Only two Gergis proxies (both tree ring) go back to the medieval period: Oroko Swamp, New Zealand and Mt Read, Tasmania, both from Ed Cook. Although claims of novelty have been made for the Gergis reconstruction, neither of these proxies is “new”, with both illustrated in AR4 and Mt Read being used as early as Mann et al 1998 and Jones et al 1998.
In today’s post, I’ll look in more detail at the Oroko tree ring chronology, which was used in three technical articles by Ed Cook (Cook et al 2002 Glob Plan Chg; Cook et al 2002 GRL; Cook et al 2006) to produce temperature reconstructions. In Cook’s earliest article (2002 Glob Plan Chg), Cook showed a tree ring chronology which declined quite dramatically after 1957. Cook reported that there was a very high correlation to instrumental summer temperature (Hokitika, South Island NZ) between 1860 and 1957, followed by a “collapse” in correlation after 1957 – a decline attributed by Cook to logging at the site. For his reconstruction of summer temperature, Cook “accordingly” replaced the proxy estimate with instrumental temperature after 1957, an artifice clearly marked in Cook’s original articles, but not necessarily in downstream multiproxy uses.
Gergis et al 2012 (which corresponds to PAGES2K up to a puzzling one year offset) said that they used “disturbance-corrected” data for Oroko:
“for consistency with published results, we use the final temperature reconstructions provided by the original authors that includes disturbance-corrected data for the 213 Silver Pine record…( E. Cook, personal communication)
By “disturbance correction” , do they mean the replacement of proxy data after 1957 by instrumental data? Or have they employed some other method of “disturbance correction”?
Assessment of this question is unduly complicated because Cook never archived Oroko measurement data or, for that matter, any of the chronology versions or reconstructions appearing in the technical articles. Grey versions of the temperature reconstruction (but not chronology) have circulated in connection with multiproxy literature (including Mann and Jones 2003, Mann et al 2008, Gergis et al 2012 and PAGES2K 2013). In addition, two different grey versions occur digitally in Climategate letters from 2000 and 2005, with the later version clearly labeled as containing a splice of proxy and instrumental data. The Gergis version is clearly related to the earlier grey versions, but, at present, I am unable to determine whether the “disturbance correction” included an instrumental splice or not.
There’s another curiosity. As noted above, Cook originally claimed a high correlation to instrumental temperature up to at least 1957, and, based, on their figures, the correlation to 1999 would still have been positive, even if attenuated, but Mann and Jones 2003 reported a negative correlation (-0.25) to instrumental temperature. However, Gergis et al 2012 obtained opposite results, once again asserting a statistically significant positive correlation to temperature. To the extent that there had been splicing of instrumental data into the Gergis version, one feels that claims of statistical significance ought to be qualified. Nonetheless, the negative correlation claimed in Mann and Jones 2003 is puzzling: how did they obtain an opposite sign to Cook’s original study?
As to the Oroko proxy itself, it does not have anything like a HS-shape. It has considerable centennial variability. Its late 20th century values are somewhat elevated (smoothed 1 sigma basis 1200-1965), but nothing like the Gergis 4-sigma anomaly. It has no marked LIA or MWP. It has elevated values in the 13th century, but it has low values in the 11th century, the main rival to the late 20th century, and these low 11th century values attenuate reconstructions where 11th and 20th century values are close. The HS-ness of the Gergis2K reconstruction does not derive from this series.
The Oroko Swamp site is on the west (windward) coast of South Island, New Zealand at 43S at low altitude (110 m). In December 2012, during family travel to New Zealand South Island, we visited a (scenic) fjord on the west coast near Manapouri (about 45S). These are areas of constant wind and very high precipitation. They are definitely nowhere near altitude or latitude treelines. Cook himself expressed surprise that a low-altitude chronology would be correlated to temperature, but was convinced by the relationship (see below).
In today’s post, I’ll parse the various versions far more closely than will interest most (any reasonable) readers. I got caught up trying to figure out the data and want to document the versions while it’s still fresh in my mind.
