Kuro Siwo: 33 billion $m^2$ under the water. A global multi-temporal satellite dataset for rapid flood mapping

Overview

• This paper introduces a new dataset called Kuro Siwo, which is a manually annotated multi-temporal dataset spanning 43 flood events globally.
• Kuro Siwo maps more than 338 billion $m^2$ of land, with 33 billion designated as either flooded areas or permanent water bodies.
• The dataset includes both a processed product optimized for flood mapping based on SAR Ground Range Detected, and a raw SAR Single Look Complex product with minimal preprocessing.
• The paper also provides an extensive benchmark, called BlackBench, offering strong baselines for a diverse set of flood events from Europe, America, Africa, Asia, and Australia.

Plain English Explanation

Floods caused by climate change are a severe threat to human life, infrastructure, and the environment. Recent catastrophic events in Pakistan and New Zealand have highlighted the urgent need for accurate flood mapping to guide restoration efforts, understand vulnerabilities, and prepare for future occurrences.

Satellite-based Synthetic Aperture Radar (SAR) technology can provide day-and-night, all-weather imaging capabilities that are well-suited for flood mapping. However, the lack of large, annotated datasets has limited the application of deep learning techniques for this task. To address this, the researchers have created a new dataset called Kuro Siwo, which contains manually annotated flood data from 43 events around the world.

Kuro Siwo includes over 338 billion square meters of land, with 33 billion square meters identified as either flooded areas or permanent water bodies. The dataset provides both a processed product optimized for flood mapping and a raw SAR product with minimal preprocessing, allowing researchers to explore different approaches to exploiting the phase and amplitude information in the data.

To further support research in this area, the researchers have also included a large set of unlabeled SAR samples and an extensive benchmark, called BlackBench, which offers strong baselines for a diverse set of flood events from across the globe. These resources are designed to help advance the use of deep learning for flood mapping and improve our ability to respond to and prepare for the impacts of climate change.

Technical Explanation

The paper introduces a new dataset called Kuro Siwo, which is a manually annotated multi-temporal dataset spanning 43 flood events globally. The dataset maps more than 338 billion $m^2$ of land, with 33 billion designated as either flooded areas or permanent water bodies.

Kuro Siwo includes two types of SAR data products: a highly processed product optimized for flood mapping based on SAR Ground Range Detected, and a primal SAR Single Look Complex product with minimal preprocessing. The latter is designed to promote research on the exploitation of both the phase and amplitude information in the data and to offer maximum flexibility for downstream task preprocessing.

To leverage advances in large-scale self-supervised pretraining methods for remote sensing data, the researchers have augmented Kuro Siwo with a large unlabeled set of SAR samples. This allows the dataset to be used for pretraining models, which can then be fine-tuned on the annotated flood data.

In addition to the dataset, the paper also provides an extensive benchmark, namely BlackBench, offering strong baselines for a diverse set of flood events from Europe, America, Africa, Asia, and Australia. This benchmark is designed to support the development and evaluation of flood mapping algorithms, helping to drive progress in this important area of research.

Critical Analysis

The Kuro Siwo dataset and BlackBench benchmark represent a significant contribution to the field of flood mapping using SAR data. By providing a large, manually annotated dataset and a comprehensive evaluation framework, the researchers have addressed a key limitation in the existing literature, which has been hampered by the lack of high-quality training data.

However, the paper does acknowledge some potential limitations of the dataset. For example, the annotation process can be subjective, and there may be some inconsistencies or errors in the labeling. Additionally, the dataset may not capture the full range of flood characteristics and environmental conditions, as it is limited to the 43 events included in the study.

Further research could explore methods for improving the annotation process, perhaps through the use of crowdsourcing or automated techniques. Combining the Kuro Siwo dataset with other flood datasets could also help to broaden the range of flood events and environmental conditions represented, providing a more comprehensive resource for the research community.

Conclusion

The introduction of the Kuro Siwo dataset and BlackBench benchmark represents a significant step forward in the development of deep learning-based flood mapping techniques. By providing a large, high-quality dataset and a comprehensive evaluation framework, the researchers have enabled the research community to make rapid progress in this important field.

The potential applications of this work are wide-ranging, from guiding disaster response and recovery efforts to informing long-term planning and infrastructure development. As the impacts of climate change continue to intensify, accurate and reliable flood mapping will become increasingly crucial for protecting human lives and mitigating the economic and environmental damages caused by these events.