Answers are grounded in this paper and cite their sources.
Accurate prediction of water turbidity is important for effective water quality management. However, it is a challenge task due to dynamic hydrological patterns and seasonal variability. The present study aimed to evaluate eight machine learning-based models for predicting average seasonal water turbidity . The MLPANN model presented the most accurate results among the employed techniques (RMSE: 3.047 FNU, R2: 0.932), while the performance of the models fluctuated among the seasons: GMDH, XGBoost, GBR, and RF presented the best outcomes for spring (R2: 0.941), summer (R2: 0.956), autumn (R2: 0.942), and winter (R2: 0.967), respectively. Diebold-Mariano tests confirmed significant performance differences between the models (e.g. CART vs. RF, p = 0.019), with river discharge identified as the most influential input variable (SHAP value: ~8, correlation: 0.70). Time-series analysis highlighted MLPANN’s robustness in tracking trends but exposed underestimation of extreme turbidity events (>80 FNU) and overestimation at low values (<5 FNU), with wide uncertainty intervals. Wavelet coherence further confirmed the multiscale influence of river discharge, with strong annual coherence (seasonal cycles) and intermittent short-term effects.
No takes yet. Share an insight, caveat, or question.
Shiri et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: