MLEWS: a series of software systems to implement machine learning for geo-hazards risk early warning based on meteorological factors

Sudden and frequent geo-hazards pose serious threats to life and property, highlighting the need for an integrated, multi-disciplinary early warning system (EWS). However, incorporating big data and machine learning into traditional EWSs brings challenges such as complex model structures, inefficient data fusion, and limited operational transferability. To address this, we propose a trinity machine learning-based EWS, comprising dual-driven sample construction, algorithmic refinement, and business integration. The system includes three modules: training sample construction, model learning/training, and early-warning operation. These modules can function independently or as a unified platform. A case study in Lin'an County, Zhejiang Province, demonstrates the system's practical value. The model achieves a high accuracy of 0.94–0.96 and maintains strong generalization without overfitting. For rainfall-induced hazards, yellow-warning hit rates exceed two-thirds, and the hazard density in the warning areas is 26 times that of the original statistical model. These results confirm the system’s effectiveness in real-world applications and its advantages in bridging disciplinary gaps and improving both model training and operational efficiency.