| 29 Sep 2026
A spatially and temporally disaggregated inland flood dataset with flood metrics (2000-2024)
Abstract. Floods are some of the most widespread and expensive natural disasters globally, yet current flood research suffers from a lack of detailed historical data on flood events and their impacts at fine spatial and temporal scales. We created a spatially and temporally disaggregated global flood dataset over 2000-2024 by integrating flood disaster records from EM-DAT with satellite-based flood detection and gridded population data. Starting with 4,073 inland flood events, we disaggregate each event to the admin1-month scale (first-level administrative regions by calendar month), generate satellite-derived flood maps using MODIS surface reflectance imagery, and combine these maps with population density data to calculate direct flood-exposed population. The resulting dataset contains 23,334 admin1-month flood records across 2,375 unique administrative regions in 177 countries, with satellite-derived metrics including flooded area and flooded population. Flooded population estimates based on the satellite flood maps are correlated with reported impact metrics, validating our approach despite known limitations in the flooded pixel detection algorithm.
The manuscript presents a comprehensive dataset of disaggregated inland flood events from 2000 to 2024. The dataset was created by combining event records from EM-DAT with flooded pixel detection data from MODIS and gridded population information. This resource fills a significant gap in the current body of knowledge on flood impacts and has the potential to be valuable to the scientific community. However, there are several areas that require further attention and improvement. They are presented below:
Major comments:
1. Section 2.1: A summary table of all input datasets is required. The manuscript currently uses many heterogeneous datasets, and the information about these datasets is presented across Section 2.1. This makes it difficult to follow the workflow and reproduce the results. Therefore, I recommend adding a table.
2. Lines 160–170: Please add an illustration showing how water is detected in the imagery. The flooded-pixel algorithm is the core of the dataset, but the reader cannot see what it actually does. Please include a figure with one or more examples.
3. Lines 220-225: The explanation for the low correlation requires support from the literature. The two reasons given are currently asserted without evidence. Please provide citations for any validation work that compares impact reports with satellite-derived flood exposure data. This also applies to other assumptions in the results.
4. Section 4, lines 283–303: There is an effective absence of discussion, and it should be significantly expanded. Please compare this dataset with similar existing products, and clearly state what this dataset contributes. It would also be helpful to speculate on how the product could be improved in future work, summarize the main sources of uncertainty.
Minor comments:
1. The number of admin1-month records is inconsistent: 23,334 (abstract) vs 23,432 (Sect. 2.2.5) vs 23,112 (Sect. 3.2).
2. Line 150: duplicated phrase: "We assign a unique ID to each of the admin1-month events a unique ID.."