Using Machine Learning Techniques in Multi -Hazards Assessment of Golestan National Park, Iran

Abstract

Golestan National Park is one of the oldest biosphere reserves exposed to environmental hazards due to growing demand, geographical location of the park, mountainous conditions, and developments in the last five decades. The purpose of this study was to evaluate potential environmental hazards using machine-learning techniques. In this study, maximum entropy, random forest, boosted regression tree, generalized additive model, and support vector machine methods were applied to model environmental hazards and evaluate the impact of affecting agents, their area of influence, and interactions. After data collection and preprocessing, the models were implemented, tuned, and trained, and their accuracies were determined using the receiver operating characteristic curve. The results indicate the high importance of climatic and human variables, including rainfall, temperature, presence of shepherds, and villagers for fire hazards, elevation, transit roads, temperature, and rainfall for the formation of floodplains, and elevation, transit roads, rainfall, and topographic wetness index in the occurrence of landslides in the national park. The boosted regression tree model with a ROC value of 0.98 for flooding, 0.97 for fire, and 0.93 for landslide hazards, had the best performance. The modeling estimated that, on average, 16.2% of the area of Golestan National Park has a high potential for landslides, 14% has a high potential for fire, and 7.2% has a high potential for flooding.