You cannot manage what you cannot see. ZRC SAZU built the project's eyes: a 30-year satellite memory of every significant wildfire, and centimetre-scale drone surveys that track how the land heals after the flames — feeding straight into the live platform.
From 682 cross-border wildfire records (Miren-Kostanjevica, Komen and Devin-Nabrežina), ZRC SAZU kept the 238 larger than half a ha and reconstructed 171 burned area perimeters using burn indices (dBAIS2, dBAI) computed on Google Earth Engine — turning Sentinel-2 and Landsat imagery into a comparable, cross-border wildfire history.
That historical archive trains the daily risk engine: a Random Forest ignition model (1995–2024 data, AUC 0.854) scores wildfire probability across roughly 40,000 grid points every day, modulated by the Fire Weather Index into a clear red / yellow / green map.
Where satellites see the region, drones see the plot. Four seasonal surveys turn the pilot areas into high resolution, repeatable evidence of whether restoration actually works.
Green, red, red-edge and near-infrared bands, positioned to centimetre accuracy.
Surveys on 27 May, 16 July, 7 November 2025 and 13 February 2026 over Nova vas (6.99 ha) and Kremenjak (~28.6 ha across three flights), at 3.65–4.02 cm per pixel.
The vegetation index, mapped season by season, shows exactly where cleared and grazed ground is greening back — measuring whether the pilots are working.
Hundreds of photos per cycle become 3D point clouds, surface models and orthophotos, archived openly for re-use.
Infordata integrates ZRC SAZU's burned-area polygons, drone orthophotos and NDVI layers into the Karst Firewall platform as interactive map layers, alongside Copernicus Sentinel-2 base imagery and active-fire feeds — so a 30-year archive and last week's drone flight sit on the same screen as today's risk map, ready for first responders on both sides of the border.
Source: deliverables D2.3.1 "Remote-sensing database" and D2.3.2 "Near-real-time pilot-area maps" (WP2), ZRC SAZU with Infordata, 2026.