PubAg

Main content area

Using CUDA to accelerate uncertainty propagation modelling for landslide susceptibility assessment

Author:
Sandric, Ionut, Ionita, Cristian, Chitu, Zenaida, Dardala, Marian, Irimia, Radu, Furtuna, Felix Titus
Source:
Environmental modelling & software 2019 v.115 pp. 176-186
ISSN:
1364-8152
Subject:
Monte Carlo method, computer software, geographic information systems, landscapes, landslides, models, topography, uncertainty, weight-of-evidence
Abstract:
The study is focused on modelling uncertainty propagation from GIS data sources and on assessing their influence on landslide susceptibility modelling. A complete set of tools was developed and written in C++ programming language, Python, and based on NVIDIA CUDA technology for terrain analysis. These tools are using Monte Carlo simulations to generate noise in elevation values and spatial delineation of landslides bodies. The uncertainty propagation is assessed using pixel based cumulative probabilities statistics at the pixel level. Thus, for each pixel, from the landslide susceptibility map, an estimation of landslide susceptibility uncertainty was obtained and spatially visualised. The results show that weight of evidence is a robust method and is not significantly influenced by small-scale variations in the primary topographic attributes. The toolbox and the source code are available under the MIT license.
Agid:
6347596