from dr Judith Curry’s Climate Etc.
by Frank Bosse and Nic Lewis
A more recent work (MB Freund et al. 2023, thereafter MBF23) in “Naturkommunikation Erde und Umwelt” examines the variability of summer drought events since 1600 -Index (SPEI) from 1950 to today back to 1600.
The paper describes and uses a multi-proxy network across large parts of Europe (see Fig. 1 of MBF23) to reconstruct the history of summer droughts for a longer historical period. It finds interesting results about the dependence of these events on volcanoes and solar forcing. It is worth reading and we were interested in whether the title of the headline is justified and also this statement in the abstract:
“We show that the recent summer drought in Europe (2015-2018) is highly unusual in a multi-century context…”
Thanks to the authors, the SPEI reconstruction year data used are available, allowing us to perform calculations to verify these claims.
An apparent first “confirmation” of the title of the paper’s headline appears in Figure 3a in MBH23:
Fig.1: A reproduction of Fig. 3a from MBH23. Annual European averages of SPEI data in blue/red, low-pass filter output in black.
The black line in this figure shows the effect of applying 13-year low-pass smoothing, so it refers to the recent past. In fact, the 13-year Chebyshev filter used shows a “dramatic” drop down after 2010 to a much lower precipitation index than at any other time during the reconstruction period from 1600 to 2018. However, if you look closely, you can also find dry periods, before 1950, the Beginning of the classic SPEI dataset marked “SPEI”, or marked dark gray before 1880 in Fig.1, and the low-pass filter did not respond as it did after 2010.
The reason for this behavior is quite simple: all smoothing filters struggle with the beginning and end of a filtered data set. They estimate the output because there are no predecessors/successors in the raw data. To test the effects of these properties, we used the same data with a similar filter (Loess) and made a comparison with Fig. 1, but discontinued the filtering in 1949:
Fig. 2: Fig.1, but with a smoothed SPEI index up to 1949.
If the paper was written in 1950 it would find “unusual recent hydroclimate”, in 2023 it finds the same for the most recent conditions due to a filter problem. The beginning after 1600 is also very unusually wet in the filter exit for the same reason.
The dip at the end of Fig.3a of MBH23 is not real, but an artefact of the filter used.
A simple running mean filter, which has no output in the early years but no artifacts, gives a fairer smoothing of the fluctuations over 1600-2018:
Fig. 3: Summer SPEI data (black) filtered with a lagging mean (red). The historical minimum of this filter is shown as a broken red line. Clearly visible minima in the 1870s and 1680s and at the end of the period 1600-2018.
Fig. 3 shows the opposite result of the MBF23 heading: until 2018 (the last data point in the set in MBF23) it indicates that the recent European summer hydroclimate was NOT unusual, the SPEI index was in the range of natural variability.
To show that the claim in the abstract (“2015-2018 highly unusual”) is also not true, we looked more closely at the data and calculated these 4-year averages over the entire period.
It turned out that the 4-year average in the SPEI data was more negative in many periods than in the years 2015-2018, for which this average is -0.273:
Since 1900 there have been four such periods, all in the years before 1950: 1947-1950; 1946-1949; 1945-1948; 1944-1947. Indeed, the period prior to 1950 (not heavily influenced by anthropogenic forcing) was characterized by very dry summers, which are not mentioned by a word in MBF23.
Before 1900 there are also some periods:
1892-1895; 1760-1763; 1759-1762; 1738-1741; 1688-1691.
“The 2015-2018 European summer drought was NOT highly unusual in a multi-century context,” as the summary erroneously claims.
To further support this point, we also examined whether average values over longer periods of time are “highly unusual”.
A 5-year lagging average turned out to give 10 periods between 1600 and 1950, a time span mainly influenced by natural fluctuations, with more negative SPEI values than the most recent period up to 2018; a 10-year average gives 9 such periods before 1951. And a 3-year lagging average gives as many as 57 periods before 1951 with more negative SPEI values than the most recent period.
We also looked at the variability of annual data after 1950 (the ‘native SPEI’ time span) and before this year the time span of the reconstruction of the ‘European hydroclimate based on a network of tree-ring stable isotopes of oxygen and carbon ratios’ in MBF23. We calculated running 21-year standard deviations (sigma) of the annual data (Fig. 4):
Fig. 4: The variability of the annual SPEI data. Average values before 1950 and after that year are marked with a dashed line. Notice the jump.
The lower temporal variability of the reconstruction raises some doubts as to whether the reconstruction of the SPEI 1600…1950 makes sense in order to compare the more recent native SPEI data 1:1 with the historical reconstruction data before 1950. It appears that the reconstruction, even if otherwise valid, significantly underestimates the natural variability. This is a common problem with proxy-based reconstructions. This causes the magnitude of fluctuations during the post-1950 instrumental SPEI era to be exaggerated compared to natural variability, such that normal fluctuations may appear unusual.
MBF23 is a very valuable paper when it comes to describing the variability of European summer droughts since 1600. Neither the title “European tree ring isotopes indicate an unusual recent hydroclimate” nor the claim in its abstract that “the recent European summer drought (2015–2018) is highly unusual in the context of several centuries” are justified by the data used in the paper.
The lower temporal and spatial resolution of the pre-1950 reconstruction relative to the determined post-1950 SPEI raises some doubts as to whether the comparison of a few years after 1950 with the historically reconstructed values is appropriate.
MBF23 should be corrected and renamed because some important conclusions, including the headline claim in its title, are not supported by a proper statistical analysis of the SPEI values that their reconstruction method produces. The recent European drought up to 2018 remained within the range of natural fluctuations.