City of Musicians

A reader calling himself “Raven” recently linked to this page by Hans Erren which graphs some central European temperature records going back to the 1700s. It features this graph:

This shows annual average temperature for five central European cities. The data are stated to be from GISS, but since the GISS website no longer provides data prior to 1880 the author links to the GHCN version 1 where the early data are still available.

The scale of the graph is not for all these data sets; Erren has offset some of them so that they align with each other. The alignment is indeed quite good, showing how strongly temperature changes in nearby locations correlate with each other. In effect, any one of the data series contains almost all the information in the five data sets. But Erren has also adjusted some of the five stations to match his model of “urban heating” based on population. For Basel, he has plotted the GISS adjusted version while for de Bilt and Uccle he has applied his own adjustment. He has also extended the Basel data by appending data from Zurich.

Later on the page he plots temperature anomaly rather than absolute temperature, all series normalized to a 1901-1930 baseline. This is probably a better choice than offsetting some of the series, but it makes little difference in the appearance of the graph (other than to change the numbers on the y-axis).

He also plots these, and a few other, data sets with trends, one for the entire time span, another for the period 1880 to now:

On this basis Raven states

We have temperature data going back to the 1700s in parts of Europe (which is comparable to today).

He also says:

This data makes it clear that the warming trend would be much less significant if data prior to 1800 was incorporated into the graph. That makes a graph starting in 1900 as “deceptive” as a graph starting in 1998.

Will these claims stand up to scrutiny? First, the graph I displayed which Raven calls “deceptive” doesn’t start in 1900, it starts in 1850. But that’s a minor point. Let’s look at some of this data more closely; since all the stations shown correlate strongly with each other, we’ll examine the first location plotted, a city for which I have very fond personal memories. It’s the city of musicians: Vienna.

The aforelinked website provides an Excel file containing the data and the graphs. This made it easy to remove the other data sets and produce a graph identical to the first, but with only the Vienna data:

The graph is not quite up to date, so I retrieved all available data from the original sources and made an up-to-date version (these are annual averages of temperature anomaly using the entire time span as a baseline):

The graphs are similar but not identical. So I investigated where the difference originated, comparing the data I used (from referenced sources) to the data used by Erren (included in his Excel file). First I took the monthly data I used and computed annual averages, except for years which didn’t have data for all 12 months. Then I did the same for Erren’s tabulated monthly data, and computed the difference between his values and mine. Here’s the result:

From this it’s clear that either Erren’s data, or that in the linked sources, includes an adjustment not included in the other. But the difference between his data and mine is negative for the early part of the data, zero for later data, and nowhere more than 0.1 deg.C. Using his data introduces a warming trend which is not present in mine; this means that if I use the data from linked sources, my results will underestimate total warming over time compared to Erren’s data. In fact the total warming estimated from using my data will be 0.1 deg.C less by the time we get to the present, which will favor the notion that earlier times were “comparable to today.” I’m perfectly willing to yield that advantage to Raven’s claim, essentially cheating in his favor. So on this basis, I’ll stick with the data from linked sources.

But that doesn’t account for the difference in the graphs. So I compared the annual averages used by Erren (which he also provides) to the annual averages from my data:

These differences are much larger, both positive and negative, and show no trend or pattern except that already observed, which cheats in favor of Raven’s claim. Clearly the annual averages plotted by Erren don’t agree with annual averages computed from the monthly data he supplies himself! This was a real puzzler, until I discovered that Erren plots annual averages for the climatological year (December of the previous year through November of the given year) rather than the calendar year.

Calendar-year averages and climatological-year averages are both completely valid, and omitting the adjustments in Erren’s data in order to cheat in Raven’s favor certainly won’t unfairly test his claims. So I stuck with the data from linked sources (which I’ve called “my” data) and with the calendar year. This certainly enables us to test — fairly — whether or not the Vienna data support Raven’s claim that past temperatures were comparable to today.

So let’s look again at my graph which is similar to Erren’s, for Vienna. One of the things I don’t like about the original graphs is that the vertical size is quite small, and a good portion of it is taken up by the title and legend, leaving a very small y-axis. This makes changes over time harder to see and reduces the visual impact of those changes. Here’s a version with a much larger y-axis:

Now we can see much more clearly the warming of recent times. We can also see that although individual years in earlier times are almost comparable to today, longer-term averages are not, and that the amount of variation in earlier times is greater than today.

The greater year-to-year variation in earlier data is probably due to less precision for earlier data. This makes it all the more important to smooth the data in a way which clarifies the longer-term trends (the climate) rather than the year-to-year fluctuations (the weather). I’ll smooth the data on a 10-year time scale, in two different ways. First, I’ll use a Savitsky-Golay filter (labelled “Smooth” and plotted in red). Lest anyone suspect that some complicated filtering method has introduced artificial trends, I’ll also smooth the data by one of the simplest of methods: computing 10-year moving averages (plotted in blue). Here’s the result:

Now we start to see much more clearly that the temperature of earlier times is not comparable to today. This becomes starkly obvious when we plot just the smoothed versions and expand the y-axis even more:

The 10-year time scale smoothed value today is a full deg.C warmer than any 10-year time scale value before 1980. I could repeat all the foregoing for the other time series, but that’s a heckuva lot of work and since they’re all so strongly correlated we can predict with extremely high confidence that the result will be the same. I have in the past, just to satisfy my own curiosity, done similar analysis of other very long temperature records including the longest of all, Central England Temperature (CET), which goes back not merely to the 1700s but to the 1600s. The result is the same. It’s abundantly clear that Raven’s first claim is false: earlier temperatures from long time series are not comparable to today.

As for the second claim, it’s quite true that the average trend over the last two-and-a-quarter centuries is less than the average trend over the last century-and-a-quarter. But that’s only emphasizes that the long-term trend is dominated by the recent warming! If temperature were exactly constant for all time until recently, then as we extend linear regression (which estimates the average trend rather than the present trend) further back in time it’s estimate of the average rate would go to zero. The whole point is that the modern trend is much faster warming than the long-term temperature changes, because it’s caused by man-made influence on the climate system due to the side effects greenhouse gas emissions. Hence the reduced average rate for the longer time span makes the modern trend all the more significant rather than less: Raven’s second claim illustrates utter failure to understand the significance of modern warming.

Raven’s mistaken claim about earlier temperatures being comparable to today is due to visual inspection of a graph which shows only annual averages, so that short-term changes are emphasized, and with a temperature axis so small that all changes have much reduced visual impact. Raven’s mistaken interpretation of the meaning of long-term trends over different spans of time is simply … incomprehensible.