An assessment of the potential for centennial-scale reconstruction of atmospheric circulation from selected New Zealand tree-ring chronologies

New Zealand's location and orography result in strongly-contrasting regional climates that are highly sensitive to variations in atmospheric circulation. Coupled with numerous remnant fragments of native forest, containing several long-lived tree species, this makes New Zealand ideally suited to multi-centennial investigation of changes in atmospheric circulation, at annual resolution. The potential of three such species (Agathis australis, Halocarpus biformis, Libocedrus bidwillii) was investigated. Mean sea level pressure (MSLP) composite mapping (1950–1992) was used to develop hypothesised relationships between tree-ring master chronologies and selected regional atmospheric circulation indices. The strongest multi-species relationship identified was with the Southern Oscillation Index (SOI), for a three-season window starting in the austral winter (JJA) prior to growth initiation. The stationarity of hypothesised relationships was tested using additional MSLP composite mapping (1901–1950) and split-record regression analysis, including investigation of the stationarity of relationships at the respective poles of the relevant circulation indices. Significant stationarity issues were encountered, leading, in particular, to the rejection of L. bidwillii as an atmospheric circulation proxy. Split-record calibration and verification was used to develop regression-based transfer functions for stationary relationships, but only the bivariate relationship between A. australis and the SOI passed additional nineteenth century verification tests. The derived 400-year SOI reconstruction suggests that the twentieth century may have been the most robust ENSO century. There is also evidence of decadal-scale periodicity in ENSO robustness from the mid-eighteenth century, possibly following a relatively quiescent seventeenth century. More generally, the results reinforce the imperative of explicitly addressing stationarity in palaeoclimatology. They also demonstrate the utility of gridded products (e.g. HadSLP2) and compositing, especially in the context of hypothesis generation.