Workflow-induced uncertainty in probabilistic landslide hazard maps

Workflow-induced uncertainty in probabilistic landslide hazard maps Anil Yildiz ([email protected]th-aachen.de) & Julia Kowalski Methods for Model-based Development in Computational Engineering | RWTH Aachen University | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Dublin, Ireland | Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 2 Model-based decision support Quantitative risk assessment •risk = f(hazard, exposure, vulnerability) •hazard = f(probability of occurrence, intensity) •Probability of occurrence •Triggering mechanism, e.g. rainfall •Data-driven, e.g. historical data •Intensity •Maximum flow height •Maximum flow velocity •Maximum flow height & velocity Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 3 Model-based decision support Quantitative risk assessment •risk = f(hazard, exposure, vulnerability) •hazard = f(probability of occurrence, intensity) •Probability of occurrence •Triggering mechanism, e.g. rainfall •Data-driven, e.g. historical data •Intensity •Maximum flow height •Maximum flow velocity •Maximum flow height & velocity Computational models Shallow flow Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 4 Workflow decisions Workflow decisions Numerical approach Eulerian Lagrangian … Solution approach Finite element Finite difference Finite volume Smoothed particle hydrodynamics … Rheology Voellmy Voellmy-Salm Bingham Herschel-Bulkley … Topography Point clouds Digital elevation models Fine grid vs. Coarse grid … Calibration Expert knowledge Brute force calibration Grid-type search Bayesian inference … Uncertainty quantification Point estimate method Monte Carlo simulations First Order Reliability Method First Order Second Moment … Prior distributions Uniform Gaussian Beta, Gamma Assumed vs. informed … Intensity estimation Max. flow height Max. flow velocity Max. flow height & velocity Dynamic pressure … Surrogate modelling Gaussian process Training size Cross-validation Prediction … Modelling approach Depth-averaged continuum mechanics Free-surface continuum mechanics Discrete particles Multi-phase … Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 5 Workflow decisions Workflow decisions Modelling approach Depth-averaged continuum mechanics Free-surface continuum mechanics Discrete particles Multi-phase … Numerical approach Eulerian Lagrangian … Solution approach Finite element Finite difference Finite volume Smoothed particle hydrodynamics … Rheology Voellmy Voellmy-Salm Bingham Herschel-Bulkley … Topography Point clouds Digital elevation models Fine grid vs. Coarse grid … Calibration Expert knowledge Brute force calibration Grid-type search Bayesian inference … Uncertainty quantification Point estimate method Monte Carlo simulations First Order Reliability Method First Order Second Moment … Prior distributions Uniform Gaussian Beta, Gamma Assumed vs. informed … Intensity estimation Max. flow height Max. flow velocity Max. flow height & velocity Dynamic pressure … Surrogate modelling Gaussian process Training size Cross-validation Prediction … Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 6 Uncertainty Quantification Modified from Yildiz et al. 2023. Computationally-feasible uncertainty quantification in model-based landslide risk assessment. Front. Earth Sci: Geohazards and Georisks Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 7 Case study: Rebaixader, Spain •High torrential activity •Torrential flows after rainfall due to sediments and steep slopes •Short and intense rainstorms in summer triggered flows with a short reaction time •Availability of long-term field monitoring data sets Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 8 Case study: Rebaixader, Spain •Autonomous Body National Center for Geographic Information (CNIG) •Digital Terrain Model (DTM) –5 m •Digital Surface Model (DSM) –5 m •Canopy Height Model (CHM = DSM - DTM) •Solved with r.avaflow •200 simulations Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 9 Simulation results Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 16 Risk assessment 𝐯𝒉 < 1.0 1.0 < 𝐯𝒉 ≤ 3.0 3.0 < 𝐯𝒉 ≤ 12.0 𝐯𝒉 > 12.0 𝒉 < 0.5 Low Not defined High Very high 0.5 < 𝒉 ≤ 1.5 Not defined Medium High Very high 1.5 < 𝒉 ≤ 3.0 High High High Very high 𝒉 > 3.0 Very high Very high Very high Very high Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 17 Uncertainty in risk assessment Jakob et al. (2012) Tang et al. (2022) Anil Yildiz | PSimPy: GP emulation-based sensitivity analysis, uncertainty quantification and calibration of landslide simulators | [email protected] | www.mbd.rwth-aachen.de | ICASP14 | MS 20: Probabilistic modelling of natural hazards and associated risks | 12.07.2023 | Slide 18 Conclusions •Uncertainty can affect up to nearly 20% difference in marking high risk zones •Poorly described criteria can alter the decision-making process of risk zoning •With any workflow decision comes an uncertainty •A flexible and modular workflow is required to assess the workflow-induced uncertainty •Certain comparisons benefit from surrogate modelling •Such an assessment can only be achieved through reproducible workflows •Reproducibility based on workflow documentation in yaml format •Metadata scheme to foster re-usability and benchmarking currently under development Thank you for your attention This work was partially funded by Deutsche Forschungsgemeinschaft (DFG) within the framework of the research project OptiData: Improving the Predictivity of Simulating Natural Hazards due to Mass Movements –Optimal Design and Model Selection (Project no. 441527981). The authors would like to extend their gratitude to Marcel Hürlimann (UPC Barcelona) and Claudia Abanco (Universitat de Barcelona) for the fruitful discussions on Rebaixader catchment.