Scholarlink

<p>Accurate lake level forecasting is crucial for sustainable water resource management amidst growing climate variability and anthropogenic impacts. Existing methods often struggle to balance accuracy, computational efficiency, and generalizability across diverse lake systems, with traditional models often requiring site-specific calibration and advanced machine learning techniques being computationally demanding and sensitive to time series structural components. This study proposes a novel framework integrating grid search-optimized Seasonal Autoregressive Integrated Moving Average (GS-SARIMA), Long Short-Term Memory (LSTM), and Extreme Gradient Boosting (XGB) models, optimized using the Augmented Weighted Mean Vector Optimizer (AWMVO). The optimization process is enhanced with composite cost functions and data-driven feature engineering. The framework’s novelty lies in combining optimization with statistical insights from lake-specific datasets, enhancing model robustness and adaptability. The models’ performances are evaluated by several metrics, e.g., the Kling-Gupta efficiency (KGE) and mean absolute percentage error (MAPE). The GS-SARIMA model achieved the highest test KGE (mean: 0.965; range: 0.946–0.989) and a consistently low MAPE (mean: 0.039%), excelling in capturing seasonal and nonseasonal patterns, but restricted by site-specific sensitivity. AWMVO-LSTM demonstrated superior generalizability across lakes, maintaining comparable accuracy with ability transfer, with a mean test KGE of 0.933 (range: 0.890–0.960) and mean MAPE of 0.046%, but it is the most computationally expensive model. XGB, while computationally efficient, showed the lowest average test KGE (0.838) and highest MAPE (mean: 0.073%, max: 0.176%), particularly underperforming in high variability systems, such as Lake Ontario. These results highlight essential trade-offs among computational efficiency, accuracy, and generalizability. This study emphasizes the value of optimizing predictive models with advanced techniques customized to hydrological and statistical characteristics. Future work should prioritize hybrid modeling and the integration of advanced feature engineering to address region-specific challenges, ensuring robust and transferable predictions for diverse hydrological systems.</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

Overcoming hydrological forecasting challenges through augmented adaptive deep algorithms: a case study of the great lakes across Canada and the U.S

  • 期刊论文

  • 发表日期：

    2025年10月14日

查看全文
  • Mohammad Zeynoddin,
  • Hossein Bonakdari,
  • Afshin Amiri,
  • Silvio José Gumiere,
  • Tadros Ghobrial

摘要

Accurate lake level forecasting is crucial for sustainable water resource management amidst growing climate variability and anthropogenic impacts. Existing methods often struggle to balance accuracy, computational efficiency, and generalizability across diverse lake systems, with traditional models often requiring site-specific calibration and advanced machine learning techniques being computationally demanding and sensitive to time series structural components. This study proposes a novel framework integrating grid search-optimized Seasonal Autoregressive Integrated Moving Average (GS-SARIMA), Long Short-Term Memory (LSTM), and Extreme Gradient Boosting (XGB) models, optimized using the Augmented Weighted Mean Vector Optimizer (AWMVO). The optimization process is enhanced with composite cost functions and data-driven feature engineering. The framework’s novelty lies in combining optimization with statistical insights from lake-specific datasets, enhancing model robustness and adaptability. The models’ performances are evaluated by several metrics, e.g., the Kling-Gupta efficiency (KGE) and mean absolute percentage error (MAPE). The GS-SARIMA model achieved the highest test KGE (mean: 0.965; range: 0.946–0.989) and a consistently low MAPE (mean: 0.039%), excelling in capturing seasonal and nonseasonal patterns, but restricted by site-specific sensitivity. AWMVO-LSTM demonstrated superior generalizability across lakes, maintaining comparable accuracy with ability transfer, with a mean test KGE of 0.933 (range: 0.890–0.960) and mean MAPE of 0.046%, but it is the most computationally expensive model. XGB, while computationally efficient, showed the lowest average test KGE (0.838) and highest MAPE (mean: 0.073%, max: 0.176%), particularly underperforming in high variability systems, such as Lake Ontario. These results highlight essential trade-offs among computational efficiency, accuracy, and generalizability. This study emphasizes the value of optimizing predictive models with advanced techniques customized to hydrological and statistical characteristics. Future work should prioritize hybrid modeling and the integration of advanced feature engineering to address region-specific challenges, ensuring robust and transferable predictions for diverse hydrological systems.