The rainfall-induced probability threshold for shallow landslides is typically established through statistical analysis and the application of comparable rainfall threshold models, such as rainfall intensity–duration (I–D), cumulative rainfall–duration (E–D), and cumulative rainfall–intensity (I–E). In this study, 311 rainfall-triggered soil landslides in Shangrao City were analyzed to statistically characterize their rainfall conditions. To address the limitation of conventional studies and enhance threshold reliability, an integrated optimization framework was further proposed, combining Pearson correlation-guided parametric calibration, statistical model comparison, and physical simulation validation. First, Pearson correlation analysis confirmed R₃ (R = 0.575) and α = 0.8 as the optimal parameters for the study area. Then, comparative evaluation of the three threshold models showed the I–D model had the highest predictive accuracy. Finally, GeoStudio seepage-stability simulation validated the I–D threshold, ensuring it aligns with the hydromechanical processes governing slope instability.The critical threshold curves were delineated via the ordinary least squares (OLS) method, and quantitative analysis was employed to identify the optimal model for landslide early warning in Shangrao. The results indicate that: (1) The parameter Re in the threshold model, derived from landslide events in the study area, showed the strongest correlation with the 3-day antecedent cumulative rainfall (R3). This suggests that the 3-day rainfall accumulation can serve as a key indicator for defining landslide early warning criteria, with optimal performance achieved when the effective rainfall model parameter reaches 0.8; (2) Among the three rainfall threshold models, the I-D model demonstrated superior accuracy and is considered more suitable for landslide monitoring and early warning in the study area; (3) A three-level early warning system was developed: Alarm Level (I > 71.287D−0.770), Alert Level (39.286D−0.770 < I < 71.287D−0.770), and Attention Level (15.355D−0.770 < I < 39.286D−0.770). Corresponding expressions for each warning level were formulated, and a methodology for optimizing rainfall threshold parameters was proposed to enhance landslide monitoring and early warning capabilities.

Optimization of rainfall threshold models for shallow soil landslides in Shangrao, South China: a Pearson correlation-guided parametric calibration and physical validation study

期刊论文
发表日期：

2026年7月15日

查看全文

Jianjun Gan,
Qing Liu,
Yu Kaining,
Liu Fangzhou,
Yanwen Xin,
Hao Chen

摘要

The rainfall-induced probability threshold for shallow landslides is typically established through statistical analysis and the application of comparable rainfall threshold models, such as rainfall intensity–duration (I–D), cumulative rainfall–duration (E–D), and cumulative rainfall–intensity (I–E). In this study, 311 rainfall-triggered soil landslides in Shangrao City were analyzed to statistically characterize their rainfall conditions. To address the limitation of conventional studies and enhance threshold reliability, an integrated optimization framework was further proposed, combining Pearson correlation-guided parametric calibration, statistical model comparison, and physical simulation validation. First, Pearson correlation analysis confirmed R₃ (R = 0.575) and α = 0.8 as the optimal parameters for the study area. Then, comparative evaluation of the three threshold models showed the I–D model had the highest predictive accuracy. Finally, GeoStudio seepage-stability simulation validated the I–D threshold, ensuring it aligns with the hydromechanical processes governing slope instability.The critical threshold curves were delineated via the ordinary least squares (OLS) method, and quantitative analysis was employed to identify the optimal model for landslide early warning in Shangrao. The results indicate that: (1) The parameter R_e in the threshold model, derived from landslide events in the study area, showed the strongest correlation with the 3-day antecedent cumulative rainfall (R₃). This suggests that the 3-day rainfall accumulation can serve as a key indicator for defining landslide early warning criteria, with optimal performance achieved when the effective rainfall model parameter reaches 0.8; (2) Among the three rainfall threshold models, the I-D model demonstrated superior accuracy and is considered more suitable for landslide monitoring and early warning in the study area; (3) A three-level early warning system was developed: Alarm Level (I > 71.287D^−0.770), Alert Level (39.286D^−0.770 < I < 71.287D^−0.770), and Attention Level (15.355D^−0.770 < I < 39.286D^−0.770). Corresponding expressions for each warning level were formulated, and a methodology for optimizing rainfall threshold parameters was proposed to enhance landslide monitoring and early warning capabilities.