Моделирование пространственно-временной динамики площади водоёма (на примере Каттакурганского водохранилища) на основе индексов NDWI, NDVI, EVI и ансамблевых методов обучения

The paper models the spatiotemporal dynamics of the Kattakurgan Reservoir area using satellite-based water and vegetation indices combined with ensemble machinelearning methods. Multi-year time series of NDWI, NDVI, and EVI are used together with meteorological variables (air temperature, humidity, precipitation) from the ERA5 and CHIRPS datasets on the Google Earth Engine platform for 2018–2023. The reservoir water-surface area, derived from satellite imagery, serves as the target variable for training and validating a Random Forest model. The study includes a train–test split, hyperparameter tuning, and performance evaluation using R2, RMSE, and MAE. The results indicate high accuracy: R2 = 0.962 for NDWI prediction, and values close to 1 for reservoir area, with RMSE about 5–7 km2 and MAE 3–5 km2. Forecasts of NDWI and reservoir area for 2024 are produced to assess expected changes in water extent and support water-management decisions and climate adaptation planning.