The Gergis2K Reconstruction and Oroko
First, for orientation, in Figure 1 below, I’ve compared the PAGES2K Australasian (“Gergis2K”) reconstruction with the PAGES2K Oroko chronology (expressed as a temperature reconstruction as in the PAGES2K archive). CA readers will recall that, for the calculation of the regional average, the Gergis2K reconstruction is converted to smoothed SD units, and, in those units, its blade closes at a monster 4-sigma. In contrast, the Oroko reconstruction has considerably more centennial variability and does not close at monster values.
Figure 1. Top panel – PAGES2K Australia reconstruction (in deg C.) Bottom panel – PAGES2K Oroko temperature reconstruction (converted to deg C basis 1961-1990) supplemented by Neukom et al 2014 data. The Neukom data was used to extend the PAGES2K version at both ends.
Cook et al 2002 Glob Plan Chg
This is the earlier in genesis of the two 2002 articles. In Cook’s Oroko articles, he typically shows the chronology in one figure and the temperature reconstruction in another figure. In the Glob Plan Chg article, Cook showed an STD chronology (trees individually standardized) based on a smaller network, while the 2002 GRL article reported a RCS chronology (one standardization curve) on an expanded network. (STD chronologies have less centennial variability than RCS chronologies – which can have other defects – as interannual variance is incorporated into inter-tree variance.)
The chronology illustrated in their Figure 3 (excerpted below in my Figure 2) shows a noticeable decline in the late 20th century – shown more clearly in the zoom in the right panel. Cook surmised that there had been logging at the site around 1957 (the 1957 breakpoint is marked in red) and attributed the subsequent decline to the effects of this logging. Much of the article is devoted to discussion of circumstantial evidence of the logging. Although Cook said that there was “no evidence of cut stumps along the transect line”, they reported that there were “remains of a small-gauge tramline in part of the swamp that was used to log selected L. colensoi during the 20th Century”, deducing the disturbance primarily by analyzing year of death of many trees.
Figure 2. From Cook et al 2003 (Glob Plan Chg) Figure 3, showing Oroko chronology in chronology units.
For the period 1894-1957, Cook reported high correlations of the chronology to summer temperature, selecting (from several alternatives) a correlation of 0.61 with Jan-March temperature as a basis for reconstruction of JFM temperature.
In the resulting temperature reconstruction (shown in their Figure 6 – see below), Cook replaced proxy data after 1957 with temperature data. This was clearly stated in the caption and additionally illustrated in a separate panel B (not shown here) limited to the instrumental period.
Figure 3. Temperature reconstruction of Cook et al 2002 (Glob Plan Chg) as shown in their Fig. 6. Original caption: Reconstruction of January–March temperatures from Oroko Swamp L. colensoi tree rings (A). The calibration is based on joint 1866–1957 tree-ring and temperature data only (B) because of the post-1957 disturbance in the tree rings. The actual temperatures, scaled to the same amplitude as the tree-ring estimates, have been appended to the tree-ring estimates to allow for an evaluation of temperatures up to 1999.
Cook et al 2002 (GRL)
The nearly contemporary Cook et al 2002 (GRL) showed an RCS chronology on an expanded dataset (increased from 145 to 260 ringwidth series) in its Figure 1 (excerpted below). It was mentioned in a 2000 Climategate email (CG2-3759) from Cook to Briffa, which included a digital version that is closely related to (but not identical) to the illustration. (It is overplotted on the right panel shown below.)
As in the earlier article, the chronology shows a post-1957 decline, though the decline in this version is attenuated from the STD chronology published earlier in the year: the difference appears larger than I would expect merely from the difference in methods. This version shows elevated values in the late 12th and early 13th centuries, but has very low values in the early 11th century, which is a very warm period in most NH reconstructions.
Figure 4. From Cook et al 2002 (GRL) Figure 1. Original caption: The Oroko Swamp Austral summer temperature reconstruction, with overlaid 40-year smoothing (thick red line). Annual error estimates in blue (±1 standard error of the estimate from regression) are also included (yellow). After 1957, the actual Hokitika data have been appended to the reconstruction (dark blue line) to allow the full 20th century to be compared to the previous 1,000 years. The thin black horizontal line is the 20th century mean of Hokitika instrumental temperatures. The three vertical arrows mark the years of Alpine fault earthquakes (1460, 1630, and 1717). Yellow in right panel – overprint of digital information in Climategate CG2-3759.
