How Climate Future Viewer builds the data

1. Why this exists

Most people never read an IPCC report, but everyone has an address. We aggregate the same CMIP6 multi-model ensemble that drives those reports, project the year-2050 result onto a city-scale grid cell, and present it next to a real historical baseline. The goal is to make the climate future legible at the human scale of "my home".

2. Future projection — Copernicus C3S CMIP6 ensemble

We use the Copernicus Climate Data Store dataset sis-extreme-indices-cmip6. This bundles bias-adjusted climate-extreme indices computed by the ETCCDI from 24 CMIP6 global climate models, each running the SSP2-4.5 scenario (a moderate-emissions pathway broadly consistent with current policies). For every model we take the txxETCCDI variable — the annual maximum of daily maximum temperature, denoted TXx — at horizontal resolution near 2.8° (~280 km), realization r1i1p1f1.

3. Historical baseline — NASA POWER (MERRA-2 reanalysis)

Climate models do not predict the past, they re-simulate it, so we anchor every city to a real historical reference. NASA POWER serves daily 2-m maximum temperature derived from the MERRA-2 reanalysis, which assimilates global surface observations onto a regular grid. For each city we pull 1981–2010, compute the annual maximum (TXx) for each year, and average over the 30-year window. That value is our "today" baseline.

4. Ensemble band — what the spread means

The 24 CMIP6 models do not agree perfectly. We report three numbers from the ensemble: the median (our headline 2050 value) and the 10th and 90th percentiles (the p10–p90 band). When the band is narrow the models broadly agree; when it is wide there is real scientific uncertainty about how hot 2050 will be at that location. Always read the band, not just the median.

5. Horizon — what "2050" really means

Climate output for a single year is noisy. "2050" in this site is shorthand for the 20-year average of TXx across 2041–2060. Averaging this window cancels short-term internal variability and isolates the forced climate signal. SSP2-4.5 is the only scenario shown in v1; we plan to add SSP1-2.6 (low) and SSP5-8.5 (high) for explicit comparison.

6. Limitations — be honest with the reader

• Grid resolution. The ~280-km CMIP6 cell sometimes maps a coastal city (Tokyo, Shanghai, Mumbai) onto a cell dominated by ocean, biasing the projected TXx low. Phase 2 will move these cities to CORDEX regional downscaling at ~25 km.
• Single scenario. Only SSP2-4.5 is displayed. Real climate planning needs to span SSP1-2.6 to SSP5-8.5 — that comparison is on the v2 roadmap.
• Single realization. We use r1i1p1f1 per model. Multi-realization averaging would shrink the band and is planned for v2.
• One metric. TXx is the annual hottest single day. It does not tell you about heat-day count, humidity-adjusted heat (WBGT), nighttime minima, or sequential heatwaves. Those are tracked as separate phase-2 metrics.

7. Attribution

• Climate projections derived from Copernicus C3S CMIP6 ensemble (sis-extreme-indices-cmip6).
• Historical baseline from NASA POWER (MERRA-2 reanalysis).
• Numbers shown on this site are scenario projections, not predictions, and should not be cited as forecasts.

8. License

Copernicus C3S data are released under the Copernicus open license. NASA POWER data are public domain. The derivative numbers we display (per-city ensemble medians, percentile bands, deltas) are our pipeline output — you may cite or screenshot them under fair use, with attribution to Climate Future Viewer by Volt AI.

9. Methodology version

Currently v1 (released 2026-05-08). Any change that moves a numeric value by more than 0.1 °C will bump the version and add an entry to the changelog so previously shared screenshots remain interpretable.