Detrending climate data prior to climate–growth analyses in dendroecology: A common best practice?

Tree growth varies closely with high–frequency climate variability. Since the 1930s detrending climate data prior to comparing them with tree growth data has been shown to better capture tree growth sensitivity to climate. However, in a context of increasingly pronounced trends in climate, this practice remains surprisingly rare in dendroecology. In a review of Dendrochronologia over the 2018–2021 period, we found that less than 20 % of dendroecological studies detrended climate data prior to climate-growth analyses. With an illustrative study, we want to remind the dendroecology community that such a procedure is still, if not more than ever, rational and relevant. We investigated the effects of detrending climate data on climate–growth relationships across North America over the 1951–2000 period. We used a network of 2536 tree individual ring-width series from the Canadian and Western US forest inventories. We compared correlations between tree growth and seasonal climate data (Tmin, Tmax, Prec) both raw and detrended. Detrending approaches included a linear regression, 30-yr and 100-yr cubic smoothing splines. Our results indicate that on average the detrending of climate data increased climate–growth correlations. In addition, we observed that strong trends in climate data translated to higher variability in inferred correlations based on raw vs. detrended climate data. We provide further evidence that our results hold true for the entire spectrum of dendroecological studies using either mean site chronologies and correlations coefficients, or individual tree time series within a mixed-effects model framework where regression coefficients are used more commonly. We show that even without a change in correlation, regression coefficients can change a lot and we tend to underestimate the true climate impact on growth in case of climate variables containing trends. This study demonstrates that treating climate and tree-ring time series “like-for-like” is a necessary procedure to reduce false negatives and positives in dendroecological studies. Concluding, we recommend using the same detrending for climate and tree growth data when tree-ring time series are detrended with splines or similar frequency-based filters.