Summer temperatures in the Canadian Rockies during the last millennium: a revised record

Abstract

We present a significant update to a millennial summer temperature reconstruction (1073–1983) that was originally published in 1997. Utilising new tree-ring data (predominantly Picea engelmannii), the reconstruction is not only better replicated, but has been extended (950–1994) and is now more regionally representative. Calibration and verification statistics were improved, with the new model explaining 53% of May–August maximum temperature variation compared to the original (39% of April–August mean temperatures). The maximum latewood density data, which are weighted more strongly in the regression model than ringwidth, were processed using regional curve standardisation to capture potential centennial to millennial scale variability. The reconstruction shows warm intervals, comparable to twentieth century values, for the first half of the eleventh century, the late 1300s and early 1400s. The bulk of the record, however, is below the 1901–1980 normals, with prolonged cool periods from 1200 to 1350 and from 1450 to the late 19th century. The most extreme cool period is observed to be in the 1690s. These reconstructed cool periods compare well with known regional records of glacier advances between 1150 and the 1300s, possibly in the early 1500s, early 1700s and 1800s. Evidence is also presented of the influence of solar activity and volcanic events on summer temperature in the Canadian Rockies over the last 1,000 years. Although this reconstruction is regional in scope, it compares well at multi-decadal to centennial scales with Northern Hemisphere temperature proxies and at millennial scales with reconstructions that were also processed to capture longer timescale variability. This coherence suggests that this series is globally important for the assessment of natural temperature variability over the last 1,000 years.

Appendix

1.1 Developing the RCS chronology for the MXD data

Regional curve standardisation (RCS) attempts to capture low-frequency information that is greater than the mean length of the tree-ring series used to build a chronology (Esper et al. 2003). In its most basic form, RCS aligns all available tree-ring series by cambial age to develop a mean series of ‘average behaviour’ that is typical for the species and region. Deviations of individual series from this regional curve are interpreted as ecological or climatological signals that result in higher or lower growth rates. By averaging the detrended series, the common ‘climate’ signal is accentuated, while the ‘ecological’ noise is minimised. However, great care needs to be taken when utilising the RCS method because (1) tree-ring data-sets may represent different populations (i.e. differing growth rates or growth trend types) that cannot be modelled by a single mean regional curve without introducing significant bias into the final chronology (Esper et al. 2003) and (2) an unquantifiable systematic bias may also be introduced into a RCS chronology when the mean biological age of all samples varies over time which results in inflation or depression of index values at either end of the sample time-series (Briffa and Melvin 2004; Melvin 2004).

To address the potential occurrence of differing ‘populations’ within the MXD data, the MXD series were divided into three groups, to allow better assessment of potential bias from the RCS method (Table 4). These groups are (1) all series from living trees at the Peyto and the Icefield sites; (2) snag data from the Icefield site and (3) snag data from the Peyto and Robson sites. For each sample, pith offset data were estimated (Aruani and J. Esper, 2004, personal communication) to reduce potential bias in generating the cambial age-aligned curves. The pith offsets for the living samples are generally higher than those for the snag material (mainly discs), as many cores did not reach the pith. Separate RCS curves were developed for each group (Fig. 7a). The Peyto–Robson snags were treated as a separate group because their mean MXD values are generally higher than for the other two data-sets. Using a single, regional curve would inflate the final indexed values for the Peyto–Robson data, possibly introducing bias in the earliest part of the reconstruction. The 95% confidence envelopes (Fig. 7b) for each age-aligned curve show that the mean functions are generally robust over their entire length.

Table 4 Grouping of tree-ring MXD series for RCS

Fig. 7

RCS chronology development. a Mean regional cambial age-aligned curves for the three-sub-group data-sets; b Mean regional age-aligned curves with two sigma error bars; c Comparison of the sub-group RCS chronologies; d Mean cambial age of all MXD samples for each year

The age-aligned curves (Fig. 7a) were smoothed with a cubic spline of 10% their length and used to standardise each maximum density series in their respective groups. This is the same method used by Esper et al. (2003). The final group RCS chronologies are compared in Fig. 7c. The correlation between the living tree and snag series from sites near the Columbia Icefield is reasonable (r=0.61, 1608–1897) but that between Icefields and the Peyto–Robson snags is more noisy (r=0.41, 1072–1312), probably reflecting the low replication of both series in this overlap period (Fig. 3a). The reasonably good fit between the three group chronologies (Fig. 7c) implies that the low frequency trends captured using RCS are real and not an artefact of the detrending procedure. Importantly, the good comparison between the Icefield Snag and Living RCS chronologies around 1700 (Fig. 7c) indicates that the low index values over this period (see also Fig. 3a) portray a relatively robust non-biased signal and are therefore likely not artificially depressed due to under-fitting of the age-aligned curves (Briffa and Melvin 2004; Melvin 2004) because of the low mean cambial age of the samples at this time (Fig. 7d).

The final MXD RCS chronology was developed by averaging the individual detrended series from all groups (Fig. 3a). An alternate MXD RCS chronology, generated without pith offset information was very similar to that presented here (r=0.986, 869–1994). This observation supports comments and analysis presented by Esper et al. (2003) that the lack of pith-offset information does not pose a serious bias when using RCS so long as sample replication is reasonably high.