FloodBrazil-1: a multisource and multitemporal remote sensing dataset for flooding monitoring

Sentinel_2_folder_year_NORMALIZED.zip: These folders contain Sentinel-2A and 2B/MSI imagery for each year of the study area. The selected spectral bands are Green (B2), Blue (B3), Red (B4), Red Edge 4 (B8A), and SWIR 1 (B11), all at processing level 2A and normalized.   Sentinel_2_folder_year_PATCHES.zip: These folders contain 64x64 spatial patches extracted from the Sentinel-2A and 2B/MSI imagery. Each patch consists of a stacked tensor containing the full temporal sequence and all selected spectral bands.

Each patch is stored in a folder named “Location_xxxx”. For example, in the patch located in the folder “Location_0001”, the first five bands correspond to the first month of the time series, the next five bands correspond to the second month, and so forth. Each folder “Location_xxxx” contains .npy and .json files with metadata describing the patch.   Sentinel_1_folder_year_NORMALIZED.zip: These folders include Sentinel-1A/C-SAR imagery for each year of the study area, with VV polarization.

The SAR data were preprocessed using the SNAP software and normalized using min–max scaling.Sentinel_1_folder_year_PATCHES.zip: These folders contain 64x64 spatial patches extracted from the Sentinel-1A/C-SAR imagery. Each patch consists of a stacked tensor containing the full temporal sequence. Each patch is stored in a folder named “Location_xxxx”, which includes .npy and .json files with metadata describing the patch.Dem_attributes_folder_NORMALIZED.zip:  This folder includes the five attributes extracted from Copernicus DEM (GLO‑30) for the study area.

The rasters represent Slope, Flow Accumulated, Topographic Wetness Index (TWI), Valley Depth and Multiresolution Index of Valley Bottom Flatness (MRVBF). All attributes were calculated using QGIS software and normalized using min–max scaling or percentile-based normalization. Dem_attributes_folder_PATCHES.zip: These folders contain 64x64 spatial patches extracted from the attributes derived from Copernicus DEM (GLO‑30).

Each patch consists of a stack of the five attributes and is stored in a folder named “Location_xxxx”, which includes .npy and .json files with metadata describing the patch. Label_mask_folder_year.zip: These folders contain raster files representing the label masks for each year of the study area. The classes include “Permanent water bodies”, “Flooded areas”, “Partially flooded areas”, and “Others”, represented by pixels values 0, 1, 2 and 3, respectively. Label_mask_folder_year_PATCHES.zip: These folders contain 64x64 spatial patches extracted from the label masks rasters.

Each patch consists of a stacked tensor containing the full temporal sequence. Each patch is stored in a folder named “Location_xxxx”, which includes .npy and .json files with metadata describing the patch. MNDWI_folder_year: These folders contain the results of the MNDWI index.Python_notebook_folders.zip: This folder contains scripts for preprocessing the entire dataset, including extracting spatial patches from time series data, splitting the data into training, validation, and test sets, and implementing the deep learning model used to map flood-prone areas.