{"title":"Volumetric estimation of landslide-induced terrain change using conditional GAN with multi-temporal DEMs and satellite imagery","authors":"Yu-En Yang, Teng-To Yu","doi":"10.1016/j.jag.2025.104864","DOIUrl":null,"url":null,"abstract":"<div><div>Estimating landslide volumes remains challenging because of the limited availability of high-resolution terrain data. Traditional methods often rely on empirical area-to-volume conversion formulas, which introduce significant uncertainties due to terrain variability and simplified assumptions. This study proposes a deep learning framework that integrates multitemporal satellite imagery, LiDAR-derived Digital Elevation Models (DEMs), terrain attributes, and a conditional generative adversarial network (cGAN) to simulate the DEM of Difference (DoD) and estimate landslide-induced volumetric changes. The input DEMs were resampled to 20-meter resolution for publicly available coverage exceeding 78,183 km<sup>2</sup>. A total of 2,881 map frames contained two DEM epochs, and 348 frames had three epochs, with time intervals typically ranging from 5 to 8 years. From these, 198 deep-seated landslide cases were extracted for analysis, covering approximately 7.12 km<sup>2</sup>, including 4.89 km<sup>2</sup> of erosion area and 2.23 km<sup>2</sup> of deposition area. The proposed model achieved an overall classification accuracy of 0.66, with F1-scores of 0.46 for erosion, 0.30 for deposition, and 0.78 for background. Volumetric estimations revealed a consistent underestimation trend, with median erosion errors of approximately − 50 % and deposition errors approaching − 100 % across the five-fold cross-validation. The framework effectively captures three-dimensional spatial distributions and enables accurate volumetric estimation without the need for post-event DEMs, offering a practical solution for data-scarce regions. Additionally, it enhances sediment volume assessments that are crucial for disaster prevention and sediment management, bridging the gap between empirical estimation and modern deep learning techniques.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"144 ","pages":"Article 104864"},"PeriodicalIF":8.6000,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"International journal of applied earth observation and geoinformation : ITC journal","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1569843225005114","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"REMOTE SENSING","Score":null,"Total":0}
引用次数: 0
Abstract
Estimating landslide volumes remains challenging because of the limited availability of high-resolution terrain data. Traditional methods often rely on empirical area-to-volume conversion formulas, which introduce significant uncertainties due to terrain variability and simplified assumptions. This study proposes a deep learning framework that integrates multitemporal satellite imagery, LiDAR-derived Digital Elevation Models (DEMs), terrain attributes, and a conditional generative adversarial network (cGAN) to simulate the DEM of Difference (DoD) and estimate landslide-induced volumetric changes. The input DEMs were resampled to 20-meter resolution for publicly available coverage exceeding 78,183 km2. A total of 2,881 map frames contained two DEM epochs, and 348 frames had three epochs, with time intervals typically ranging from 5 to 8 years. From these, 198 deep-seated landslide cases were extracted for analysis, covering approximately 7.12 km2, including 4.89 km2 of erosion area and 2.23 km2 of deposition area. The proposed model achieved an overall classification accuracy of 0.66, with F1-scores of 0.46 for erosion, 0.30 for deposition, and 0.78 for background. Volumetric estimations revealed a consistent underestimation trend, with median erosion errors of approximately − 50 % and deposition errors approaching − 100 % across the five-fold cross-validation. The framework effectively captures three-dimensional spatial distributions and enables accurate volumetric estimation without the need for post-event DEMs, offering a practical solution for data-scarce regions. Additionally, it enhances sediment volume assessments that are crucial for disaster prevention and sediment management, bridging the gap between empirical estimation and modern deep learning techniques.
期刊介绍:
The International Journal of Applied Earth Observation and Geoinformation publishes original papers that utilize earth observation data for natural resource and environmental inventory and management. These data primarily originate from remote sensing platforms, including satellites and aircraft, supplemented by surface and subsurface measurements. Addressing natural resources such as forests, agricultural land, soils, and water, as well as environmental concerns like biodiversity, land degradation, and hazards, the journal explores conceptual and data-driven approaches. It covers geoinformation themes like capturing, databasing, visualization, interpretation, data quality, and spatial uncertainty.