Scholarlink

<p>Time series forecasting models are essential components of early warning systems, which play a crucial role in supporting decision-makers in mitigating the impacts of natural hazards, particularly in water resource management and agricultural planning. The reliability of these systems depends on the accuracy of future predictions. This study addressed the effect of input window size on the predictive performance of four models—Seasonal Autoregressive Integrated Moving Average with exogenous variables (SARIMAX), Multi-Layer Perceptron (MLP), Sequence-to-Sequence Long Short-Term Memory (Seq2Seq-LSTM), and Bidirectional Long Short-Term Memory (BiLSTM). In addition, the post-forecast evaluation was conducted to assess the accuracy of forecast results compared to observed data in 2024. The results showed that varying input window sizes led to a significant improvement in forecasting accuracy, and each model performed optimally with specific window sizes. All models using a 12-month window accurately forecasted the study area conditions for the first six months of 2024. However, the inherent uncertainty in predicting climate variables makes it challenging to determine an optimal window size, especially for long-term applications. The MLP model outperformed with window sizes of 3 and 6&#xa0;months, the Seq2Seq model excelled with 12, 15, and 18-month windows, and the BiLSTM model provided stable performance using 12 and 15-month window sizes. The window sizes of 30 and 36&#xa0;months increased the risk of overfitting in the models, which requires retuning. The post-forecast evaluation results highlighted the importance of model maintenance through regular updates and retraining to ensure the models adapt to new shifts in data patterns.</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

The impact of window size on the performance and accuracy of time series forecasting models for meteorological drought prediction

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Leila Fazeldehkordi,
  • Jie-Lun Chiang

摘要

Time series forecasting models are essential components of early warning systems, which play a crucial role in supporting decision-makers in mitigating the impacts of natural hazards, particularly in water resource management and agricultural planning. The reliability of these systems depends on the accuracy of future predictions. This study addressed the effect of input window size on the predictive performance of four models—Seasonal Autoregressive Integrated Moving Average with exogenous variables (SARIMAX), Multi-Layer Perceptron (MLP), Sequence-to-Sequence Long Short-Term Memory (Seq2Seq-LSTM), and Bidirectional Long Short-Term Memory (BiLSTM). In addition, the post-forecast evaluation was conducted to assess the accuracy of forecast results compared to observed data in 2024. The results showed that varying input window sizes led to a significant improvement in forecasting accuracy, and each model performed optimally with specific window sizes. All models using a 12-month window accurately forecasted the study area conditions for the first six months of 2024. However, the inherent uncertainty in predicting climate variables makes it challenging to determine an optimal window size, especially for long-term applications. The MLP model outperformed with window sizes of 3 and 6 months, the Seq2Seq model excelled with 12, 15, and 18-month windows, and the BiLSTM model provided stable performance using 12 and 15-month window sizes. The window sizes of 30 and 36 months increased the risk of overfitting in the models, which requires retuning. The post-forecast evaluation results highlighted the importance of model maintenance through regular updates and retraining to ensure the models adapt to new shifts in data patterns.