The chironomid triennial

Over at the Subfossil Chironomid group on Facebook, Dr Larocque-Tobler posted a link to Zhang et al. (2017), describing it as impressive. I hadn’t seen the published version of Zhang et al, so I popped over fully prepared to be impressed.

Zhang et al report a transfer function for reconstructing July air temperature from chironomid assemblages using a 100-lake calibration set from the south-eastern margin of the Tibetan Plateau. The authors then apply this transfer function to a high-resolution chironomid stratigraphy from the high-elevation (3900m) Tiancai Lake. They report that the correlation between the reconstruction and the instrumental record from Lijiang (55km away and 1500m lower) is statistically significant (r = 0.45, p < 0.05,  n = 31).

This is the key figure.

Zhang et al Figure 6. (a) Chironomid-based mean July temperature (MJT) reconstruction results from Tiancai Lake based on two transfer function models: the solid black line is the reconstruction based on the weighted-average partial least squares (WA-PLS) bootstrap model with two components and the dashed black line is the reconstruction based on the weighted-average with inverse deshrinking (WAinv) bootstrap model. Red solid line is the instrumental data from Lijiang weather station, corrected applying the lapse rate and solid grey line is the three-sample moving average of the data set. Reconstruction of diagnostic statistics for the 100 lake data set where (b) displays the goodness-of-fit statistics of the fossil samples with MJT. Dashed lines are used to identify samples with “poor fit” (> 95th percentile) and “very poor fit” (> 90th percentile) with temperature [note: the percentiles are the wrong way round and should be poor fit >90th and very poor fit >95th]. (c) Nearest modern analogues for the fossil samples in the calibration data set, where the dashed line is used to show fossil samples with “no good” (5 %) modern analogues. (d) Percentage of chironomid taxa in fossil samples that are rare in the modern calibration data set (Hill’s N2< 2). (e) Comparison between the chironomid-based transfer function reconstructed trends (represented by MJT anomalies) with the instrumental data from Lijiang weather station (in red solid line, with three-sample moving average). The black solid line represents the reconstruction based on the WA-PLS bootstrapped model with two components using 100-lake calibration set.

The problem is obvious once you have found it – it helped immensely having some archived data. Compare the grey line in panel a with the red line in panel e. These are both supposed to be the three-sample moving average of the temperature data from Lijiang (a lapse-rate corrected; e anomalies) and should therefore have exactly the same shape, but the resemblance is limited. (Panel a was not shown in the Climate of the Past Discussion paper, so the reviewers at least can be absolved of any responsibility for not noticing this.)

Using the archived data from dropbox, I can confirm that panel a is correct, except that the curve should extend before 1960 as the instrumental temperature series starts in 1951. It took a while to work out what the authors had done in panel e: at least a couple of minutes.

If instead of taking the three-year moving average, you take every third year, you can get a plot that looks almost exactly like panel e. Interpolate the triennial data and the correlation with the reconstruction is very similar to what the authors report. This is obviously a very lucky error: nobody thinks that chironomids are only sensitive to July temperature every third year. This analysis implicitly assumes that the chironomids can predict temperature up to two years ahead! Now that would be impressive.

Reconstruction from Zhang et al (black) and every third year temperature data from Lijiang (red). Both series are presented as anomalies. cf panel e above

With the promised three-year moving average, the correlation is much weaker (r = 0.21, p = 0.28). I am not impressed.