Deep Self-Supervised Disturbance Mapping with the OPERA Sentinel-1 Radiometric Terrain Corrected SAR Backscatter Product

TL;DR

This work addresses the challenge of disturbance mapping from SAR data at near-global scales without labeled data. It introduces a self-supervised vision transformer trained on OPERA RTC-S1 imagery to estimate a per-pixel distribution and uses a Mahalanobis-distance-based disturbance metric to delineate changes from a baseline time-series. Across three diverse disasters (PNG landslide, Chile wildfires, and Bangladesh floods), the transformer consistently outperforms a prior RNN approach and the classical log-ratio method, achieving PR AUCs above 0.65 and F1 scores above 0.6. The approach demonstrates strong potential for operational, near-global disturbance monitoring using analysis-ready RTC-S1 data, with broad implications for disaster response and environmental monitoring.

Abstract

Mapping land surface disturbances supports disaster response, resource and ecosystem management, and climate adaptation efforts. Synthetic aperture radar (SAR) is an invaluable tool for disturbance mapping, providing consistent time-series images of the ground regardless of weather or illumination conditions. Despite SAR's potential for disturbance mapping, processing SAR data to an analysis-ready format requires expertise and significant compute resources, particularly for large-scale global analysis. In October 2023, NASA's Observational Products for End-Users from Remote Sensing Analysis (OPERA) project released the near-global Radiometric Terrain Corrected SAR backscatter from Sentinel-1 (RTC-S1) dataset, providing publicly available, analysis-ready SAR imagery. In this work, we utilize this new dataset to systematically analyze land surface disturbances. As labeling SAR data is often prohibitively time-consuming, we train a self-supervised vision transformer - which requires no labels to train - on OPERA RTC-S1 data to estimate a per-pixel distribution from the set of baseline imagery and assess disturbances when there is significant deviation from the modeled distribution. To test our model's capability and generality, we evaluate three different natural disasters - which represent high-intensity, abrupt disturbances - from three different regions of the world. Across events, our approach yields high quality delineations: F1 scores exceeding 0.6 and Areas Under the Precision-Recall Curve exceeding 0.65, consistently outperforming existing SAR disturbance methods. Our findings suggest that a self-supervised vision transformer is well-suited for global disturbance mapping and can be a valuable tool for operational, near-global disturbance monitoring, particularly when labeled data does not exist.

Figures (15)

• Figure 1: Visualization of the disturbance metric and how it is computed for single channel (VV) time-series. This example shows data from a landslide that occurred in Enga Province, Papua New Guinea on May 24, 2024. We use a sequence of baseline imagery $x_1, x_2, ... x_T$ (left) from before the event to predict a distribution for each pixel in the image with mean $\mu_{T+1}$ and standard deviation $\sigma_{T+1}$ (center right). These parameters are used to quantify disturbance in the observed image $x_{T+1}$ (center right, bottom) to compute one-dimensional transformer metric via the Mahalanobis distance (right), where higher values (measured in standard deviations or SD) indicate a higher likelihood of disturbance. Our methodology is able to precisely delineate the landslide disturbance extents (bottom right corner of the Disturbance Map).
• Figure 2: A flow chart illustrating the transformer metric proposed in this paper. The metric is computed for dual polarization images (see Fig. \ref{['fig:schematic']} for 1-dimensional visualization). The first arrow is where the deep-learning model is utilized to estimate the per-pixel distribution of the $x_{T+1}$.
• Figure 3: PlanetScope (optical) imagery planet_data of before and after the landslide in Papua New Guinea. The red polygon shows the extent of the landslide (mapped manually by the authors).
• Figure 4: Landsat 8 and 9 false color imagery (bands 6, 5, 3) of before and after the fires in the Valparaíso region of Chile. The above imagery is taken from NASA Earth Observatory nasa_inferno_scars_valparaiso.
• Figure 5: Left: Pre-landslide SAR imagery (VV), May 22 2024. Middle: Post-landslide SAR imagery (VV), June 3 2024. The landslide is seen in the bottom right corner. Right: Target (ground truth) damage map overlaid on ESRI Composite World Imagery esri_world_imagery. The damaged area is shown in yellow. Note that the ESRI imagery is a composite and thus does not capture the recent landslide, and is used as a reference only.