Cook observed high correlations between this chronology and Jan-March temperature at Hokitika (which is also one of seven stations in the NIWA seven-station composite recently in blog news). The correlation was 0.62 in 1894-1957. Cook observed that the reconstruction verified (according to usual dendro methods) according to high verification r (and r2), RE and CE values in a 1866-1893 verification period, but that there was a “collapse” in the fidelity of the response in the 1958-1999 period, which he attributed to growth being “severely disturbed by known selective logging of the site”.
Figure 2. Cook et al 2002 (GRL) figure 2. Original caption: Calibration and verification results. The Oroko Swamp chronology has been calibrated against January- March mean temperatures from Hokitika, yielding a regression R2 of 0.38. The correlation between actual and estimated temperatures in the 1866-1893 verification period indicates that the tree-ring estimates are valid. Tree growth after 1957 was severely disturbed by known selective logging of the site, which resulted in a collapse in the fidelity of the tree-ring response to climate.
For the GRL temperature reconstruction, Cook once again replaced proxy data after 1957 by instrumental data after 1957, though but the smoothed version only went to 1957. Overplotted onto this figure is data from a 2005 Climategate email, sent by Cook to AR4 editors and shown in a figure in AR4 (also later used in Mann et al 2008). Cook explained in the covering email:
JANUARY-MARCH TEMPERATURES RECONSTRUCTED FROM OROKO SWAMP, NEW ZEALAND SILVER PINE TREE RINGS BE ADVISED THAT THE DATA AFTER 1958 ARE INSTRUMENTAL TEMPERATURES
The values appear to closely match Cook et al 2002 GRL values up to 1957, but, curiously, the amplitude in Figure 3 after 1957 (from experiment) is approximately 1.5 times the amplitude in the underlying data.
Figure 4. Cook et al 2002 (GRL) Figure 3 showing temperature reconstruction with splice of instrumental temperature after 1975. Original caption: The Oroko Swamp Austral summer temperature reconstruction, with overlaid 40-year smoothing (thick red line). Annual error estimates in blue (±1 standard error of the estimate from regression) are also included (yellow). After 1957, the actual Hokitika data have been appended to the reconstruction (dark blue line) to allow the full 20th century to be compared to the previous 1,000 years. The thin black horizontal line is the 20th century mean of Hokitika instrumental temperatures. The three vertical arrows mark the years of Alpine fault earthquakes (1460, 1630, and 1717). Yellow/gold – overplotted is the version contained in the 2005 Briffa email (also Mann et al 2008). After 1957, the values shown in the figure have a 50% greater amplitude than figures in the digital file.
Mann and Jones 2003
Mann and Jones 2003 was the first multiproxy series to consider use of the Oroko series, which was listed as one of 5 candidate long reconstructions: Fewer long series are available for the Southern Hemisphere (SH), where we make use of temperature reconstructions from 5 distinct regions….
(3) ‘RCS’-based tree-ring temperature reconstruction [warm-season; Cook et al., 2002] for southern New Zealand back to AD 900.
However, the Oroko series was eliminated from their network because Mann and Jones calculated a negative 0,25 correlation between the Oroko reconstruction and local temperature (see value in the Mann and Jones 2003 location map excerpted below). This was opposite to the value obtained by Cook and subsequently by Gergis. Mann and Jones described their procedure as follows:
Local (decadal) correlations were calculated between each proxy record and the instrumental grid-box surface temperature records for the regions they represent over the period 1901-1980 (see Figure 1). Proxy records exhibiting negative or approximately zero local correlations (SH record #2 and #3) were eliminated from further consideration in the study.
Figure 5. Location map of Mann and Jones 2003. Correlations to instrumental temperature shown in parentheses, with negative correlation -0.25 shown for Oroko, New Zealand. In their text, they stated: “Local (decadal) correlations were calculated between each proxy record and the instrumental grid-box surface temperature records for the regions they represent over the period 1901–1980 (see Figure 1). Proxy records exhibiting negative or approximately zero local correlations (SH record #2 and #3) were eliminated from further consideration in the study.”
