Thirty-year ensemble mean, with the spread between models

In short

C3S publishes, for each cell of a 0.25° grid, the annual value of each climate indicator, separately for every member of a model ensemble. A single year is not a fact about a place: in the cell containing 40.915715, 14.792024, under RCP8.5, heatwave days run 44 in 2080, 25 in 2081, 3 in 2082 and 51 in 2083. This method reduces those series to a number that can be read: the mean change expected between two thirty-year periods, computed separately for each model and then averaged across models, together with the range the models span.

The steps

1. The cells covering Italy are cut from the European grid C3S publishes.
2. For each of the nine ensemble members, the annual values are averaged over the 1981-2010 reference period, read from the same file as the future. This is not a convenience: the past values differ between one scenario’s file and the other’s even for the same model, and differencing within a single file cancels that difference.
3. For each member, the annual values are averaged over the thirty years 2036-2065.
4. The two are subtracted: the difference is that member’s anomaly, in the indicator’s own unit.
5. Across the nine members, the mean of the anomalies is taken — the value reported — together with their minimum and maximum, which are the range reported beside it.
6. The reference climate shown to the reader comes not from the models but from the ERA5 reanalysis, which is observed: a modelled reference would depend on the scenario, and a report showing two would print two different pasts.

What this number is not

• It is a scenario, not a forecast. It describes how the climate would change if emissions followed a given pathway; pathways are assumptions, which is why at least two are always shown.
• It is the value of a cell about 21 × 28 km across, not of the building. Nothing in these data distinguishes a hillside from a valley floor.
• The range is the spread between models, not a measurement error. Where it crosses zero the models do not agree on the sign of the change, and no change is established: this is frequently the case for precipitation.
• There is no overall score, and none will be produced. Indicators with different units have no common scale, and inventing one would be a TerraOverview claim and not a C3S one.

Where the code is

• pipeline/python/fetch_ecde.py
• pipeline/python/normalize_ecde.py
• packages/sources/src/adapters/c3s-ecde-climate.ts