Method and limits

← Weather

A degree-day total means nothing without its base and its convention, and one grid cell at a capital is a proxy for a country, not a measurement of it.

Two degree-day conventions, and they disagree

A degree day needs a daily temperature, and there are two answers to what that is. One convention uses the day's mean; the other uses the midpoint of the day's high and low, which is what a station-based series reports because it is all a max/min thermometer can record. Neither is more correct, and this warehouse ships both rather than picking one silently.

Over 779 complete country-years the two conventions differ by 1.6% on average and by as much as 22.2% at the extreme.

The extreme is a small-denominator effect rather than a measurement problem, and the table below shows it: the worst disagreements are all places with barely any heating season, where a few degree days either way is a large share of a small total. Restricted to country-years with a real winter — a thousand degree days or more — the widest gap is 5.1% and 97 of the country-years fall below that line.

A capital is not a country

The archive holds one grid cell per country, at its capital city. That is a coarse proxy for national heating demand and the model says so with a number rather than a caveat.

The temperatures come from ERA5-Land, which answers on a 0.1-degree grid, so a request snaps to the nearest cell centre and the response reports where it actually landed. Averaged over the capitals that displacement is 3.7 km, and the furthest capital sits 7.5 km from the cell that answered for it.

That displacement is small and it is not the approximation that matters. The approximation that matters has no number here at all: Madrid's climate is not Spain's, and one cell stands in for a whole national population. Comparing a country against itself across years — which is what every degree-day column on the weather section is used for — survives that. Comparing two countries against each other does not, and no column in this section should be read as a ranking of national climate.

Limitations

• Europe only. The scope was chosen to match the Eurostat electricity price series exactly, so the two join with no gaps. Joined to the global emissions data it leaves the rest of the world null, the same way the electricity price column already does.
• A reanalysis, not a forecast model, in every year. The request names what answers it: ERA5-Land for the temperatures, ERA5 for precipitation, wind and radiation. Left unnamed, Open-Meteo answers the years from 2017 out of a weather-forecast model instead, which is finer and runs closer to today, and which would put a change of model inside every comparison that crosses 2017. A year's heating degree days differ by more than 5% between the two in over a quarter of country-years, and by a third or more in Malta and Cyprus.
• One cell per country. See A capital is not a country: fine for a country against itself, weak for one country against another. A population-weighted average over many cells is the honest version and costs many times the API budget this source has.
• The recent tail is preliminary. Open-Meteo serves the reanalysis about six days behind real time, from preliminary runs that Copernicus replaces with final ones two to three months later, so rows inside the last ninety days can change value between builds. Rows older than that are frozen, because they are carried forward between releases rather than refetched.
• Degree days are a demand proxy, not demand. They know nothing about building stock, insulation, occupancy or what a country heats with. A cold country with well-insulated housing and a mild one without can land the same way round here and the opposite way round on a gas bill.
• A correlation is not a causal test, in either direction. For prices, the finding is that the simplest weather explanation does not fit, which is what a control variable is for; it is not evidence for any particular alternative. For emissions the fit is real, but it carries whatever else moved with the winters: the gas crisis of 2022 and 2023 cut demand in two mild years, and some of that fall is in the fit too.
• Winters are continent-wide. Most of the emissions signal is the whole panel having a cold or a mild year together. Comparing countries within one year, which strips that out, leaves a much weaker relationship, so read the per-country chart as each country against its own history.

The tables are marts.fct_country_weather_year (these pages) and staging.stg_weather_daily (the daily grain underneath it, one row per country and date). Both ship in the data release, along with the raw landing table, which is the one table in this warehouse a rebuild cannot reproduce inside the source's daily budget.