Optical Remote Sensing for Global Flood Disaster Mapping: A Critical Review Towards Operational Readiness

Flood hazards and their disastrous consequences disrupt economic activity and threaten human lives globally. From a remote sensing perspective, since floods are often triggered by extreme climatic events, such as heavy rainstorms or tropical cyclones, the efficacy of using optical remote sensing data for disaster and damage mapping is significantly compromised. In many flood events, obtaining cloud-free images covering the affected area remains challenging. Nonetheless, considering that floods are the most frequent type of natural disaster on Earth, optical remote sensing data should be fully exploited. In this article, firstly, we will present a critical review of remote sensing data and machine learning methods for global flood-induced damage detection and mapping. We will primarily consider two types of remote sensing data: moderate-resolution multi-spectral data and high-resolution true-color or panchromatic data. Big and semantic databases available for advanced machine learning to date will be introduced. We will develop a set of best-use case scenarios for using these two data types to conduct water-body and built-up area mapping with no to moderate cloud coverage. We will cross-verify traditional machine learning and current deep learning methods and provide both benchmark databases and algorithms for the research community. Last, with this suite of data and algorithms, we will demonstrate the development of a cloud-computing-supported computing gateway, which houses the services of both our remote-sensing-based machine learning engine and a web-based user interface. Under this gateway, optical satellite data will be retrieved based on a global flood alerting system. Near-real-time pre- and post-event flood analytics are then showcased for end-user decision-making, providing insights such as the extent of severely flooded areas, an estimated number of affected buildings, and spatial trends of damage. In summary, this paper’s novel contributions include (1) a critical synthesis of operational readiness in flood mapping, (2) a multi-sensor-aware review of optical limitations, (3) the deployment of a lightweight ML pipeline for near-real-time mapping, and (4) a proposal of the GloFIM platform for field-level disaster support.