基于XGBoost的降雨型滑坡区域性气象预警模型OA

[Objective]In response to issues in rainfall-induced landslide early warning in Huangmei County,such as subjectivity in factor weighting of traditional methods,insufficient dynamic response,and the limited ability of single models to explain mechanisms and capture complex relationships,this study constructs a regional meteorological early warning model that integrates the advantages of physical mechanisms and data-driven approaches.[Methods]Based on 10 conventional environmental factors and the failure probability calculated by the TRIGRS model,this study uses the XGBoost algorithm to develop a landslide susceptibility assessment model.By combining historically observed rainfall data associated with landslides and simulation results,differentiated E-D(Effective Duration)rainfall threshold curves are established based on zoning by the ratio of unstable area.These are then converted into multi-temporal-scale cumulative rainfall early warning indicators,forming a four-level meteorological early warning system linked with emergency response.[Results]Verification shows that the accuracy of the XGBoost fusion model reaches 91.2%,significantly outperforming both the single physical model and traditional machine learning models.The differentiated threshold curves effectively distinguish rainfall events of different risk levels.Based on this,a four-level cumulative rainfall early warning index is established,covering five time scales:3 h,6 h,12 h,24 h,and 48 h.[Conclusions]The model integrates the explicability of physical mechanisms with the capability of machine learning to capture non-linear relationships,improving the accuracy and dynamic response capability of rainfall-induced landslide early warning in Huangmei County.The differentiated E-D threshold curves based on zoning by unstable area ratio overcome the limitations of traditional single thresholds,making warnings more targeted.After being converted into multi-temporal-scale cumulative rainfall indicators,the operability of the warning system is enhanced.This model provides a feasible technical solution for rainfall-induced landslide early warning in Huangmei County and other areas with similar geological conditions.