Methods & data
How ClimMate works
ClimMate is a physical climate risk framework built on the same four modules as the Oasis Loss Modelling Framework. The hazard module is new: a multi-model ensemble of machine-learning downscaled CMIP6 projections for Australasia.
- Exposure
Where and what
An OED location file. Each location is keyed to one 0.25° grid cell, the
areaperil. - Hazard
How severe, how often
Annual indices from every model chain; extreme-value (GEV) curves for acute hazards.
- Vulnerability
How much damage
Damage ratio as a function of hazard intensity, by OED occupancy and construction.
- Financial
Who pays
Site deductible and limit (
LocDed6All,LocLimit6All) applied to ground-up loss. - Results
Loss metrics
Average annual loss and exceedance-probability curves, with the spread across the ensemble.
The hazard layer is pre-computed once; everything else runs in the browser, so the site is a static website with no server-side compute.
The climate data
Daily near-surface temperature, precipitation and 10 m wind from CMIP6 global climate models (GCMs), downscaled to 0.25° over CORDEX region 9 (Australasia, grid AUS-25i, 264 × 480 cells, 12°N–53.75°S, 88°E–152.25°W). All four emulators were trained once on ERA5 (1985–2013, validated on 2014); nothing is tuned per GCM or per scenario.
| Downscaling model | Architecture |
|---|
Coverage: driving GCM × experiment
Indicators
Chronic indicators are 20-year means of an annual index. Acute indicators get a generalised extreme value (GEV) distribution fitted by L-moments to the 20 annual maxima of each model chain, so you can read any return period (for example the 1-in-100-year wettest day).
| Hazard | Indicator | Units | Type | Definition |
|---|
How uncertainty is shown
A model chain is one GCM run downscaled by one model. Every number on the site is a statistic across the chains available for that scenario and horizon: the median, with the 10th–90th percentile range. Location views show every chain, so you can see whether the spread comes from the GCMs or from the downscaling.
Change is computed for each chain against that chain’s own 1995–2014 baseline before taking percentiles (the delta method). GCMs carry biases into the downscaled fields, so the change signal is more robust than absolute future values.
Horizons are 20-year windows: baseline 1995–2014, 2030 (2021–2040), 2050 (2041–2060) and 2090 (2081–2100). The SSP5-8.5 runs end in 2070.
Limitations
- Wind is daily-mean, not gusts. Speeds are computed from daily-mean wind components, so they understate both mean speed and peak gusts. Tropical cyclones are not resolved at this scale.
- Flood is a rainfall proxy. Pluvial (surface-water) damage is driven by how far extreme daily rainfall exceeds each chain’s baseline 1-in-20-year design level. There is no hydrology, river routing or terrain.
- Heat uses daily mean temperature. Maximum and minimum temperature and humidity are not yet available, so there is no wet-bulb, heatwave (EHF) or fire-weather index.
- Vulnerability functions are illustrative. They show how the framework works and must be replaced with calibrated curves before any decision use.
- Portfolio aggregation is an upper bound. Losses at all sites are summed at the same return period (full dependence). Event-based aggregation is on the roadmap.
- No sea-level rise, coastal inundation or riverine flood.
Roadmap
- Event-based Oasis model files (
footprint,vulnerability,occurrence) built from the daily fields, so the fulloasislmftoolchain can run on the same hazard. - Bias adjustment against ERA5 and AGCD, and global-warming-level horizons (1.5, 2 and 3 °C).
- New downscaled variables: daily maximum and minimum temperature, humidity and wind gust, which unlock heatwaves, wet-bulb heat stress and fire weather.
- Drought indices (SPI/SPEI) from monthly totals.
Data access
The site reads static files in the climmate-web/0.1 format: one small binary grid per map layer and one block of model-chain values per 16 × 16-cell tile. The same files can be used directly from Python or JavaScript. Current dataset: ….