NEURAL NETWORK MULTIFACTOR MODEL FOR OPERATIONAL FORECASTING OF LANDSLIDE ACTIVITY

Abstract

This article discusses a methodology for constructing nonlinear multivariate neural network models based on a multilayer perceptron (MLP) for the rapid forecasting of landslide activity. Using long-term monitoring of the Tectonic landslide (Mayluu-Suu) as an example, two problems are solved: classifying the nature of landslide activity over a 15-day period and regression forecasting the displacement velocity over the next five days. The author conducts a detailed analysis of the neural network's sensitivity to various hydrometeorological factors and historical movement parameters, identifying the most significant predictors for each problem. The study demonstrated the high effectiveness of the neural network approach in predicting landslide velocity regimes, which is important for risk management in hazardous areas.