Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Portraying the Need for Temporal Data in Flood Detection via Sentinel-1

Xavier Bou, Thibaud Ehret, Rafael Grompone von Gioi, Jeremy Anger

TL;DR

This paper addresses the challenge of flood detection by highlighting the ill-posedness of relying on a single SAR image to separate permanent water from flood water. It extends the MMFlood dataset to a multi-date setting by adding one year of Sentinel-1 observations around each flood event and reframes flood mapping as temporal anomaly detection. An unsupervised, ViBe-inspired method is proposed: a two-stage pipeline that first segments water in each image and then uses a per-pixel background history to detect anomalous (flood) water while updating the background over time, enabling low-cost, real-time capable detection. The study reveals inconsistencies in MMFlood annotations when viewed across time and provides a practical baseline for future multi-date flood mapping research, with significant implications for improving flood response and monitoring using SAR time series.

Abstract

Identifying flood affected areas in remote sensing data is a critical problem in earth observation to analyze flood impact and drive responses. While a number of methods have been proposed in the literature, there are two main limitations in available flood detection datasets: (1) a lack of region variability is commonly observed and/or (2) they require to distinguish permanent water bodies from flooded areas from a single image, which becomes an ill-posed setup. Consequently, we extend the globally diverse MMFlood dataset to multi-date by providing one year of Sentinel-1 observations around each flood event. To our surprise, we notice that the definition of flooded pixels in MMFlood is inconsistent when observing the entire image sequence. Hence, we re-frame the flood detection task as a temporal anomaly detection problem, where anomalous water bodies are segmented from a Sentinel-1 temporal sequence. From this definition, we provide a simple method inspired by the popular video change detector ViBe, results of which quantitatively align with the SAR image time series, providing a reasonable baseline for future works.

Portraying the Need for Temporal Data in Flood Detection via Sentinel-1

TL;DR

This paper addresses the challenge of flood detection by highlighting the ill-posedness of relying on a single SAR image to separate permanent water from flood water. It extends the MMFlood dataset to a multi-date setting by adding one year of Sentinel-1 observations around each flood event and reframes flood mapping as temporal anomaly detection. An unsupervised, ViBe-inspired method is proposed: a two-stage pipeline that first segments water in each image and then uses a per-pixel background history to detect anomalous (flood) water while updating the background over time, enabling low-cost, real-time capable detection. The study reveals inconsistencies in MMFlood annotations when viewed across time and provides a practical baseline for future multi-date flood mapping research, with significant implications for improving flood response and monitoring using SAR time series.

Abstract

Identifying flood affected areas in remote sensing data is a critical problem in earth observation to analyze flood impact and drive responses. While a number of methods have been proposed in the literature, there are two main limitations in available flood detection datasets: (1) a lack of region variability is commonly observed and/or (2) they require to distinguish permanent water bodies from flooded areas from a single image, which becomes an ill-posed setup. Consequently, we extend the globally diverse MMFlood dataset to multi-date by providing one year of Sentinel-1 observations around each flood event. To our surprise, we notice that the definition of flooded pixels in MMFlood is inconsistent when observing the entire image sequence. Hence, we re-frame the flood detection task as a temporal anomaly detection problem, where anomalous water bodies are segmented from a Sentinel-1 temporal sequence. From this definition, we provide a simple method inspired by the popular video change detector ViBe, results of which quantitatively align with the SAR image time series, providing a reasonable baseline for future works.
Paper Structure (7 sections, 3 equations, 3 figures)

This paper contains 7 sections, 3 equations, 3 figures.

Table of Contents

  1. Introduction
  2. The problem with flood detection datasets
  3. Extending a single image dataset
  4. Proposed algorithm
  5. Water segmentation
  6. Detection of anomalous water bodies
  7. Conclusion

Figures (3)

  • Figure 1: Sample from the MMFlood dataset mmflood (EMSR358-0-6). The left image was acquired on 2019/05/10 (not part of MMFlood), the middle image is from 2019/05/16 during the flood event. On the right, the MMFlood label shows only a partial annotation of the flooded areas. Note that from only the middle image it is not possible to infer which are the permanent bodies, a multi-date input is essential for flood mapping.
  • Figure 2: Water segmentation process. First, a speckle filter is used to denoise the raw SAR image. Then, a threshold filter is applied to generate a binary segmentation. Lastly, only connected components of relevant size are kept.
  • Figure 3: Qualitative results of the proposed unsupervised method for MMFlood scenes EMSR358-0-6 (a) and EMSR468-0-1 (b). Four observations during the flood event are shown for each, containing two observations prior to the event and two afterwards. The input SAR images are shown on the top row of the examples, while the output flood segmentation maps are displayed on the bottom.