Automatic extraction of flood inundation areas from SAR images: a case study of Jilin, China during the 2017 flood disaster

ABSTRACT

Flood is one of the most frequent and widespread natural hazards globally, which can cause tremendous economic damage and human casualties. As such, flood event monitoring is essential, for which Synthetic Aperture Radar (SAR), with high spatial resolution as well as all-weather and all-time capabilities, can provide high-quality data support. However, algorithms for automatic flood inundation mapping that do not require ancillary data are limited. In this study, we propose a hybrid methodology that combines automatic thresholds selection, pixel- and object-based classification, and bidirectional region growing method for extracting flood inundation areas. This is a fully automatic approach that does not require the assistance of ancillary data. Firstly, the gamma distribution is used to estimate the probability density function (PDF) of ‘open water’ and to set thresholds. Then, we introduce a two-step classification approach, applying the pixel- and object-based classifications; the former is easy to implement with low computational complexity and stable performance, whereas the latter can reduce noise pixels and is less sensitive to SAR intrinsic speckle. The two-step classification is employed to yield core flooded and non-flooded regions that are used as seeds for region growing. Furthermore, we propose a bidirectional region growing approach that grows regions for flooded and non-flooded regions simultaneously to eliminate areas of uncertainty, while minimizing under- and over-detection. We verified the proposed approach by applying it to real flood events that occurred in Jilin, China on 13 July 2017 and 20 July 2017. The experimental results demonstrate the effectiveness and reliability of the proposed approach.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Key R&D Program of China [2017YFB0502901]. Part of this work was undertaken in the framework of the HREOS project [03-Y20A10-9001-15/16]. The authors would like to thank the National Disaster Reduction Center of China for providing the test cases.