Hong Kong Landslide Susceptibility Mapping in a Meta-learning Way (tf2).

Table of Contents

• Background
• Data
• Dependencies
• Usage
• Contact
• Citation

Background

Landslide susceptibility assessment (LSA) is vital for landslide hazard mitigation and prevention. Recently, there have been vast applications of data-driven LSA methods owing to the increased availability of high-quality satellite data and landslide statistics. However, two issues remain to be addressed, as follows: (a) landslide records obtained from a landslide inventory (LI) are mainly based on the interpretation of optical images and site investigation, resulting in current datadriven models being insensitive to slope dynamics, such as slow-moving landslides; (b) Most study areas contain a variety of landslide-inducing environments (LIEs) that a single model can not accommodate well. In this study, we proposed the utilization of InSAR techniques to sample weak landslide labels from slow-moving slopes for LI augmentation; and meta-learn intermediate representations for the fast adaptation of LSA models corresponding to different LIEs. We performed feature permutation to identify dominant landslide-inducing factors (LIFs) and fostered guidance for targeted landslide prevention schemes. The results obtained in Hong Kong revealed that deformation in several mountainous regions are closely associated with the majority of recorded landslides. By augmenting the LI using InSAR techniques, the proposed method improved the perception of potential dynamic landslides and achieved better statistical performance. The discussion highlights that slope and stream power index (SPI) are the key LIFs in Hong Kong, but the dominant LIFs will vary under different LIEs. By comparison, the proposed method entails a fast-learning strategy and extensively outperforms other data-driven LSA techniques, e.g., by 3-6% in accuracy, 2-6% in precision, 1-2% in recall, 3-5% in F1-score, and approximately 10% in Cohen Kappa.

​ Fig. 1: Overflow

Data

• The landslide inventory can be found here.
• The related thematic information can be found here.
• The nonlandslide/landslide sample vectors are filed into ./src_data/ where samples_HK.csv and samples_HK_noTS.csv are datasets with and without augmented slow-moving landslides, respectively.

Dependencies

The default branch uses tf2 environment:

• cudatoolkit 11.2.2
• cudnn 8.1.0.77
• python 3.9.13
• tensorflow 2.10.0

Install required packages

python -m pip install -r requirements.txt

Usage

• For the scene segmentation and task sampling stage, see ./scene_sampling.py, the result would be output into ./metatask_sampling folder.
• For the meta learner, see ./meta_learner.py.
• For the model adaption and landslide susceptibility prediction, see ./predict_LSM.py. The intermediate model and adapted models of blocks would be saved in folder ./checkpoint_dir and ./models_of_blocks, respectively.The adapted models will predict the susceptibility for each sample vector in ./src_data/grid_samples_HK.csv.
• The ./tmp folder restores some temp records.
• For the figuring in the experiment, see ./figure.py, the figures would be saved in folder ./figs.

Contact

To ask questions or report issues, please open an issue on the issue tracker.

Citation

If this repository helps your research, please cite the paper. Here is the BibTeX entry:

@article{CHEN2023107342,
title = {Landslide susceptibility assessment in multiple urban slope settings with a landslide inventory augmented by InSAR techniques},
journal = {Engineering Geology},
volume = {327},
pages = {107342},
year = {2023},
issn = {0013-7952},
doi = {https://doi.org/10.1016/j.enggeo.2023.107342},
author = {Li Chen and Peifeng Ma and Chang Yu and Yi Zheng and Qing Zhu and Yulin Ding},
}

The preprint can be found: here