Generating multi-temporal landslide inventories through a general deep transfer learning strategy using HR EO data

Abstract

Mapping of landslides over space has seen an increasing attention and good results in the last decade. While current methods are chiefly applied to generate event-inventories, whereas multi-temporal (MT) inventories are rare, even using manual landslide mapping. Here, we present an innovative deep learning strategy which employs transfer learning that allows for the Attention Deep Supervision Multi-Scale U-Net model to be adapted for landslide detection tasks in new areas. The method also provides the flexibility of re-training a pretrained model to detect both rainfall- and earthquake-triggered landslides on new target areas. For the mapping, we used archived Planet Lab remote sensing images spanning a period between 2009 till 2021 with spatial resolution of 3–5 m to systematically generate MT landslide inventories. When we examined all cases, our approach provided an average F1 score of 0.8 indicating that we successfully identified the spatiotemporal occurrences of landslides. To examine the size distribution of mapped landslides we compared the frequency-area distributions of predicted co-seismic landslides with manually mapped products from the literature. Results showed a good match between calculated power-law exponents where the difference ranges between 0.04 and 0.21. Overall, this study showed that the proposed algorithm could be applied to large areas to generate polygon-based MT landslide inventories.

Evaluation of U-Net transfer learning model for semantic segmentation of landslides in the Colombian tropical mountain region

“…The prevalence of clouds, shadows, atmospheric noise, and image artefacts like haze makes difficult at times to accurately and regularly map landslides. [1] stated that a possible solution to the problem of cloud coverage is to complement the optical image mapping with a parallel landslide detection procedure based on SAR imagery. This was corroborated and validated in this study since the results of 7-bands image model which consider RGB data, geo-indices (slope and curvature) and VV amplitude pre-and -post landslide events, showed better performance and adjustment detecting landslides respect to ground truth landslide footprints.…”

“…In mountainous regions, natural hazards such as landslides, avalanches, floods, and debris flows can cause significant property damage and human casualties [1]. Making their accurate detection and mapping is crucial for risk management and mitigation efforts.…”

“…[2] proposed a dual-encoder U-Net for landslide detection using Sentinel-2 and Digital Elevation Model (DEM) data. [1] presented an innovative deep learning strategy employing transfer learning that allows for the Attention Deep Supervision Multi-Scale U-Net model to be adapted for landslide detection tasks. [3] tested the applicability and effectiveness of a U-Net for semantic segmentation of Landsat 8 satellite imagery to the identification of landslide scars.…”

An ensemble neural network approach for space-time landslide predictive modelling

“…In fact, automated mapping tools are increasingly ubiquitous. DL architectures are regularly used to produce landslide inventories in space and time (Bhuyan et al, 2023, Novellino et al, 2024 while discriminating between landslide types (Rana et al, 2023) and the likely trigger (Rana et al, 2021). These tools are also frequently published openly to promote the benchmarking against same procedures in a consistent manner (Amateya et al, 2021;Das et al, 2023;Rana et al, 2022), thus encouraging the adoption and enhancement of the developed approaches.…”

Assessing multi-hazard susceptibility to cryospheric hazards: lesson learnt from an Alaskan example

“…We are currently testing deep-learning architectures to map RTS and ALD occurrences within the same study area. These architectures are being trained to recognize the same inventory mapped by (Swanson, 2021) through the optical information collected by PlanetScope (e.g., Bhuyan et al, 2023) and Rapid Eye (e.g., Kearney et al, 2020) products. Such tools can unlock multi-temporal RTS and ALD inventory mapping.…”