Counteracting flawed landslide data in statistically based landslide susceptibility modelling for very large areas: a national-scale assessment for Austria

Abstract

The reliability of input data to be used within statistically based landslide susceptibility models usually determines the quality of the resulting maps. For very large territories, landslide susceptibility assessments are commonly built upon spatially incomplete and positionally inaccurate landslide information. The unavailability of flawless input data is contrasted by the need to identify landslide-prone terrain at such spatial scales. Instead of simply ignoring errors in the landslide data, we argue that modellers have to explicitly adopt their modelling design to avoid misleading results. This study examined different modelling strategies to reduce undesirable effects of error-prone landslide inventory data, namely systematic spatial incompleteness and positional inaccuracies. For this purpose, the Austrian territory with its abundant but heterogeneous landslide data was selected as a study site. Conventional modelling practices were compared with alternative modelling designs to elucidate whether an active counterbalancing of flawed landslide information can improve the modelling results. In this context, we compared widely applied logistic regression with an approach that allows minimizing the effects of heterogeneously complete landslide information (i.e. mixed-effects logistic regression). The challenge of positionally inaccurate landslide samples was tackled by elaborating and comparing the models for different terrain representations, namely grid cells, and slope units. The results showed that conventional logistic regression tended to reproduce incompleteness inherent in landslide training data in case the underlying model relied on explanatory variables directly related to the data bias. The adoption of a mixed-effects modelling approach appeared to reduce these undesired effects and led to geomorphologically more coherent spatial predictions. As a consequence of their larger spatial extent, the slope unit–based models were able to better cope with positional inaccuracies of the landslide data compared to their grid-based equals. The presented research demonstrates that in the context of very large area susceptibility modelling (i) ignoring flaws in available landslide data can lead to geomorphically incoherent results despite an apparent high statistical performance and that (ii) landslide data imperfections can actively be diminished by adjusting the research design according to the respective input data imperfections.

Literature review and bibliometric analysis on data-driven assessment of landslide susceptibility

“…Recent advances in computer hardware and storage have made it possible to run DdLSM using very detailed raster resolutions, even for very large areas. The interplay between inventory positional accuracy and resolution of the input data should be one of the initial concerns (Lima et al 2021). The role of predictors resolution and its effects on DdLSM, were assessed in publications like Arnone et al (2016); Claessens (2005); Durić et al (2019); Lee et al (2004); Palamakumbure et al (2015); Shirzadi et al (2019).…”

“…For this reason, assuring inventory's best positional accuracy is extremely important. It is known that historical and accurate landslide inventories are virtually never available, especially when dealing with very large areas (Herrera et al 2018;Lima et al 2021;Lin et al 2021; van den Eeckhaut and Hervas 2014). However emerging landslide mapping techniques that enable, immediate and precise event-based cataloging or techniques allowing past landslide extraction from highresolution digital terrain models can positively contribute (Guzzetti et al 2012).…”

“…However, there are nowadays automatic tools able to perform this delineation, guaranteeing its reproducibility (Alvioli et al 2016). Although the modelling unit representation is a particular researcher's decision, this choice should take into account the landslide inventory, predictors' characteristics and the accuracy (Lima et al 2021;Zêzere et al 2017).…”

Estimating global landslide susceptibility and its uncertainty through ensemble modelling

“…In this study, we create a statistical LSS model using MELR (Zuur, 2009), as previously also employed by , and at national scale by Lin et al (2021) and Lima et al (2021). Logistic regression is the most commonly used approach for statistical LSS mapping (Reichenbach et al, 2018), and is associated with strong generalizing capabilities (Brenning, 2005)a necessity when working at the global scale.…”

“…and input uncertainty (here: 'How correct is the input to these equations?'). Model uncertainty stems from heuristic choices that are necessary in the process of model creation, including the choice of the statistical modelling approach, the selection of predictor variables, training data sampling and training data quality (see for example Steger et al (2015); Pourghasemi and Rossi (2016); Zêzere et al (2017); Depicker et al (2020); Lima et al (2021)). In order to estimate these model-intrinsic errors for a chosen modelling approach, cross validation (CV) is a widely used method where data is divided into a number of subsets, that are subsequently used for training and testing of the model.…”