Fractional water detection with VIIRS and deep learning to enhance global flood monitoring

Fractional water detection with VIIRS and deep learning to enhance global flood monitoring Please cite: TO BE COMPLETED Overview This repository contains the validation data from Saunders et al. “Fractional water detection with VIIRS and deep learning to enhance global flood monitoring”, submitted to Remote Sensing of Environment. Structure The data are presented according to the four validation and evaluation stages in the manuscript. The data for each validation stage is provided in a separate zip file, which should be downloaded and extracted prior to use. The corresponding validation results for all stages are provided in the “Results.zip” file. Description Specific data attributes are described separately below for each validation stage. General data attributes: Values: With the exception of VIIRS_NOAAGMU_Mask and files provided in the folder VIIRS_SR, all data represent fractional water extent (FWE) estimates, continuous values between 0 and 1, at nominal pixel resolution of 375 meters. File format: All data are provided in GeoTIFF format. Date format: All dates are provided in yyyymmdd format. Fractional water extent data: VIIRS_MLFWD: VIIRS Machine Learning Fractional Water Detection (ML-FWD). VIIRS_NASA: NASA NRT VIIRS Global Flood Product, 1-day composite without cloud shadow removal (Slayback, 2025), resampled from 250 m to 375 m. VIIRS_NOAAGMU: NOAA-GMU VIIRS Flood Product, 1-day (Li et al., 2018). We set areas of ‘Normal Water’ (with values of 4 in the corresponding VIIRS_NOAAGMU_Mask) to values of 1. S2_DW: Dynamic World V1, derived from Sentinel-2 imagery, label=0 (Water) (Brown et al., 2022), resampled from 10 m to 375 m. S1_PG: flood maps derived from Sentinel-1 imagery using the Paul & Ganju (2021) algorithm, resampled from 10 m to 375 m. Categorical mask data: For each example in the second to fourth validation stages, we provide a categorical mask extracted from the NOAA-GMU VIIRS Flood Product, which is can be used to filter out cloudy or permanent water pixels when calculating validation results: VIIRS_NOAAGMU_Mask: 1: Cloud; 2: Cloud Shadow; 3: Ice/Snow; 4: Normal Water; 5: Ice Water; NaN; Nodata. Surface reflectance data: Surface reflecatnce imagery from VIIRS is provided for each validation data example in the folder ‘VIIRS_SR’. We provide the six surface reflectance bands used by VIIRS ML-FWD: I bands: 375 m resolution, I1 (Red), I2 (NIR), I3 (SWIR 1). M bands: 750 m resolution, M3 (blue), M4 (green), M11 (SWIR 2). 1_Global_Water_Instances Models: ML-FWD; Dynamic World Categorical mask: None Scope: 9,701 chips; 976 grid locations Output: 1 tif file per chip (grid-date) and model Folder naming: GridID/Date/ Results file: 1_Global_Water_Instances.csv; validation metrics comparing ML-FWD to the Dynamic World reference, per chip (grid-date combination). 2_Global_Flood_Database_Events Models: ML-FWD; NASA; NOAA-GMU; Sentinel-1 Categorical mask: NOAA-GMU mask Scope: 939 chips, 100 flood events (up to 10 chips per event) from the Global Flood Database (Tellman et al., 2021) Output: 1 tif file per chip and model Folder naming: DFOEventID/ChipID_Date/ Results file: 2_Global_Flood_Database_Events.csv; validation metrics comparing ML-FWD, NASA, NOAA-GMU to the Sentinel-1 reference, per DFOEventID and chip. Three mask options were used: (A) excluding cloudy pixels; (B) excluding cloudy and permanent water pixels; (C) excluding only invalid Sentinel-1 pixels. NaN or blank results are due to no valid pixels present after the mask has been applied. 3_Selected_Flood_Events Models: ML-FWD; NASA; NOAA-GMU; Dynamic World; Sentinel-1 Categorical mask: NOAA-GMU mask Scope: 9 flood events Output: 1 tif file per flood and model Folder naming: DFOEventID_Country_Date/ Results file: 3_Selected_Flood_Events.csv; validation metrics comparing ML-FWD, NASA, NOAA-GMU to the Dynamic World and Sentinel-1 references, per DFOEventID. Two mask options were used: (A) excluding cloudy pixels; (B) excluding cloudy and permanent water pixels. VIIRS and Dynamic World data are from the same day; Sentinel-1 data may include observations from up to 2 days before and after the event date. 4_Problem_Scenarios Models: ML-FWD; NASA Categorical mask: NOAA-GMU mask Scope: 19 problem scenarios Output: 1 tif file per problem scenario and model Folder naming: ScenarioID_Country_Date/ Results file: 4_Problem_Scenarios.csv; qualitative expert rating scores for ML-FWD and NASA, per scenario and rating category (WD: water detection; CS: cloud shadow; TSV: terrain shadow and volcanic material). NaN or blank results are due to the scenario not receiving a score for that category (for example, the scenario did not contain clouds so a cloud shadow rating was not assigned). References Brown, C.F., Brumby, S.P., Guzder-Williams, B., Birch, T., Hyde, S.B., Mazzariello, J., Czerwinski, W., Pasquarella, V.J., Haertel, R., Ilyushchenko, S., Schwehr, K., Weisse, M., Stolle, F., Hanson, C., Guinan, O., Moore, R., Tait, A.M., 2022. Dynamic World, Near real-time global 10 m land use land cover mapping. Sci Data 9, 251. https://doi.org/10.1038/s41597-022-01307-4 Li, S., Sun, D., Goldberg, M.D., Sjoberg, B., Santek, D., Hoffman, J.P., DeWeese, M., Restrepo, P., Lindsey, S., Holloway, E., 2018. Automatic near real-time flood detection using Suomi-NPP/VIIRS data. Remote Sensing of Environment 204, 672–689. https://doi.org/10.1016/j.rse.2017.09.032 Paul, S., Ganju, S., 2021. Flood Segmentation on Sentinel-1 SAR Imagery with Semi-Supervised Learning. https://doi.org/10.48550/arXiv.2107.08369 Slayback, D., 2025. MODIS/VIIRS NRT Global Flood Products: User Guide Revision E. Tellman, B., Sullivan, J.A., Kuhn, C., Kettner, A.J., Doyle, C.S., Brakenridge, G.R., Erickson, T.A., Slayback, D.A., 2021. Satellite imaging reveals increased proportion of population exposed to floods. Nature 596, 80–86. https://doi.org/10.1038/s41586-021-03695-w