High-resolution Bayesian mapping of landslide hazard with unobserved trigger event

Opitz, Thomas; Bakka, Haakon; Huser, Raphaël; Lombardo, Luigi

doi:10.1214/21-aoas1561

Cited by 12 publications

(3 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Specifically, adjacency matrix (see Fig. 3 of Opitz et al 2022) can control this information which is then passed to the model as a latent covariate. The relation above corresponds to a Generalized Additive Mixed Models (GAMM, Steger et al 2021), which we implement here in its Bayesian form via INLA (Bakka et al 2018).…”

Section: Landslide Intensity Modelingmentioning

confidence: 99%

On the estimation of landslide intensity, hazard and density via data-driven models

Napoli¹,

Tanyaş²,

Castro‐Camilo³

et al. 2022

Preprint

View full text Add to dashboard Cite

Maps that attempt to predict landslide occurrences have essentially stayed the same since Brabb, E.E., Pampeyan, E.H. and Bonilla, M.G. (1972) Landslide susceptibility in San Mateo County, California (No. 360), US Geological Survey. The tools have certainly changed in fifty years. But, the geomorphological community addressed and still addresses this issue by estimating whether a given slope is potentially stable or unstable. This concept corresponds to the landslide susceptibility, a paradigm that entirely neglects how many landslides may trigger within a given slope, how large these landslides may be and what proportion of the given slope they may disrupt. Modeling how many landslides may occur per mapping unit has been recently proposed via the landslide intensity concept, which has later been shown to closely correlate to the planimetric extent of landslides per mapping unit. In this work, we take this observation a step further as we use the relation between landslide intensity and planimetric extent to generate maps that predict the aggregated size of landslides per mapping unit, and the proportion they may affect. Our findings suggest that it may be time for the geoscientific community as a whole, to revise the use of susceptibility assessment in favour of more informative analytical schemes. Our chain of landslide intensity, hazard and density may in fact lead to substantially improve decision-making processes related to landslide risk.

show abstract

Section: Landslide Intensity Modelingmentioning

confidence: 99%

On the estimation of landslide intensity, hazard and density via data-driven models

Napoli¹,

Tanyaş²,

Castro‐Camilo³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Specifically for statistical studies a common assumption is the choice of a suitable distribution reflecting the data on landslides. For this reason, susceptibility models assume a Bernoulli probability distribution (Steger et al, 2016;Steger et al, 2017), whereas intensity models based on counts assume the Poisson probability distribution (Lombardo et al, 2019;Opitz et al, 2022) instead. When it comes to model landslide area, the choice is not straightforward.…”

Section: Introductionmentioning

confidence: 99%

A data-driven framework for landslide size space-time modelling

Fang¹,

Wang²,

Westen³

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Landslide susceptibility assessment using data-driven models has predominantly focused on predicting where landslides may occur and not on how large they might be. The spatio-temporal evaluation of landslide susceptibility has only recently been addressed, as a basis for predicting where and when landslides might occur. The present study combines these new developments by proposing a data-driven model capable of estimating how large landslides may be, for the Taiwan territory in a fourteen year time window. To solve this task, our model assumes that landslide sizes follow a Log-Gaussian probability distribution in space and time. Spatially the area is subdivided into 46074 slope units, with 14 annual timesteps from 2004 to 2018. Based on this subdivision, the model we implemented regressed landslide sizes against a covariate set that includes temporally static and dynamic properties. In the validation of our model, we nested a wide range of cross-validation (CV) procedures, such as a randomized 10fold-CV, a spatially constrained CV, a temporal leave-one-year-out CV, and a spatio-temporal CV. The final performance was described both numerically as well as in map forms. Overall, our space-time model achieves interpretable and satisfying results. With the availability of more complete landslide inventories, both temporally and spatially, we envision that spatio-temporal landslide size prediction will become the next challenge for geomorphologists to finally address a fundamental component of the landslide hazard definition. And, because of its spatio-temporal nature, we also envision that it may lead to simulation studies for varying climate scenarios.

show abstract

“…Another appealing advantage of statistical-based models is their capability to capture and display spatial effects (Song et al, 2020), such as spatially varying coefficients models (e.g., Geographically Weighted Regression, Fotheringham et al, 2003) or (e.g., Spatially Varying Regression, Opitz et al, 2022). However, restricted by the data size and the relationships' complexity, statistical models are usually computationally challenging when dealing with big spatial data (Lombardo et al, 2019).…”

Section: Introductionmentioning

confidence: 99%

On the use of explainable AI for susceptibility modeling: examining the spatial pattern of SHAP values

Wang¹,

Zhang²,

Dahal³

et al. 2023

Preprint

View full text Add to dashboard Cite

Hydro-morphological processes (HMP, any natural phenomenon contained within the spectrum defined between debris flows and flash floods) are globally occurring natural hazards which pose great threats to our society, leading to fatalities and economical losses. For this reason, understanding the dynamics behind HMPs is needed to aid in hazard and risk assessment. In this work, we take advantage of an explainable deep learning model to extract global and local interpretations of the HMP occurrences across the whole Chinese territory. We use a neural network architecture and interpret the model results through the spatial pattern of SHAP values.In doing so, we can understand the model prediction on a hierarchical basis, looking at how the predictor set controls the overall susceptibility as well as doing the same at the level of the single mapping unit.Traditional statistical tools usually lead to a clear interpretation at the expense of large performance, which is otherwise reached via machine/deep learning solutions, though at the expense of interpretation. Explainable AI is the key to combine both strengths and in this work, we explore this combination in the context of HMP susceptibility modeling. Specifically, we demonstrate the extent to which one can enter a new level of data-driven interpretation, supporting the decision-making process behind disaster risk mitigation and prevention actions.

show abstract

High-resolution Bayesian mapping of landslide hazard with unobserved trigger event

Cited by 12 publications

References 50 publications

On the estimation of landslide intensity, hazard and density via data-driven models

On the estimation of landslide intensity, hazard and density via data-driven models

A data-driven framework for landslide size space-time modelling

On the use of explainable AI for susceptibility modeling: examining the spatial pattern of SHAP values

Contact Info

Product

Resources

About