Forecast of ice phenomena on the rivers of the lake Sevan basin

Introduction. The study of ice conditions on rivers is an important task to ensure safe navigation, the operation of a number of hydraulic structures, the laying of oil and gas pipelines, as well as to minimize economic damage from dangerous ice phenomena, which are often accompanied by significant flooding of the territory (floods). One of the main types of ice forecasts can be attributed to forecasts of the timing (dates) of the beginning and end of freeze-up. Despite the fact that both parameters are an element of the same physical phenomenon, they differ significantly in their formation factors. The purpose of the research is to study and apply multivariate statistics methods (decision tree methods and the naive Bayesian classifier) to forecast ice phenomena on the rivers of the Lake Sevan basin, which are characterized by complex processes of opening and freezing, the methods used are based on the use of fairly short time series obtained from the hydrometeorological observation network. Methods and materials of the study. Decision trees and the naive Bayesian classifier method were considered. The advantages of decision trees include automatic parameter selection, interpretability, controllability, finding nonlinear relationships, and the ability to work with a small data set. The naive Bayesian classifier method is based on the application of Bayes' theorem with a naive assumption of independence of variables, where a probability measure and its distribution are taken for each feature. The advantage of this approach is ease of implementation, low computational costs, and a small amount of data for training and classification. The source material was the data of actual hydrometeorological observations (air temperature, water level, ice phenomena on rivers) of the Center for Hydrometeorology and Monitoring of the SNCO of the Ministry of Environment of the Republic of Armenia. Research results. An analysis of ice phenomena for the rivers of the Lake Sevan basin is presented. Testing of the proposed methods showed that the actual onset of ice phenomena begins later than the forecast dates. Also, the models are worse at determining the end date of ice phenomena. When comparing the relative number of correct forecasts for the entire period of the test sample, the naive Bayesian classifier method has a slight advantage in accuracy. This is explained by its ability to better adapt to the variability of time parameters. The decision tree model of classification provides a more visual representation of the decision-making process, which can be useful for analyzing the influence of individual factors on the system as a whole. Conclusions. The scientific study shows that the transformation of forecasting the timing of ice phenomena and river breakup into a binary classification problem is a fairly effective solution when working with a limited set of hydrometeorological data and the absence of a pronounced linear relationship. The proposed methods are quite promising for use in forecasting ice conditions

Bayesian inference · Bayesian network · Bayesian probability · Decision tree · Hydrometeorology · Multivariate statistics · Posterior probability · Structural basin · Aquatic and Environmental Studies · Soil and Environmental Studies · Water Resources and Management