Xây dựng bản đồ cảnh báo sạt lở đất theo thời gian thực cho tỉnh Lào Cai sử dụng các nguồn dữ liệu mở và công nghệ học máy

Abstract

In the context of increasing frequency and destructive intensity of landslides in Lào Cai Province - an area characterized by steep and highly dissected terrain - the development of an early warning system for landslides is an urgent requirement to mitigate losses of life and property. However, current warning efforts face considerable challenges due to the limited availability of traditional observation data. This study leverages open-access, near real-time datasets in combination with the Random Forest machine learning algorithm - an efficient ensemble learning method for classification - to construct a landslide hazard warning map for Lào Cai Province. The input dataset includes observed landslide locations along with conditioning factors such as slope, aspect, rainfall, soil moisture, soil type, land cover, distance to roads and rivers, as well as real-time rainfall forecasts. Model evaluation shows an overall accuracy of 85% and a Kappa coefficient of 0.69, with ROC and PRC analyses indicating high reliability, thereby demonstrating the feasibility and effectiveness of this approach. The findings highlight the potential of open-source data-driven machine learning in developing early warning systems for landslide-prone mountainous regions where observational data remain scarce, such as Lào Cai.