Cook et al 2006
Cook et al 2006 (J Quat Sci) also showed an Oroko chronology compared against a ring width chronology from Ahaura, about 100 km away, arguing in favor of a common pattern. Once again, the post-1957 chronology shows a decline. The 2000 RCS chronology version is overplotted (yellow) on right panel, showing that the revised chronology remains close to the 2000 versions.
Figure 6. Excerpt from Cook et al 2006 Figure 7, top panel, showing Oroko chronology, inclusive of post-1957 decline.
As in the earlier articles, the temperature reconstruction of Cook et al 2006 includes spliced instrumental data, shown after 1957 in the diagram below. In this case, values from the 2005 Climategate email (equivalent to Mann et al 2008) are shown in gold prior to 1957; after 1957, the amplitudes are amplified by 1.5 for reasons that are not evident. The values go up, rather than down (as in the chronology shown above.)
Figure 7. Excerpt from Cook et al 2006 Figure 6, showing Oroko temperature reconstruction of Jan-Mar temperatures. After 1957, instrumental values are splice, additionally being shown in the diagram with increased amplitude.
The Oroko version sent to IPCC authors by Cook was subsequently shown in IPCC AR4 Figure 6.12, truncated in 1957 as shown below, with the commentary shown below:
Another tree ring reconstruction, of austral summer temperatures based on data from South Island, New Zealand, spans the past 1.1 kyr and is the longest yet produced for the region (Cook et al., 2002a). Disturbance at the site from which the trees were sampled restricts the calibration of this record to the 70 years up until 1950, but both tree rings and instrumental data indicate that the 20th century was not anomalously warm when compared to several warm periods reconstructed in the last 1 kyr (around the mid-12th and early 13th centuries and around 1500).
Figure 6. Excerpt from IPCC AR4 Figure 6.12 showing two Southern Hemisphere proxies, including Oroko (New Zealand.)
Mann et al 2008
Mann et al 2008 used exactly the same data as Cook had sent to the IPCC authors, including the instrumental data spliced together with proxy data: the data is plotted below. Unlike the original chronologies, the series closes on an uptick.
Figure 7. Oroko temperature reconstruction from Mann et al 2008 data: instrumental data is spliced after 1957 to proxy data before 1957.
Neukom and Gergis 2012
Neukom and Gergis 2012 is cited on several occasions in PAGES2K, but no data was archived in connection with this article. Worse, Neukom refused to provide data even on request. Neukom saved a short (1900-2000) portion of the Oroko series (attributed to Cook, pers comm), which matches the Climategate 2005/Mann et al 2008 version up to 1957, but its post-1957 portion doesn’t appear to match either the earlier or later versions.
PAGES2K and Gergis et al 2012
The PAGES2K version of Oroko is a temperature reconstruction, that is highly related to the earlier version, but which has been recalculated somewhere along the way: its subset up to 1957 has a correlation of 0.80 with the earlier version. Both versions have centennial variation and “cold” 11th centuries. Unlike the earlier temperature reconstructions (in which temperature data was spliced) it shows a decline, opening the possibility that it uses “actual” data after 1957, rather than spliced instrumental data: its appearance looks related to the chronology in Cook et al 2006 (as opposed to the spliced instrumental data), seemingly with an upward bodge reminiscent of the “Briffa bodge” of Cook’s friend. Perhaps”disturbance correction” mentioned in Gergis et al 2012.
Figure 8 Oroko temperature reconstruction from PAGES2K. Its post-1957 is reminiscent of the chronology in Cook et al 2006, rather than the spliced instrumental data.
Both Gergis et al 2012 and Gergis2K purported to reconstruct summer SONDJF temperature, but there is 0 correlation between the two versions as extracted from the respective archives.
However, if the Gergis 2012 version (normalized) is dated one year earlier, it has an exact correlation to the PAGES2K version (deg C).
