Evaluation and prediction of compound geohazards in highly urbanized regions across China's Greater Bay Area

Abstract

Natural hazards could have devastating consequences globally, making hazard assessment and spatial prediction crucial for enhancing the resilience of urbanized regions. However, current disaster prediction and assessment research often neglect the compound effects between multiple geohazards highly in urbanized regions. To address the concern, we employed comparative methodology, evaluating four machine learning algorithms—Extreme Gradient Boosting (XGBoost), Random Forest (RF), Back Propagation Neural Network (BP), and Long Short-Term Memory (LSTM)—in the creation of Geohazard Susceptibility Maps (GSM) for the highly urbanized Guangdong-Hong Kong-Macao Greater Bay Area (GBA). Additionally, the study investigated the triggering mechanisms and the compound interaction between multiple geohazards using the conditional vine copula model. The results showed that the XGBoost model outperformed other models (AUC = 0.89) for predicting multiple geohazards. Geohazards were predominantly concentrated in urban areas in the GBA, with surface subsidence being the most severe, followed by collapse and landslide. The primary triggers for multi-geohazards include distance to roads, slope length, and lithology, with slope length and lithology identified as the primary causative factors in urban areas. Urbanization within the GBA increased the probability of multi-geohazards by 10%, compared to their univariate counterparts. Urban regions exhibited increased risks of landslides, surface subsidence, and collapse by approximately 31%, 44%, and 32%, respectively compared to non-urban regions. Additionally, compound geohazards in the GBA were primarily triggered by heavy rainfall, resulting in the formation of landslide-collapse and collapse-landslide geohazard chains. The probability of compound geohazards is approximately 5% lower than that of univariate geohazards. This is because compound geohazards necessitate a higher cumulative rainfall, and the rainfall threshold was approximately 2–3 times higher than that of univariate geohazards. In the cascading hazard pattern, the occurrence of primary geohazards during local heavy rain increased the probability of secondary geohazards by approximately 10%. The study provides essential insights for mitigating compound geohazards in urbanized areas.