Assessment of landslide-triggering rainfall thresholds is useful for early warning in prone areas. In this paper, it is shown how stochastic rainfall models and hydrological and slope stability physically based models can be advantageously combined in a Monte Carlo simulation framework to generate virtually unlimited-length synthetic rainfall and related slope stability factor of safety data, exploiting the information contained in observed rainfall records and field-measurements of soil hydraulic and geotechnical parameters. The synthetic data set, dichotomized in triggering and non-triggering rainfall events, is analyzed by receiver operating characteristics (ROC) analysis to derive stochastic-input physically based thresholds that optimize the trade-off between correct and wrong predictions. Moreover, the specific modeling framework implemented in this work, based on hourly analysis, enables one to analyze the uncertainty related to variability of rainfall intensity within events and to past rainfall (antecedent rainfall). A specific focus is dedicated to the widely used power-law rainfall intensity–duration (I–D) thresholds. Results indicate that variability of intensity during rainfall events influences significantly rainfall intensity and duration associated with landslide triggering. Remarkably, when a time-variable rainfall-rate event is considered, the simulated triggering points may be separated with a very good approximation from the non-triggering ones by a I–D power-law equation, while a representation of rainfall as constant–intensity hyetographs globally leads to non-conservative results. This indicates that the I–D power-law equation is adequate to represent the triggering part due to transient infiltration produced by rainfall events of variable intensity and thus gives a physically based justification for this widely used threshold form, which provides results that are valid when landslide occurrence is mostly due to that part. These conditions are more likely to occur in hillslopes of low specific upslope contributing area, relatively high hydraulic conductivity and high critical wetness ratio. Otherwise, rainfall time history occurring before single rainfall events influences landslide triggering, determining whether a threshold based only on rainfall intensity and duration may be sufficient or it needs to be improved by the introduction of antecedent rainfall variables. Further analyses show that predictability of landslides decreases with soil depth, critical wetness ratio and the increase of vertical basal drainage (leakage) that occurs in the presence of a fractured bedrock.

Uncertainty in rainfall datasets and landslide inventories is known to have negative impacts on the assessment of landslide-triggering thresholds. In this paper, we perform a quantitative analysis of the impacts of uncertain knowledge of landslide initiation instants on the assessment of rainfall intensity–duration landslide early warning thresholds. The analysis is based on a synthetic database of rainfall and landslide information, generated by coupling a stochastic rainfall generator and a physically based hydrological and slope stability model, and is therefore error-free in terms of knowledge of triggering instants. This dataset is then perturbed according to hypothetical reporting scenarios that allow simulation of possible errors in landslide-triggering instants as retrieved from historical archives. The impact of these errors is analysed jointly using different criteria to single out rainfall events from a continuous series and two typical temporal aggregations of rainfall (hourly and daily). The analysis shows that the impacts of the above uncertainty sources can be significant, especially when errors exceed 1 day or the actual instants follow the erroneous ones. Errors generally lead to underestimated thresholds, i.e. lower than those that would be obtained from an error-free dataset. Potentially, the amount of the underestimation can be enough to induce an excessive number of false positives, hence limiting possible landslide mitigation benefits. Moreover, the uncertain knowledge of triggering rainfall limits the possibility to set up links between thresholds and physio-geographical factors.

Rainfall intensity-duration landslide-triggering thresholds have become widespread for the development of landslide early warning systems. Thresholds can be in principle determined using rainfall event datasets of three types: (a) rainfall events associated with landslides (triggering rainfall) only, (b) rainfall events not associated with landslides (non-triggering rainfall) only, (c) both triggering and non-triggering rainfall. In this paper, through Monte Carlo simulation, we compare these three possible approaches based on the following statistical properties: robustness, sampling variation, and performance. It is found that methods based only on triggering rainfall can be the worst with respect to those three investigated properties. Methods based on both triggering and non-triggering rainfall perform the best, as they could be built to provide the best trade-off between correct and wrong predictions; they are also robust, but still require a quite large sample to sufficiently limit the sampling variation of the threshold parameters. On the other side, methods based on non-triggering rainfall only, which are mostly overlooked in the literature, imply good robustness and low sampling variation, and performances that can often be acceptable and better than thresholds derived from only triggering events. To use solely triggering rainfall—which is the most common practice in the literature—yields to thresholds with the worse statistical properties, except when there is a clear separation between triggering and non-triggering events. Based on these results, it can be stated that methods based only on non-triggering rainfall deserve wider attention. Methods for threshold identification based on only non-triggering rainfall may have the practical advantage that can be in principle used where limited information on landslide occurrence is available (newly instrumented areas). The fact that relatively large samples (about 200 landslides events) are needed for a sufficiently precise estimation of threshold parameters when using triggering rainfall suggests that threshold determination in future applications may start from identifying thresholds from non-triggering events only, and then move to methods considering also the triggering events as landslide information starts to become more available.

A key component for landslide early warning systems (LEWSs) is constituted by thresholds providing the conditions above which a landslide can be triggered. Traditionally, thresholds based on rainfall characteristics have been proposed, but recently, the hydrometeorological approach, combining rainfall with soil moisture or catchment storage information, is becoming widespread. Most of the hydrometeorological thresholds proposed in the literature use the soil moisture from a single layer (i.e., depth or depth range). On the other hand, multi-layered soil moisture information can be measured or can be available from reanalysis projects as well as from hydrological models. Approaches using this multi-layered information are lacking, perhaps because of the need to keep the thresholds simple and two-dimensional. In this paper, we propose principal component analysis (PCA) as an approach for deriving two-dimensional hydrometeorological thresholds that use multi-layered soil moisture information. To perform a more objective assessment we also propose a piecewise linear equation for the identification of the threshold's shape, which is more flexible than traditional choices (e.g., power law or bilinear). Comparison of the receiver operating characteristic (ROC) (true skill statistic, TSS) of thresholds based on single- and multi-layered soil moisture information also provides a novel tool for identifying the significance of multi-layered information on landslide triggering in a given region. Results for Sicily island, considering the ERA5-Land reanalysis soil moisture data (available at four different depth layers), corroborate the advantages of the hydrometeorological approach gained in spite of the coarse spatial resolution and the limited accuracy of reanalysis data. Specifically, the TSS of traditional precipitation intensity–duration thresholds is equal to 0.5, while those of the proposed hydrometeorological thresholds is significantly higher (TSS=0.71). For the analyzed region, however, multi-layered information seems not to be relevant, as performances in terms of TSS are similar to those obtained with single-layer soil moisture at the upper depths, namely 0–7 and 7–28 cm, which can imply that in Sicily landslide phenomena are mainly influenced by soil moisture in most shallow soil layers.