<code>cor(ts.union(lag(gerg,lag=1),pages),use=use0)[1,2]# 1 </code>
In other words, the Oroko ring width (reconstructing SONDJF) is assigned to the following NH calendar year in Gergis et al 2012 and to the previous NH calendar year in PAGES2K. The same pattern holds with the AUS regional reconstruction: if the Gergis et al 2012 version is dated one year earlier, it matches the PAGES2K.
To further confuse matters (and I’ve triplechecked this), it appears to me that the Neukom et al 2014 Oroko version (and SH reconstruction) is in effect assigned to the calendar year one year before that i.e. it implies that summer temperatures of the preceding year force ring widths, not temperatures of the current summer.
<code>cor(ts.union(lag(lag(gerg )),neuko),use=use0)[1,2] #1 </code>
The Neukom et al 2014 version also contains a substantial extension of the reconstruction into the first millennium, an extension that is thus far unreported elsewhere, and a couple of additional recent values that continue the decline: the series closes slightly below its long-term average.
Figure 9. Oroko temperature reconstruction extracted from Neukom et al 2014. The normalized values of N14 are converted to temperature reconstruction by the exact linear relationship between the versions.
Revisiting Temperature Correlations
It’s not entirely certain that the PAGES2K version is an unspliced chronology. In the figure below, I’ve done a comparison of reconstruction to Hokitika summer temperature in the style of the original Cook article, but using the more recent PAGES version and updated Hokitika temperature (using a slightly dated CRU version.) In the updated calculation, I obtained slightly higher 1894-1957 correlation than reported by Cook ( 0.72 versus 0.62) as well as in the 1866-1893 period (0.67 versus 0.65 – though, for this step, I didn’t bother interpolating to missing data.)
In the post-1957, I got a much higher correlation (0.39 versus the previous 0.11). The import of this calculation depends on precisely how the “disturbance correction” was implemented.
Figure 10. Comparison of reconstruction to Hokitika summer temperature in style of Cook et al 2002 Figure 2, using updated data.
What, if anything, does this mean?
First, the continuing failure of IPCC authors to archive measurement data and chronologies as used makes the interpretation of these results far more onerous and uncertain than ought to be acceptable for results later cited in policy documents. PAGES2K used yet another grey version of the Oroko chronology. Because the properties of “disturbance correction” are unknown, it is impossible to assess whether the “statistically significant” relationship to temperature claimed by Gergis et al is impacted by a temperature splice or not.
Second, while Cook clearly marked his own replacement of proxy data with temperature data, it really is an unacceptable practice. The poor practice is exacerbated when spliced data is used downstream without the downstream authors attaching red flags. This happened with the Briffa bodge of the Tornetrask data and happened again with Mann et al 2008’s use of the spliced series.
Third, the negative correlation asserted by Mann and Jones 2003 seems unlikely: one wonders where it came from.
On a previous occasion (here), I noted that Gergis2K only had two proxies that reached back to the medieval period. While late 20th century values are elevated, they are not exceptional and do not explain the 4-sigma blade of the Gergis2K reconstruction. In a subsequent post, I’ll discuss the construction of the Gergis2K blade, which I have not yet parsed. Readers will recall that undetrended screening applied to the Gergis et al 2012 network resulted in only a few passing proxies. Gergis2K claim to have used undetrended screening, but permitted relationships with any gridcell within 500 km and/or with lag/lead of a year. I will try to determine whether their definition of statistical “significance” was amended to reflect this expansion of targets (I doubt it.) As previously noted, the Gergis2K network changed from the Gergis et al 2012 only on the edges: 20 of 27 proxies carried over; a few shuffled out and a few shuffled in.
But while screening undoubtedly plays a role, I think that the main methodological issue will be one that it has been underdiscussed: how to combine short proxies with long proxies. The vast majority of Gergis2K proxies are short corals with strong trends. Stapling such data onto long tree ring data with centennial variability, but negligible trend, will create a HS. This sort of stapling has occurred in previous reconstructions – Jones et al 1998 is an example, but has received less attention than more exotic techniques. At some point, I’ll try to figure out Gergis’ methodology and see the effect of their (presumptive) stapling.