Scholarlink

<p>Reservoir sedimentation represents a major threat to water resource sustainability in semi-arid regions such as northern Morocco. This study introduces a hybrid modeling framework that integrates the Modified Universal Soil Loss Equation (MUSLE) with a Long Short-Term Memory (LSTM) neural network to estimate daily sediment yield (SY) in the Tleta watershed, located upstream of the Ibn Battouta Dam. The model was developed using a 12-year dataset (2007–2018) combining remote sensing information—including daily precipitation and NDVI derived from Google Earth Engine—with in situ hydrological observations. MUSLE-derived SY estimates, calibrated against triennial bathymetric surveys, served as reference targets for LSTM training. The hybrid approach achieved R² = 0.93 during training and R² = 0.87 during testing, demonstrating strong capacity to reproduce seasonal variations and storm-generated sediment peaks. Unlike previous SWAT–ANN or standalone ML approaches, the proposed MUSLE–LSTM framework leverages GEE-derived variables to enable sediment monitoring in data-scarce environments with minimal field instrumentation. It is important to acknowledge that, in the absence of direct suspended-sediment measurements, the reported accuracy metrics reflect the model’s ability to reproduce MUSLE outputs rather than independently observed sediment fluxes—a limitation inherent to data-scarce contexts that is explicitly discussed herein. The framework offers a promising and cost-efficient tool to support sediment management and mitigate reservoir siltation in northern Morocco.</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

Hybrid MUSLE–LSTM Modeling of Sediment Yield Using Google Earth Engine-Derived Precipitation and NDVI in the Tleta Watershed, Morocco

  • 期刊论文

  • 发表日期：

    2026年7月25日

查看全文
  • Oumayma Bassairate,
  • Mohamed Chikhaoui,
  • Mustapha Naimi

摘要

Reservoir sedimentation represents a major threat to water resource sustainability in semi-arid regions such as northern Morocco. This study introduces a hybrid modeling framework that integrates the Modified Universal Soil Loss Equation (MUSLE) with a Long Short-Term Memory (LSTM) neural network to estimate daily sediment yield (SY) in the Tleta watershed, located upstream of the Ibn Battouta Dam. The model was developed using a 12-year dataset (2007–2018) combining remote sensing information—including daily precipitation and NDVI derived from Google Earth Engine—with in situ hydrological observations. MUSLE-derived SY estimates, calibrated against triennial bathymetric surveys, served as reference targets for LSTM training. The hybrid approach achieved R² = 0.93 during training and R² = 0.87 during testing, demonstrating strong capacity to reproduce seasonal variations and storm-generated sediment peaks. Unlike previous SWAT–ANN or standalone ML approaches, the proposed MUSLE–LSTM framework leverages GEE-derived variables to enable sediment monitoring in data-scarce environments with minimal field instrumentation. It is important to acknowledge that, in the absence of direct suspended-sediment measurements, the reported accuracy metrics reflect the model’s ability to reproduce MUSLE outputs rather than independently observed sediment fluxes—a limitation inherent to data-scarce contexts that is explicitly discussed herein. The framework offers a promising and cost-efficient tool to support sediment management and mitigate reservoir siltation in northern Morocco.