Сравнение методов прогнозирования весеннего половодья на примере р. Пур в створе п. Самбург

Traditional hydrological forecasting methods based on linear relationships between snow cover, precipitation and runoff often demonstrate insufficient accuracy, especially under conditions of variable climate and extreme hydrometeorological events. This article analyzes the effectiveness of classical approaches using the example of the Pur River (Samburg), which revealed significant errors in forecasting the runoff depth (up to 27 %), flood volume (up to 29 %) and maximum water discharges (up to 105 %). These results highlight the limitations of existing methods and the need to implement modern technologies.

As an alternative, the use of neural network models capable of accounting for nonlinear relationships is considered. Neural networks demonstrated high forecasting accuracy with an error of 5 cm for maximum water levels. Their key advantages include adaptability to new data and resilience to anomalous conditions.

The results of the study confirm that neural network methods can be successfully used in operational practice to improve hydrological forecasts. The article also offers recommendations for the implementation of hybrid approaches that combine physical-statistical models with machine learning, and for the development of automated monitoring systems. The transition to artificial intelligence technologies is an important step in increasing the reliability of forecasts and minimizing the risks associated with floods and other hydrological phenomena.