ABSTRACT In this study, a sequence‐to‐sequence (Seq2Seq) deep learning model was developed to predict the chlorophyll‐ a concentration, which serves as a quantitative indicator of algal blooms, and its prediction performance was evaluated at eight different time steps from t + 1 to t + 28 days to analyze how varying prediction intervals affect model performance. The results demonstrated that the one‐step‐ahead prediction model performed best, achieving a Nash–Sutcliffe efficiency (NSE) of 0.908, whereas performance decreased with increasing prediction time step, with the NSE declining to 0.255 for 28‐day‐ahead predictions. This study also applied Shapley additive explanations analysis, a representative explainable artificial intelligence (XAI) method for quantitatively interpreting machine learning results, to further investigate the effects of environmental variables on model performance across these prediction intervals. The analysis revealed that flow rate tended to exhibit greater importance in short‐term predictions, whereas sunshine duration became more influential in longer term predictions. Overall, this study quantitatively interprets machine learning model results using XAI, contributing to the enhanced practical applicability of machine learning models in the field.
Kim et al. (Mon,) studied this question.