Evaluating performance of simplified physically based models for shallow landslide susceptibility

Abstract

Abstract. Rainfall-induced shallow landslides can lead to loss of life and significant damage to private and public properties, transportation systems, etc. Predicting locations that might be susceptible to shallow landslides is a complex task and involves many disciplines: hydrology, geotechnical science, geology, hydrogeology, geomorphology, and statistics. Two main approaches are commonly used: statistical or physically based models. Reliable model applications involve automatic parameter calibration, objective quantification of the quality of susceptibility maps, and model sensitivity analyses. This paper presents a methodology to systemically and objectively calibrate, verify, and compare different models and model performance indicators in order to identify and select the models whose behavior is the most reliable for particular case studies.The procedure was implemented in a package of models for landslide susceptibility analysis and integrated in the NewAge-JGrass hydrological model. The package includes three simplified physically based models for landslide susceptibility analysis (M1, M2, and M3) and a component for model verification. It computes eight goodness-of-fit indices by comparing pixel-by-pixel model results and measurement data. The integration of the package in NewAge-JGrass uses other components, such as geographic information system tools, to manage input-output processes, and automatic calibration algorithms to estimate model parameters.The system was applied for a case study in Calabria (Italy) along the Salerno-Reggio Calabria highway, between Cosenza and Altilia. The area is extensively subject to rainfall-induced shallow landslides mainly because of its complex geology and climatology. The analysis was carried out considering all the combinations of the eight optimized indices and the three models. Parameter calibration, verification, and model performance assessment were performed by a comparison with a detailed landslide inventory map for the area. The results showed that the index distance to perfect classification in the receiver operating characteristic plane (D2PC) coupled with the model M3 is the best modeling solution for our test case.

Sensitivity analysis and calibration of a dynamic physically based slope stability model

“…Further performance indicators could be used for this task instead (e.g. Formetta et al, 2016;Mergili et al, 2017). However, for validating the results of physically based slope stability models, a performance indicator which is independent of a threshold (such as the AUC) can be misleading.…”

“…The coordinates of the point in the ROC plot where the FOS falls below 1.0 represent the correctly predicted fractions of observed landslides (true positives; TP) and non-landslides (true negatives; TN). The basic idea of the calibration procedure is to identify parameter value combinations which result in an optimum prediction of observed landslides and non-landslides, at a FOS threshold falling below 1.0, by minimizing the distance to the perfect classification (D2PC; Formetta et al, 2016;Mergili et al, 2017; The identification of "behavioural model runs" out of the 10 000 calibration runs is based on the following observations and assumptions:…”

A hydroclimatological approach to predicting regional landslide probability using Landlab

“…While using transient hydrologic models provided slight improvements in the prediction of landslide locations, overall, statistical comparisons of model outputs between steadystate and transient models revealed fairly similar degrees of success (Gorsevski et al, 2006;Zizioli et al, 2013;Anagnostopoulos et al, 2015;Bordoni et al, 2015;Formetta et al, 2016). In some applications, model complexity increased the accuracy of predicted landslide locations at the expense of overestimating instability on unsaturated hillslopes (e.g., Godt et al, 2008;Bellugi, 2011).…”

Probabilistic landslide ensemble prediction systems: lessons to be learned from hydrology

“…Vertical groundwater flow and one-dimensional slope stability in a two-dimensional array of noninteracting columns can subsequently be computed independently for each cell, which is a prime example for parallelization purposes Alvioli and Baum, 2016). In addition to TRIGRS v2.1, which received its parallel implementation from Alvioli and Baum (2016) only recently, other models for physically based landslide applications use a parallelized module: NewAge-JGrass (Formetta et al, 2016) or r.slope.stability (Mergili et al, 2014a). In our case study, the computational time for one model iteration is about 45 min, which is far too long for computing a large set of different ensemble members in an operational real-time application.…”

“…These models usually contain a hydraulic and a slope stability component with different degrees of simplification (Formetta et al, 2016). In most cases, the target variable is the slope safety factor (FoS), which is useful as it enables decision makers to take actions when it falls short of a certain threshold (the slope is unstable with FoS < 1.0, but higher thresholds are used in practice).…”