the other models, followed by SWAT (NSE = 0.81), DPC Tree (NSE = 0.80), and IHACRES (NSE = 0.79). The newly proposed BA-DPC Tree algorithm, when integrated with IHACRES models, demonstrates significant practical implications for predicting <i>Q</i>. ML models performed best when incorporating all relevant meteorological and hydrological data, as evidenced by the lowest RMSE values. Overall, the hybrid BA-DPC Tree model proved to be more reliable, with a narrower uncertainty range.</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

Enhancing Rainfall-Runoff Simulation in Data-Scarce Watersheds: Integration of Satellite Observations, Physically-Based Hydrological Modeling, and Machine Learning

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Khabat Khosravi,
  • Aitazaz Farooque,
  • Shasha Han,
  • David M. Hannah

摘要

Accurate prediction of monthly streamflow (Q) is essential for effective water resources and watershed management, including drought preparedness and mitigation, climate impact assessment, and watershed-scale hydrological budgeting. However, selecting the optimal model for reliable Q prediction in data-scare regions remains a significant challenge. This study compared four model types applied to Alingar watershed in eastern Afghanistan: (1) a conceptual model, IHACRES (Identification of unit Hydrographs And Component flows from Rainfall, Evapotranspiration and Streamflow), (2) a semi-distributed physically-based model, SWAT (Soil and Water Assessment Tool), (3) an advanced Machine Learning (ML) model, Dual Perturb and Combine Tree (DPC Tree), and (4) its ensemble with Bootstrap Aggregation (BA-DPC Tree). To enhance model performance, three data sources were compared: ground truth data (GTD), satellite data (SD), and bias-corrected satellite data (BCSD). Additionally, effective rainfall estimated from the IHACRES model, along with rainfall, temperature, and time-lagged streamflow, were incorporated with ML models to boost predictive accuracy and enhance understanding of the physics behind the rainfall-runoff process. The results revealed that BCSD inputs produced the most accurate model output, followed by SD and GTD. The ensemble BA-DPC Tree (NSE = 0.89) outperformed the other models, followed by SWAT (NSE = 0.81), DPC Tree (NSE = 0.80), and IHACRES (NSE = 0.79). The newly proposed BA-DPC Tree algorithm, when integrated with IHACRES models, demonstrates significant practical implications for predicting Q. ML models performed best when incorporating all relevant meteorological and hydrological data, as evidenced by the lowest RMSE values. Overall, the hybrid BA-DPC Tree model proved to be more reliable, with a narrower uncertainty range.