BEYOND’s Floods Monitoring Service FloodHub monitors all the flood events in selected river basins and publishes the flood mapping results produced following the processing of Sentinel-1 images from the Hellenic National Sentinel Data Mirror Site (the first fully automated process).

Data Retrieval:

The Hellenic SENTINEL Data HUB continuously monitors the ESA Collaborative Data HUB via a dedicated Application Programming Interface (API).
Each time a SENTINEL acquisition is available (Level-1, SLC or GRD) the Hellenic SENTINEL Data HUB extracts and stores the relevant metadata as well as the original acquisition data first at a short-term and finally at a local (NOA premises) long-term storage archive (100 TB volume).

Floods Monitoring:

The floods monitoring application consists of a number of processing modules: Binary Flood Mask extraction [A], Floods classification algorithm [C], Floods post-processing algorithm [D], Diachronic Overlay analysis algorithm [E], as well as a number of input data layers.

Pre-Processing:

The Binary Flood Mask extraction algorithm [A] takes as input a high resolution Digital Elevation Model (DEM) for each AOI and exports a Binary Flood Mask. This Mask is derived after applying thresholding and noise reduction techniques to the Flood Prone Areas Index (MTI) Map which is calculated from the input DEM dataset and provides indications of flood proneness. This Mask is computed once for each AOI and is used by the Floods post-processing algorithm [D] in order to validate (clip) pixels classified as flooded by the floods classification algorithm [C].
The Diachronic Overlay analysis algorithm [E] preforms a diachronic analysis over a large number of SENTINEL acquisitions (a number of hydrological years), classified by the flood classification algorithm [C] as flooded/non-flooded pixels. Spatially consistent areas where their pixels are classified as flooded for at least 85% of all cases are further characterized as permanent water and as a final result the Permanent Water Mask is calculated. This Mask is computed once for each AOI and is also used by the Floods post-processing algorithm [D] in order to eliminate (erase) pixels classified as flooded by the floods classification algorithm [C] that actually represent areas with permanent water (lakes, rivers etc.).

Real-time:

Real-time floods monitoring algorithm periodically queries the Hellenic SENTINEL Data HUB for new satellite images (Level-1, SLC or GRD) over the predefined AOIs (Areas of Interest). Each time there is a new acquisition the Flood classification algorithm [C] applies a non-supervised classification and a product is derived which encodes pixels as flooded/non flooded. The classification product is further post-processed [D] in order to apply the “Clip Mask” (Binary Flood Mask) and the “Erase mask” (Permanent Water Mask). The final product which spatially defines temporarily flooded areas is vectorized and stored into a relational database schema.