Preprints
https://doi.org/10.5194/gmd-2022-119
https://doi.org/10.5194/gmd-2022-119
Submitted as: methods for assessment of models
18 May 2022
Submitted as: methods for assessment of models | 18 May 2022
Status: this preprint is currently under review for the journal GMD.

Transfer learning for landslide susceptibility modelling using domain adaptation and case-based reasoning

Zhihao Wang, Jason Goetz, and Alexander Brenning Zhihao Wang et al.
  • Friedrich Schiller University Jena, Department of Geography, Loebdergraben 32, 07743 Jena, Germany

Abstract. Transferability of knowledge from well-investigated areas to a new study region is gaining importance in landslide hazard research. Considering the time-consuming compilation of landslide inventories as a prerequisite for landslide susceptibility mapping, model transferability can be key to making hazard-related information available to stakeholders in a timely manner. In this paper, we compare and combine two important transfer-learning strategies for landslide susceptibility modelling: case-based reasoning (CBR) and domain adaptation (DA). CBR gathers knowledge from previous similar situations (source areas) and applies it to solve a new problem (target area). DA, which is widely used in computer vision, selects data from a source area that has a similar distribution to the target area. We assess the performances of single- and multiple-source CBR, DA and CBR-DA strategies to train and combine landslide susceptibility models using generalized additive models (GAMs) for 10 study areas with various resolutions (1 m, 10 m and 25 m) located in Austria, Ecuador, and Italy. The performance evaluation shows that CBR and combined CBR-DA based on our proposed similarity criterion was able to achieve performances comparable to benchmark models trained in the target area itself. Particularly the CBR strategies yielded favourable results in both single- and multi-source strategies. DA tended to have overall lower performances than CBR; yet, it had promising results in scenarios when the source-target similarity was low. We recommend that future transfer learning research for landslide susceptibility modelling can build on the similarity criterion we used, as it successfully helped to achieve landslide susceptibility model transfers by discovering suitable training datasets from various regions.

Zhihao Wang et al.

Status: open (until 13 Jul 2022)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse

Zhihao Wang et al.

Data sets

Paldau and Waidhofen with a 10 m resolution for "Transfer learning for landslide susceptibility modelling using domain adaptation and case-based reasoning" Raphael Knevels, Alexander Brenning, Simone Gingrich, Gerhard Heiss, Theresia Lechner, Philip Leopold, Christoph Plutzar, Herwig Proske and Helene Petschko https://doi.org/10.3390/land10090954

Burgenland with a 10 m resolution for "Transfer learning for landslide susceptibility modelling using domain adaptation and case-based reasoning" Raphael Knevels, Helene Petschko, Philip Leopold and Alexander Brenning https://doi.org/10.3390/ijgi8120551

Model code and software

Methods for "Transfer learning for landslide susceptibility modelling using domain adaptation and case-based reasoning" Zhihao Wang https://github.com/Zhihao-Wang16/GMD_slidetransfer

Zhihao Wang et al.

Viewed

Total article views: 432 (including HTML, PDF, and XML)
HTML PDF XML Total Supplement BibTeX EndNote
352 74 6 432 22 1 1
  • HTML: 352
  • PDF: 74
  • XML: 6
  • Total: 432
  • Supplement: 22
  • BibTeX: 1
  • EndNote: 1
Views and downloads (calculated since 18 May 2022)
Cumulative views and downloads (calculated since 18 May 2022)

Viewed (geographical distribution)

Total article views: 386 (including HTML, PDF, and XML) Thereof 386 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 30 Jun 2022
Download
Short summary
A lack of inventory data can be a limiting factor in developing landslide predictive models, which are crucial for supporting hazard policy and decision making. We show how case-based reasoning and domain adaptation, transfer learning techniques, can effectively retrieve similar landslide modelling situations for predicting in new areas that are data scarce. Using cases in Italy, Austria and Ecuador, our findings support applying transfer learning for areas requiring rapid model development.