Rainfall-induced shallow landslides and soil wetness: comparison of physically based and probabilistic predictions

Landslides are an impacting natural hazard in alpine regions, calling for effective forecasting and warning systems. Here we compare two methods (physically based and probabilistic) for the prediction of shallow rainfall-induced landslides in an application to Switzerland, with a specific focus on the value of antecedent soil wetness. First, we show that landslide susceptibility predicted by the factor of safety in the infinite slope model is strongly dependent on soil data inputs, limiting the hydrologically active range where landslides can occur to only ∼20 % of the country with typical soil parameters and soil depth models, not accounting for uncertainty. Second, we find the soil saturation estimate provided by a conceptual hydrological model (PREVAH) to be more informative for landslide prediction than that estimated by the physically based coarse-resolution model (TerrSysMP), which we attribute to the lack of temporal variability and coarse spatial resolution in the latter. Nevertheless, combining the soil water state estimates in TerrSysMP with the infinite slope approach improves the separation between landslide triggering and non-triggering rainfall events. Third, we demonstrate the added value of antecedent soil saturation in combination with rainfall thresholds. We propose a sequential threshold approach, where events are first split into dry and wet antecedent conditions by an N d (day) antecedent soil saturation threshold, and then two different total rainfall–duration threshold curves are estimated. This, among all different approaches explored, is found to be the most successful for landslide prediction.

Integration of observed and model derived groundwater levels in landslide threshold models in Rwanda

Incorporation of specific regional hydrological characteristics in empirical statistical landslide threshold models has considerable potential to improve the quality of landslide predictions towards reliable early warning systems. The objective of this research was to test the value of regional groundwater level information, as a proxy for water storage fluctuations, to improve regional landslide predictions with empirical models based on the concept of threshold levels. Specifically, we investigated: i) the use of a data driven time series approach to model the regional groundwater levels based on short duration monitoring observations; ii) the predictive power of single variable and bilinear threshold landslide prediction models derived from groundwater levels and precipitation. Based on statistical measures of the model fit (R2 and RMSE), the groundwater level dynamics estimated by the transfer function noise time series model are broadly consistent with the observed groundwater levels. The single variable threshold models derived from groundwater levels exhibited the highest landslide prediction power with 82–93 % of true positive alarms despite the quite high rate of false alarms with about 26–38 %. Further combination as bilinear threshold models reduced the rate of false alarms by about 18–28 % at the expense of reduced true alarms by about 9–29 % and thus, being less advantageous than single variable threshold models. In contrast to precipitation based thresholds, relying on threshold models exclusively defined using hydrological variables such as groundwater levels can lead to improved landslide predictions due to their implicit consideration of long-term antecedent conditions until the day of landslide occurrence.

Automated Landslide-Risk Prediction Using Web GIS and Machine Learning Models

Spatial susceptible landslide prediction is the one of the most challenging research areas which essentially concerns the safety of inhabitants. The novel geographic information web (GIW) application is proposed for dynamically predicting landslide risk in Chiang Rai, Thailand. The automated GIW system is coordinated between machine learning technologies, web technologies, and application programming interfaces (APIs). The new bidirectional long short-term memory (Bi-LSTM) algorithm is presented to forecast landslides. The proposed algorithm consists of 3 major steps, the first of which is the construction of a landslide dataset by using Quantum GIS (QGIS). The second step is to generate the landslide-risk model based on machine learning approaches. Finally, the automated landslide-risk visualization illustrates the likelihood of landslide via Google Maps on the website. Four static factors are considered for landslide-risk prediction, namely, land cover, soil properties, elevation and slope, and a single dynamic factor i.e., precipitation. Data are collected to construct a geospatial landslide database which comprises three historical landslide locations—Phu Chifa at Thoeng District, Ban Pha Duea at Mae Salong Nai, and Mai Salong Nok in Mae Fa Luang District, Chiang Rai, Thailand. Data collection is achieved using QGIS software to interpolate contour, elevation, slope degree and land cover from the Google satellite images, aerial and site survey photographs while the physiographic and rock type are on-site surveyed by experts. The state-of-the-art machine learning models have been trained i.e., linear regression (LR), artificial neural network (ANN), LSTM, and Bi-LSTM. Ablation studies have been conducted to determine the optimal parameters setting for each model. An enhancement method based on two-stage classifications has been presented to improve the landslide prediction of LSTM and Bi-LSTM models. The landslide-risk prediction performances of these models are subsequently evaluated using real-time dataset and it is shown that Bi-LSTM with Random Forest (Bi-LSTM-RF) yields the best prediction performance. Bi-LSTM-RF model has improved the landslide-risk predicting performance over LR, ANNs, LSTM, and Bi-LSTM in terms of the area under the receiver characteristic operator (AUC) scores by 0.42, 0.27, 0.46, and 0.47, respectively. Finally, an automated web GIS has been developed and it consists of software components including the trained models, rainfall API, Google API, and geodatabase. All components have been interfaced together via JavaScript and Node.js tool.