Machine learning powered high-resolution co-seismic landslide detection

Haojie Wang,Limin Zhang,Lin Wang,Ruilin Fan,Shengyang Zhou,Yejia Qiang,Ming Peng
DOI: https://doi.org/10.1016/j.gr.2022.07.004
IF: 6.1
2022-07-13
Gondwana Research
Abstract:Numerous co-seismic landslides can be triggered by a strong earthquake. Fast and accurate detection and mapping of these landslides are crucial for rapid risk assessment and humanitarian assistance. Traditional visual interpretation is time-consuming and heavily influenced by human judgement. This paper presents an integrated machine learning (ML) powered method for co-seismic landslide detection, which combines multi-source data, pixel-based and object-based treatments, and ML techniques. The proposed method first fuses multi-source data, including optical images, synthetic aperture radar images and digital elevation models, to produce data layers and reference landslide inventories for learning. After that, ML algorithms are coupled with pixel-based and object-based treatments to produce a series of co-seismic landslide detection models. Powerful models are subsequently recommended after comprehensive evaluation. Two case studies of earthquakes in China are presented. The first is in the epicenter area of the 2008 Wenchuan earthquake. Even trained with only 5% of local data, the proposed method still achieves an accuracy of area (AOA) of 97.77% in the trained area and an AOA of 93.17% in a new area. If trained with the entire local data, an AOA of 99.99% in the trained area and 94.56% in a new area can be obtained. The second case detects the co-seismic landslides induced by the 2017 Jiuzhaigou earthquake. An AOA of 96.55% is achieved with 5% of local data. The case studies confirm the outstanding performance and generic nature of the proposed machine learning method, greatly advancing the state of the art of high-resolution co-seismic landslide detection and the landslide science.
geosciences, multidisciplinary
What problem does this paper attempt to address?