Article
Integrating InSAR Observables and Multiple Geological
Factors for Landslide Susceptibility Assessment
Yan-Ting Lin
1
, Yi-Keng Chen
2
, Kuo-Hsin Yang
2
, Chuin-Shan Chen
2
and Jen-Yu Han
2,
*
Citation: Lin, Y.-T.; Chen, Y.-K.;
Yang, K.-H.; Chen, C.-S.; Han, J.-Y.
Integrating InSAR Observables and
Multiple Geological Factors for
Landslide Susceptibility Assessment.
Appl. Sci. 2021, 11, 7289. https://
doi.org/10.3390/app11167289
Academic Editor: Francesco Fiorillo
Received: 24 June 2021
Accepted: 5 August 2021
Published: 8 August 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
National Center for Research on Earthquake Engineering, National Applied Research Laboratories,
Taipei 106, Taiwan; yanting@ncree.narl.org.tw
2
Department of Civil Engineering, National Taiwan University, Taipei 106, Taiwan;
ykchen0814@gmail.com (Y.-K.C.); khyang@ntu.edu.tw (K.-H.Y.); dchen@ntu.edu.tw (C.-S.C.)
* Correspondence: jyhan@ntu.edu.tw; Tel.: +886-2-33664347
Abstract:
Due to extreme weather, researchers are constantly putting their focus on prevention and
mitigation for the impact of disasters in order to reduce the loss of life and property. The disaster
associated with slope failures is among the most challenging ones due to the multiple driving
factors and complicated mechanisms between them. In this study, a modern space remote sensing
technology, InSAR, was introduced as a direct observable for the slope dynamics. The InSAR-derived
displacement fields and other in situ geological and topographical factors were integrated, and their
correlations with the landslide susceptibility were analyzed. Moreover, multiple machine learning
approaches were applied with a goal to construct an optimal model between these complicated factors
and landslide susceptibility. Two case studies were performed in the mountainous areas of Taiwan
Island and the model performance was evaluated by a confusion matrix. The numerical results
revealed that among different machine learning approaches, the Random Forest model outperformed
others, with an average accuracy higher than 80%. More importantly, the inclusion of the InSAR data
resulted in an improved model accuracy in all training approaches, which is the first to be reported
in all of the scientific literature. In other words, the proposed approach provides a novel integrated
technique that enables a highly reliable analysis of the landslide susceptibility so that subsequent
management or reinforcement can be better planned.
Keywords: landslide potential; InSAR; spatial factors; machine learning; slope unite
1. Introduction
In Asian subtropical monsoon regions, July to September is a season of strong ty-
phoons. High rainfall intensity usually causes serious landslide events in mountainous
areas [
1
]. It is necessary to predict landslide occurrence and behavior and adopt appropriate
prevention policies and methods to improve disaster relief effectiveness and reduce casual-
ties and property loss during and after disasters. Landslide prediction aims to predict the
possibility of the occurrence of landslides in a specific area; available data are commonly
used, including conditional factors and historical landslides. These data are collected
from landslide inventories and static instruments, and their values are shown in spatial
analysis [
2
]. However, traditional landslide prediction, such as mathematical evaluation
models, lacks information about the temporal probability of landslides, i.e., time-series
landslide behavior. Landslide displacement time-series data can directly reflect ground
surface deformation and stability characteristics. Therefore, they have been recently used
to develop landslide prediction models. Generally, these time-series data are collected from
one-point survey equipment, such as surface extensometers and GPS devices [
3
]. However,
field GPS surveying projects, which depend on only one or two temporarily installed
reference stations, have many disadvantages [
4
]. In practice, steadily obtaining survey
data using these single reference stations is often difficult because of poor performance or
Appl. Sci. 2021, 11, 7289. https://doi.org/10.3390/app11167289 https://www.mdpi.com/journal/applsci
