ANALYTICAL MODELING OF HYDROLOGICAL CHARACTERISTICS BASED ON HYDROLOGICAL STATION DATA AND WATER BALANCES OF THE DNIESTER RIVER

Abstract

The article focuses on the analytical modeling of hydrological characteristics of the Dniester River using data obtained from open sources, including the website of the State Agency of Water Resources of Ukraine and the Ukrainian Hydrometeorological Center. The analysis covered key indicators such as the water management balances of the Dniester basin and water level data from the "Sambir" hydrological station. For further processing, the data were cleaned and structured into datasets.

The objective of the study is to apply data analysis methods to identify water management segments with critical deficit values and to forecast their hydrological fluctuations with a high level of accuracy. This involves performing spatiotemporal and exploratory data analyses, followed by anomaly-aware modeling using various time series models with regression components.

The analysis of water management balances revealed that three out of twelve segments exhibit a very high deficit, and some segments demonstrate pronounced instability of reserves across supply scenarios, indicating vulnerability to hydrological fluctuations. Special attention is given to the comparison of various machine learning models and statistical analyses, which allows selecting the most suitable tool for forecasting. Among the models evaluated, Prophet showed the best performance, demonstrating the highest efficiency according to the MAPE criterion.

The conducted research confirms the high effectiveness of integrating geoinformation technologies and statistical analysis tools in forecasting hydrological characteristics, particularly water level modeling. The application of machine learning enables a high degree of reliability in short-term forecasting, which is of critical practical importance for timely decision-making in hydrological monitoring and water management.