Integration of effective antecedent rainfall to improve the performance of rainfall thresholds for landslide early warning in Wanzhou District, China

Empirical rainfall thresholds are widely utilized in landslide early warning systems (LEWS) at different scales. However, the approach ignores complex hydrological processes that predispose slopes to instability, leading to a relatively lower performance with high false alarm rates. The objective of this study is to address this limitation by proposing an updated 3D rainfall threshold approach that combines an assessment of the peak rainfall intensity with the contribution of antecedent rainfall conditions. While the former is obtained with a traditional intensity–duration (I–D) threshold approach, the latter is based on a purposely developed effective antecedent rainfall index (EARI), representing the most proximate regional soil moisture condition related to landslides. Thus, thresholds evolved from lines in the 2D space to planes in the 3D space, which were customized for 11 alert zones in Wanzhou District, China. The results highlight that the participation of EARI operates a consistent decrease in false alarms (ranging from 3.5 % to 94.8 % compared to the I-D approach). Beyond that, the power exponent decay EARI is more reliable than a simple sum-based antecedent rainfall in correctly identifying landslide conditions, resulting in higher performances up to 52.3 % if an operational application is simulated. The updated 3D threshold can be considered a good prototype for developing a LEWS because it evaluates both triggering rainfall and antecedent hydrological conditions with good performance and robustness. The general framework of the model could also be exported to other places, given the relatively simple structure and the wide availability of the input data needed.