Talaix computes climate-extreme intelligence from documented datasets and published methods. Every component says what it rests on — and what it does not know.
Only sources that are actually integrated are listed as in use. Every result carries its source, freshness and limitations.
| Source | What it provides | How Talaix uses it | Freshness | Resolution | Kind |
|---|---|---|---|---|---|
| Copernicus Sentinel-2 via Earth Search STAC |
Real optical imagery: NDVI / NDMI / NDWI vegetation & moisture indices | Fuel-moisture estimation, vegetation condition | Latest cloud-free scene (days; shown per result) | 10 m, 5-day revisit | Observed |
| Copernicus C3S ERA5 via Open-Meteo archive |
Historical daily weather reanalysis | Climatologies, hazard history, "What changed?", validation features | Daily, lag of several days | ~11–25 km, daily | Modelled (reanalysis) |
| Open-Meteo forecast | Current & forecast weather, soil moisture, daily fire-weather series | Canadian FWI computation, hazard screening inputs | Hourly model output | ~11 km grid | Modelled / Forecast |
| GloFAS river discharge Copernicus EMS/JRC, via Open-Meteo Flood API |
Modelled river discharge, daily | Flood screening: discharge vs climatology | Daily nowcast | river network, from 1984 | Modelled |
| Ocean waves (ECMWF WAM) via Open-Meteo Marine API |
Wave height, period and direction | Coastal exposure screening | Nowcast + short forecast | marine grid | Modelled |
| EU-DEM / SRTM Copernicus/EEA & NASA, via OpenTopoData |
Elevation, slope, aspect | Terrain contribution to risk, coastal low-elevation screening | Static | 25 m (Europe) / 90 m | Observed |
| ESA WorldCover | Global land-cover classes | Burnability, built-up and cropland exposure | Yearly product | 10 m | Observed |
| NASA FIRMS requires free API key |
Active-fire detections (VIIRS/MODIS) | Observed fire events for validation & history — marked unavailable without a key | Near-real-time | 375 m | Observed |
| WorldPop | Gridded population estimates | Population exposure per hazard class | Yearly vintage | 100 m grid | Modelled estimate |
| OpenStreetMap via ohsome API; Overpass fallback |
Buildings, roads, critical facilities, land use | Exposure counting — with completeness caveats | Continuously updated | feature-level | Reported (crowdsourced) |
| Canadian FWI System Van Wagner 1987 |
Fire-weather danger indices (FFMC/DMC/DC/ISI/BUI/FWI) | Anchor of the wildfire screening score | Computed daily from the sources above | daily | Derived |
These official datasets are evaluated for integration but are not currently used — Talaix does not claim their capabilities:
Every analysis component is labelled by claim status (observed / documented / reported / modelled / inferred / unknown) and temporal class (observed / historical / forecast / modelled / projected / scenario) with acquisition time, resolution and limitations.
Talaix computes screening-level indicators from real data. We do not claim to predict disasters: levels are labelled as screening indicators until validated, and validation status is shown in every report.
Hazard screening uses documented systems and reanalysis climatologies — the Canadian FWI system for fire danger, percentile-versus- climatology screening for heat, wind, discharge and deficit.
Analyses are recorded as content-hashed runs, verifiable against the same underlying bytes; benchmark and ground-truth registries track what has — and has not — been validated.
ML components (fuel-moisture estimation, ignition-likelihood indicator) are research-track components: their outputs are labelled as modelled/inferred indicators, never as observed fact.
What-if tools (e.g. smoke-transport scenarios, protection planning) compute declared scenarios from stated inputs — always labelled SCENARIO, never mixed with observations.
The platform is a containerized, tested system: an offline test suite gates every deploy, the public API is documented, and the source and model registries are public.
Explore the roadmap, the source registry, or get in touch to discuss partnerships.