It’s hot these days, it’s unusually hot, even dangerously hot, in a lot of places a lot of the time. While America has been hit hard this summer, it seems that Europe has been hit even harder. Naturally, this causes a lot of people to speculate, or to outright declare, or to outright deny, that the frequency and severity of heat waves has been increased by man-made climate change, also known as “global warming.”
I decided to study a single location in Europe, the daily high temperature in a small region near the town of Beaune, France, known for centuries as an outstanding wine-making region.
I retrieved data from the ERA5 reanalysis, and for each year I isolated the data for summer, which is defined as the months of June, July, and August. Then I computed each year’s summer average temperature, giving this:
I probably should have left out this year (2022) because it’s not over yet (these data only go through June!), but I left it in place anyway. You might already have guessed that there is a statistically significant trend, summer is warming up. What’s not so obvious is that there is a statistically significant change in the rate of summer warming, which got faster around 1980.
The red line shows a lowess smooth, but you can verify not just an upward trend, but the acceleration around 1980, either with changepoint analysis or simply by fitting a quadratic curve using least-squares regression.
It’s also worth noting that the hottest single summer by far was 2003. That was the year when a massive fluctuation combined with the upward trend to make the hottest summer weather on record for this region.
If we’re interested in extreme heat, then we want to know more than just how the summer average has changed. In particular, we want to know just how hot it gets each year and how that may have changed over time. It turns out that the same pattern we see in the summer average, is also present in the hottest day of each year:
For other measures of how much excess heat each year brings during summer, I also defined a “hot day” as one in the top 5% of all summer days, and counted how many such days occurred each summer; again the year 2003 really sticks out with more than any other, and in addition (as in, even if you leave out the year 2003 entirely) an upward trend is clear (and statistically significant).
I also defined the “hot degrees” as the accumulated temperature over threshold for each year; again, 2003 exceeds all others and again an upward trend is clear and statistically significant (even if you leave out the year 2003 entirely):
All told, it’s abundantly clear that summer temperature is increasing, and all the given measures of extreme hit are increasing too.
Perhaps the real detail is in the actual probability density function for summer temperature. I estimated it, using both a histogram and a kernel density estimator, for each decade individually, to investigate whether, and in what way, the distribution was changing. The first decade in this record (1950-1960) is quite different from the last (2010-2020):
Furthermore, the difference is statistically significant (overwhelmingly so, according to the Kolmogorov-Smirnov test), and it’s not just because the mean value has increased; even the shape of the distribution has changed significantly, showing higher variance to accompany its higher mean value, both of which increase the probability and severity of extreme heat.
Looking at the estimated probability density for each decade, we see that the first four of them (from 1950 through 1990) are similar, but the last three (from 1990 through 2020) are markedly different:
This suggested comparing the distributions for the data before and after 1990, and it turns out to be very similar to the comparison of the first and last full decades given above:
However, the statistical significance of both results — the the average has increased, and that the variance has increased — is greater because the comparison is based on larger sets of data.
There is simply no doubt that in this wine-growing region of France, summer extreme heat has increased, in terms of hottest day of the year, number of hot days, and total excess degrees per summer. This is reflected in the increase in summer average temperature, and although that is the dominant factor in the rise of extreme heat, it is further exaggerated by the increase in the variance of summer temperature.
But this record only goes back to 1950. Fortunately, for this region we have an excellent proxy data set directly related to summer temperature. In general, the hotter the summer in Beaune, France, the sooner they harvest the wine grapes.
Those dates are on record, and if we use them as proxy data to predict the summer average temperature, it does a remarkably good job, explaining 60% of the variance of the temperature data:
Now the interesting part: the record of grape harvest date (GHD) in Beaune, France goes back to the year 1354. If we use the grape harvest dates to estimate summer temperature, and use outbursts of extremely hot summer as indicators of extreme summer heat, we can estimate it going back over 650 years:
Clearly, extreme summer heat over the last few decades has been more frequent, and more severe, than in the previous 650 years.
It’s because of man-made climate change (global warming). This summer is what a 1.2°C-heated planet looks like. But in the future this won’t seem so bad — compared to the 1.5°C-heated planet. If we get to 2°C …
This blog is made possible by readers like you; join others by donating at My Wee Dragon.
For grape harvest dates prior to 1582, I presume you adjusted for the Gregorian calendar reform? I’m sure you will have done but there was no mention of it in the article.
