Sentinel-1 Flood Maps Using A Topographic Index As Prior In Bayesian Inference
The TUWien flood mapping algorithm is a Sentinel-1 based workflow using Bayes Inference at the pixel level. However, priors in its formulation have, so far, been reduced to non-informative priors (50%-50% probability). We proposed and tested a Height Above Nearest Drainage (HAND) index based prior probability function compared to the baseline non-informed case. We have conducted experiments on six study sites for both flooded and no-flood scenarios. Full description and discussion is found in the paper: Improving Sentinel-1 Flood Maps Using A Topographic Index As Prior In Bayesian Inference.
- We propose an exponential function defined by a midpoint and steepness parameter for the HAND prior function.
- We determine optimal parameterization for the proposed function by iterating the midpoint (5 - 40) values and steepness (5 - 40).
- Each flood map is compared with reference CEMS Rapid Mapping reference dataset-- generating validation/confusion maps.
- Flood maps were generated using Bayes Inference based SAR Flood mapping algorithm implemented in python using Yeoda software package.
- Datasets are stored in GeoTiff format using LZW Compression
- Files are compressed per dataset/map product
- Files are orgnized and tiled following T3 Equi7Grid tilling system at 20m x 20m resolution.
- Folder structure: dataset/map product>(continental)subgrid>tile>files.
- Files are named follows the Yeoda filenaming convention.
- Flood - flood maps generated using different parameterization of HAND prior function and non-informed priors.
- No Flood - maps generated using different parameterization of HAND prior function and non-informed priors at no flood scenarios.
- Validation - confusion maps generated from the difference of the Flood maps generated and rasterized CEMS Rapid Mapping reference dataset.
- HAND - corresponding Height Above Nearest Drainage dataset used in the flood map generation.