This is a code-base is used to the generate coregistered datasets for DIST-S1 calibration and validation.
The actual datasets are derived from publicaly available datasets (see Datasets below).
We utilize geojsons in the external_validation_data_db
, each of which is curated from publicly available datasets.
The provenance of these datasets is included in the properties and the associated event yml
in the events/
directory.
This is still very much a work in progress and more information about its use and application will be added as it is refined.
Intall the environment and notebook kernel:
mamba env update -f environment.yml
conda activate dist-s1
python -m ipykernel install --user --name dist-s1
Examples:
python run_events.py --event all
python run_events.py --event 'benghazi_flood_2023 chiapas_fire_2024'
python run_events.py --event all --exclude_event 'bangladesh_coastal_flood_2024 yajiang_fire_2024'
The datasets should be generated in an out
directory. The total size currently is about 60 GB of data for all the possible events.
We use the following sources for generating these datasets.
- The Copernicus Emergency Management Service, specifically the rapid mapping of these events: https://rapidmapping.emergency.copernicus.eu/
- The UNOSAT data available through humanitarian data exchange: https://data.humdata.org/ (search "flood extents" for example!)
- The Wildland Fire Interagency Geospatial Services from the National Interagency Fire Center: https://data-nifc.opendata.arcgis.com/datasets/nifc::wfigs-current-interagency-fire-perimeters/about
- Hand drawn delineations
There will be additional sources used to derive forthecoming sites. For now, all the sites in this repository (i.e. in events/
) are derived from the above 3 sources. We note that all the datasets are mostly delineated using optical sensors (either Sentinel-2, Landsat, planet or other VHR sensors and the exact provenance of each dataset can be traced using the source data). There are few that use Sentinel-1 SAR sensor that we will use for disturbance mapping (e.g. demak_flood_2024
). Generally, optically-derived delineations are valuable datasets for calibrating/validating SAR disturbances as some aspects of the event will be visible in one sensor but not the other and vice versa. We highlight that validating any imagery across sensors can be impacted by the differences in acquisition time, particularly when imaging dynamic events like floods. In other words, care must be used for each event!