[Response: That was done by authors of the original study.]
Thx, it was a pleasure to read, esp. the grape harvest record thing.
Now we only need to convince our respective fellow countrymen and -women the decarbonization in 30 years is a really good idea and in 25 years might even be a better idea.
Merci beaucoup Tamino, très intéressant.
Why do you think the warming trend begins so late ? Your plot appears to show very little warming prior to about 1975 or 1980, but CO2 emissions from industry began at least 100 years before that.
The rate of CO₂ emissions during the 19th century were minimal compared to that of the last few decades, and this is reflected by the CO₂ concentration in the atmosphere:
The CO₂ has increased from 285 ppm in 1850 to 417 ppm in 2022, but HALF of that has occurred after 1988, the same year as the IPCC was established!
The graphs appear to show a slowing decline followed by an accelerated incline. The point at which it changed trend looks to be around 1970 (Using the average temp graph)
CO2 emissions may have begun in measurable quantities around 1850 but they really became significant post WW2, as can be seen here:
Since then, CO2 emissions have increased 7x from 5 Billion t to 35 Billion t.
What is even more interesting is that prior trends suggest a cyclical nature of temperature extremes, with the trend in the 20th century seeming to be moving towards lower average temperatures. This appears to have been slowed and eventually overturned as CO2 emission increased leading us to the calamity we now find ourselves in.
Read the post. Slowly.
This analysis is concentrated on specifically defined extreme climate events in the climatological data of one specific place. In a way, and others in this blog with far more knowledge and experience can correct me if wrong, in a way Tamino is actually statistically analysing “weather” rather than “climate”, to demonstrate unmistakable fingerprints of the later’s influence on the former.
Extreme events in one location are rare by definition, therefore to pick up statistically significant trends is always going to be harder than from all aggregate data because you have significantly less data points.
The point of Tamino’s post is that climate change is now so pervasive throughout the planet and of a magnitude that its effects are now clearly discernible even in character of “local weather events” where global climate is just one of (and not even perhaps the highest in relative weight ratio) variable influencing it.
This comment would make much better sense referring to Mass’ blog posting discussed in tamino’s previous post.
Yes. All observations are a combination of short term variation–weather–and longer term variation–various cycles and trends commonly called climate. O = f(weather) + f(climate) + error.
Certainly it’s obvious that the harvest proxy is related to weather all season which is more of a climate than a strictly weather measure. Tamino doesn’t use single rare events in his two definitions of warmth here: (1) average high across the season and (2) degree days across the season above criterion are both aggregated using all seasonal values.
Using rare events–as for example Mass does in the posting–indeed does maximize the contribution of weather variation which by definition minimizes the relative contributions of climate. So yes, using all data–i.e., aggregating all values over various time spans rather than selecting/cherrypicking a single specific datum/span–allows a much more sensitive test.
In the same way, the use of a single location is problematic for short term work, yes, however a near millennial look at one location does get into global climate. As well, this is the logic behind ice core data from the polar regions or data from tree rings being used as global proxies going back thousands of millennia.
In Japan there is a temple which records the cherry blossom date each year. This dataset is over a 1000 years and similarly shows warming in recent times. However it also shows warming from 1800’s. Do check it out.
I see there is also a definite visual, if not actual, variance change. No idea what it means. Restriction of range effect of warming bumping up against the annual growth/blossoming hormonal cycles maybe?
Berkeley Earth have a temperature record running 1743-2020 with the location 47.42 N, 3.55 E. Not sure of the name of this station but it must be about 30 miles NE of Beaune (47.0N, 4.8E). If the JJA temps are plotted out for this station, a very similar trace to the Beaune proxy trace is seen, decadal averages wobbling along flat for 225 years (flatter than the annual plot shown on the Berkeley Earth web page), up to the early 1980s.
Though Science Po literary theorist and sociologist of science Bruno Latour takes a dystopic look at the entwined decay of ecology and politics im Down to Earth: Politics in the New Climatic Regime he is no stranger to Beaune— or pinot noir as a palaeoclimate proxy
Down the road in Aloxe- Corton is Maison Louis Latour whose Domaine spans 48 hectares of vineyards and produces and exports more Grand Cru Burgundy than any other.
Though far from disinterested, he has a unique understanding of the degree to which a region can be hostage to climate change, for great wine often comes from places where vines are greatly stressed.
He’s always worth talking to.