<b>MACHINE LEARNING-BASED FORECASTING OF WATER QUALITY DYNAMICS IN THE ARYS RIVER AND SHARDARA RESERVOIR: A COMPARATIVE TEMPORAL ANALYSIS</b>

Water resource contamination remains one of the most pressing environmental challenges in Central Asia. This study analyzes monthly water quality data (n = 108 months, across 8 parameters) from the Arys River and Shardara Reservoir in Turkestan Oblast, Kazakhstan, covering the period 2017–2025. Applying the six-class national classification system established by Kazakhstan Order No. 70 (2024), an ensemble machine learning model comprising XGBoost, Random Forest (RF), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Prophet was developed using a soft voting strategy. The model achieved an overall accuracy of 94.1% for the Arys River and 91.2% for the Shardara Reservoir, outperforming all individual components. Forecasts for 2026–2028 project Class 2–3 conditions for the Arys River and Class 2 (good quality) for the Shardara Reservoir. SHAP (SHapley Additive exPlanations) analysis identified the key monthly-level physicochemical determinants: magnesium (Mg²⁺) and phenols for the Arys River, and sulfates (SO₄²⁻) for the Shardara Reservoir. The results provide a scientific and practical foundation for improving regional water resource management and early-warning systems in the Syr Darya basin.

Hali tarjima qilinmagan