Geological Hazard Meteorological Risk Warning Model Based on Heterogeneous Integration Machine Learning: a Case Study in Yichang City
Chang Liu,Xi Li,Tao Yang,Mengyuan Chen,Feng Pan
DOI: https://doi.org/10.1117/12.3032255
2024-01-01
Abstract:In order to enhance the predictive and forecasting capabilities of meteorological risk warning models for geologic hazards, this paper presents a new meteorological risk warning model for geologic hazards in Yichang City, Hubei Province, using heterogeneously integrated machine learning. Firstly, eight essential environmental factors were extracted from various data sources, including topographic, geological, and remote sensing imagery. Correlation tests were conducted alongside statistical analyses of landslide deformation and damage under diverse external hydrological conditions, such as rainfall intensity and duration. Consequently, nine rainfall warning indexes were derived, comprising metrics like previous period's effective rainfall, same-day inspired rainfall amount, and short-term rainfall intensity. These indexes, in conjunction with the fundamental environmental factors, formed the foundation of the landslide warning index system. Subsequently, multiple modeling techniques including logistic regression, CART decision tree, plain Bayes, a heterogeneous integration method as base models, were employed to establish the geologic hazards meteorological risk model. The results demonstrated that the integrated machine learning model achieved an accuracy rate of 0.835, surpassing the performance of the individual models. Then, the constructed model was further evaluated using historical rainfall and geologic hazards event data, which corroborated the alignment of the prediction outcomes with actual occurrences of geologic hazards. Collectively, this study's findings have important implications for the development of practical meteorological risk warning models for geologic hazards and provide a promising framework for improving regional disaster prevention, as well as emergency management planning.