In a figure that took considerable work, IPCC AR5 (First Draft) compared 5 regional proxy reconstructions to model output. In Australia, they used the Gergis (Neukom) et al 2012 reconstruction, In South America, they used a Neukom et al 2011 (Clim Dyn) reconstruction. In 2011, Neukom refused to provide me with the data versions used in this article (many of which are not public). I recently wrote to the editor of Climate Dynamics without acknowledgement. Their Table 2 lists 19 “proxies” used in their winter temperature reconstructions – one of which is Law Dome, in a remarkable highly truncated version.
Neukom et al 2011 (Climate Dynamics) has almost as many coauthors (18) as selected winter proxies: R. Neukom, J. Luterbacher, R. Villalba, M. Küttel, D. Frank, P. D. Jones, M. Grosjean, H. Wanner, J.-C. Aravena, D. E. Black, D. A. Christie, R. D’Arrigo, A. Lara, M. Morales, C. Soliz-Gamboa, A. Srur, R. Urrutia, L. von Gunten.
Neukom et al start with 144 climate “sensitive” proxies, from which 20 are selected into their winter network. As one would expect in a paper coauthored or influenced by Luterbacher, some of the “proxies” are not actually proxies, but instrumental records. Three of the “proxies” are long instrumental temperature records. Three other “proxies” are grid cells of v-wind vectors from the CLIWOC/ICOADS database.
For some reason, Neukom used a severely truncated version (only back to 1800(!)) of the Law Dome O18 series.
Neukom had purported to and had been funded to make a comprehensive collection of sh proxies and could hardly have been unaware of the longer available Law Dome O18 record. It’s cited in Neukom and Gergis 2011. Neukom’s coauthor Phil Jones was aware that earlier values of Law Dome were available, as he’d received them from Tas van Ommen in 2003. Plus Law Dome was even discussed in AR4.
And yes, inclusion of Law Dome would have “mattered”. It has an elevated medieval values. In their Australia article, they accepted a reconstruction with only two proxies (Cook’s Tasmania and Oroko tree rings.) So they should have been able to extract a winter SSA reconstruction with the longer available Law Dome reconstruction.
Law Dome is located almost due south of western Australia, half a world away from South America. Readers might well wonder why, in the “peer reviewed literature” being relied upon by IPCC, Law Dome is used as a proxy for South American winter temperature, but not a proxy for nearby Australia. I wonder the same thing.
Given the importance of screening in Gergis et al, looking at the corresponding algorithm in Neukom et al 2011 (Clim Dyn) seems timely. They describe their “methodology” as follows:
We established the set of predictors used for our summer and winter temperature reconstructions using two steps.
2.1. Identification of potential summer and winter temperature predictors
Out of the original database of 144 proxies (Tables S1-S3) that are related to SSA climate (e.g. temperature, precipitation or atmospheric pressure), we selected the predictors with temporally consistent and significant correlations with SSA summer or winter temperatures (potential predictor matrices). To evaluate the temporal stability, the 30-year running Spearman correlation coefficients of each proxy series with the 20th century instrumental CRU TS3 grid at the “best location” were calculated. As “best location” we defined the grid cell with the highest absolute correlation with the proxy over the overlapping period (between 70 and 106 years within 1901-2006). If the running correlation curve showed instabilities (i.e. changes in sign, or fluctuations in the coefficient that exceed ±0.2/decade) the relation between the proxy and the predictand was considered not stable and the proxy series was not included into the potential predictor matrix. In total, 44 (27) series were included into the potential predictor matrix for the summer (winter) temperature reconstructions. They are specified in the last two columns of Table S1 (tree ring clusters), Table S2 (individual tree ring records) and Table S3 (other records).
2.2. Selection of the final proxy sets
In a next step, we optimized the potential predictor matrices by identifying the optimal subset in terms of pre-defined reconstruction skill measures for each season. Due to computation limitations it was not possible to test all possible combinations of the available 44 (summer) and 27 (winter) temperature predictors. We therefore combined the proxies into meaningful sub-groups of maximum eight records based on their starting years. We then performed the PCR reconstructions and verifications for all possible combination of proxies within the oldest group and selected the combination with the highest skill scores. This set was then combined with all possible combinations of the next younger group and again, the set with the highest skill was evaluated. This procedure was repeated for all sub-groups. Finally, we used the leave-one-out (“add-one-more”) method to test, whether the quality could be further improved by removing (adding) some of the selected (excluded) predictors. For the SSA 5 mean reconstructions (see Table 3 in the main text) we used the average of the RE and r2 scores as measure for the quality. The spatial reconstructions were assessed based on the average number of grid cells with positive REs. We performed the evaluation by computing an SSA mean and a spatial reconstruction with each set and equally weighting both criteria. Independent use of the two criteria lead to two very similar predictor sets for the SSA mean and spatial reconstructions (not shown). Verification of the results by comparing the skill scores of selected subsets with the corresponding results of CPS and RegEM showed that the rankings of the subsets are relatively robust and similar for the three methods (not shown). The final set of 22 (20) summer (winter) temperature proxies and the temporal evolution of the number of predictors are presented in Table 1 (Table 2) and Figure 1 in the main text.
Even if they provided their networks (which they haven’t), this bizarre procedure would not be replicable in finite time. It’s also hard to figure out any conceivable statistical rationale.
There appears to be an interaction between their truncation of Law Dome and their screening procedure. It appears that they began their procedure for the accretion of proxies with the longest proxy group first. They should have included Law Dome in their long network and worked forward. It would also have to be checked in their summer network as well, the long portion of which has only four series: two Quelccaya series (accumulation, d18O), Oroko Swamp from NZ – also used in the Gergis Australian reconstruction and only one “new” series: Laguna Aculeo pigments from von Gunten et al. (2009).