Exploring the Copernicus Global Flood Monitoring system for the development of a global impact attribution and validation dataset

Global assessments of socio-economic impacts and risks from extreme weather events are often constrained by fragmented and insufficient data. The Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) provides a comprehensive framework for impact attribution and projection across sectors, including hazard indicators for extreme events.  However, for sound impact assessments, model simulations need to be properly validated against observational data. Currently, this is challenging because impact observations and records suitable for validation exercises are fragmented and often need to be collected from different sources, complicating the development of impact functions and the attribution of longer event time series. While socio-economic impacts such as fatalities, damages, and displacement are documented in global databases such as EM-DAT and the Global Internal Displacement Database (GIDD) of the Internal Displacement Monitoring Centre (IDMC), the spatial extents of affected areas are typically derived from separate remote-sensing initiatives, including the Global Flood Database (GFD) and the World Fire Atlas. To address this fragmentation, we compiled a multi-source, event-based catalogue, initially focusing on floods and tropical cyclones, with the framework designed to be extensible to additional hazards. We integrated socio-economic impact records from EM-DAT, GIDD and the Dartmouth Flood Observatory (DFO) with observational data on spatial event extents as well as simulated hazard data developed for ISIMIP. Here, we focus on exploring different data products providing spatially explicit gridded flood extents based on satellite imagery to assess their suitability for an inclusion into such an event catalogue. In particular, we test the Copernicus Global Flood Monitoring (GFM) service that leverages Sentinel‑1 SAR imagery to generate a global ~20 m ensemble flood product from three independent water detection algorithms (2015-2025). We present a comparative assessment of GFM against established datasets, such as the MODIS‑based GFD and the RAPID SAR flood mapping system and investigate the suitability of GFM for automated global risk pipelines by addressing the distinct limitations. The GFD is a valuable historical archive already linked to DFO impact records, yet its reliance on optical imagery makes it vulnerable to cloud cover and darkness. This limits effective temporal resolution, thereby increasing the likelihood that short-lived flood peaks are missed. RAPID provides high‑resolution SAR mapping, but depends on trigger‑based tasking that can miss events when hydrological or rainfall‑based triggers fail. In contrast, GFM offers systematic, near‑global, all‑weather processing of available Sentinel‑1 acquisitions. Detecting inundation in urban areas remains a persistent challenge across remote sensing products due to signal blockage—a critical limitation given that socio-economic assets are concentrated in these zones. While GFM cannot fully resolve this physical constraint, it mitigates the ambiguity by explicitly delineating structurally non-observable pixels in its Exclusion Layer, ensuring that users distinguish 'unobserved' areas from 'non-flooded' conditions. Unlike GFD or RAPID, GFM explicitly distinguishes structurally non‑observable pixels (e.g., urban areas, dense vegetation) from actual water and flags low‑confidence conditions. We showcase a workflow for spatio‑temporal matching of these footprints with reported disaster events and propose options to combine these products into a comprehensive event catalogue