Application of machine learning methods for studying ice-jam floods on mountain rivers of the Almaty region

Introduction. The article presents the results of the study of catastrophic water-ice flows that occurred on the Esentai River on 13.12.2022 and 16.12.2023. They had high flow speeds (4–5 m/s) and differed from slush ice drifts on lowland rivers. These flows resembled debris flows, because new material was involved and an avalanche-like increase of the water-ice mixture occurred. The flow movement was wave-like. These flows caused material damage. A similar phenomenon that occurred on January 6, 2006 on the Uzyn Kargaly River in Ile Alatau resulted in casualties. The main objective is to study the hydrological and meteorological conditions of the formation of water-ice flows on mountain rivers. Research methods. For this, the machine learning algorithm “Random Forest” was used. This is a method of automatically clustering data and dividing it into similar groups. It allows one to build “Decision Trees” that separate the meteorological situation and determine threshold values for each group. The results of the research. For the Central Asian climate, cooling down to 20–25 degrees below zero is an anomalous phenomenon. This causes the accumulation of anchor ice and a significant increase in the water level in the channel: 1–2 m above the winter low-water level. If cooling is replaced by a sharp warming with positive temperatures, destruction of ice structures and formation of ice-jam floods occur. Discussion. The difficulty in using this algorithm lies in interpreting the data and comparing it with already known results, since some machine “Classification Trees” take into account only general climate features. But certain threshold values are already confirmed by other scientific researches. This means that the algorithm works and produces results with a minimal statistical error. Resume. Threshold values of meteorological elements on the day of ice-jam floods were determined: daily warming gradient was more than 3.3 °C/day, air temperature anomaly was below 2.6 °C, the sum of air temperatures over the previous 5 days was 42.5 °C. They can be used to predict dangerous situations on mountain rivers.