Modern aspects of the use of artificial intelligence for predicting natural disasters on the rivers of the Russian Federation (using the example of the Amur River)

Abstract

Among all observed natural disasters, water-related disasters are the most frequent and pose a serious threat to people and socio-economic development. River floods are the most relevant for the Russian Federation, and the importance of flood control, particularly in the Far East, was repeatedly stressed by Russian President Vladimir Putin. The quality of performance of various artificial intelligence methods on the task of predicting river floods in the Amur River basin was investigated. The uniqueness of the research lies in the fact that similar studies have not previously been conducted for this river. The main task of the work was the subsequent practical application of the obtained results in flood forecasting and risk management systems. For this purpose, the best method was searched among widely used methods on the market, which have a rich choice of auxiliary solutions: gradient tree binning, linear regression without regularisation and neural networks. The study design focus on achieving maximum reproducibility of the results. The gradient boosting over the trees in the domestic implementation of CatBoost showed the highest quality. The results of this work can be extrapolated to other rivers comparable in both area and volume of data collected.

Keywords

• disaster management
• floods forecasting
• amur river
• machine learning
• linear regression
• neural network
• gradient boosting