Estimated tree longevity response to climate at risk of methodological bias

Cores are often sampled from living trees, which have an unquantified span of life ahead of them. This means that frequently, ring width data in ITRDB do not capture a tree’s full lifespan. Gao et al.² approached this by estimating tree longevity using the 99^th percentile age of all ring width series at each site, thus calculating the oldest representative age from each site. This would be a pragmatic solution to estimate longevity if all samples were taken from old growth forests. However, the ITRDB also includes ring-width data from plantations³, sub-fossils⁴ and different kinds of managed forest. In fact, old growth forests unlikely predominate ITRDB, because forests without human disturbance account for only 30% of the world’s forests⁵. When measurements are from the ~70% other forests, the oldest trees are not representative of the potential lifespan that the trees could reach under their respective climate conditions, because their age is prone to have been impacted by forest management regimes. This reflects the survivorship bias that tree longevity studies can suffer from, because forests can only reach their potential longevity if they evade extrinsic mortality from, for example, human impacts⁶. Moreover, historical forest management can moderate tree growth responses to climate change⁷, further obscuring climate-longevity relationships when management has not been considered.

The role of forest management to ITRDB ring width series can be profound. We quantified its potential role on sites in Europe² (excluding Russia), because forest management is well-documented here, and it represents a substantial portion (28%) of the entire dataset published by Gao et al.², for one continent. We found that only 5.7% of the European sites published by Gao et al.⁸ were proximal to known primary forests (Fig. 1a; 1.7% within 5 km of a primary forest point⁸, and 4% within a primary forest polygon⁹). There may be sites from primary forests not included, here—for example, 34% of sites had an undiagnosed forest regime. Nonetheless, a much larger percentage (65%) of sampled sites in Europe were diagnosed as managed¹⁰. This leads to our greatest concern, that a strong spatial variability in the distribution of managed and primary forests⁹ biases longevity interpretations, by representing the maximum potential tree longevity in a spatially-dependent way. It means that regions with lower representation of primary forests are biased to shorter longevity estimates, because forest management variably restricts forest longevities, obscuring meaningful relationships between longevity, climate, and growth strategies. To make matters worse, forest management practices are profoundly different between regions¹¹ (Fig. 1b), increasing their obscurement of meaningful longevity interpretations by having a variable, spatially dependent, role to tree longevity. For example, the tree species has a role in harvest strategy and intensity², and this is relevant to sites in Gao et al.¹² because some species have typical harvest ages longer than 100 years¹³. Furthermore, trees in even-aged managed forests have longevities strongly influenced by the forest rotation length^14,15, while uneven-aged forestry may allow trees to age closer towards their full potential longevity, depending on the strategy being used^15,16. To illustrate this effect, in Europe the sampled sites² were unevenly distributed among different forest management regimes (Fig. 1c), with the most common regimes falling under combined-objective management (29%) and close-to-nature management (27%), followed by intensive forestry (8%). Therefore, different forest management regimes have the potential to affect site longevity estimates at a broad scale. Different group sizes between management regimes strongly challenge comparisons in longevity between regimes (Fig. 1c), therefore it is not possible to compare the longevity of primary versus managed forests with the available data. Consequently, there remains an unquantified risk that forest management affects the realized longevity of trees, which can affect climate-longevity relationships. This issue likely applies at a global scale because there is evident, and variable, forest management within numerous continents (e.g., Asia, Central America, North America, South America^{17,18,19,20,21,22,23,24,25}). Indeed, at least 59% of forests globally ( > 2000 AD) have forest integrities substantially (medium-high) affected by human activity²⁶. For example, in North America (56% of sites in Gao et al.²), natural fire regimes can be interrupted by forest management in fire-prone forests²⁷, timber harvest is among the dominant disturbance regimes in the northeastern United States²⁸, and forest management drives forest variability outside its natural range in boreal forests of eastern North America²¹. While global forest management regimes have limited data availability, they can vary substantially across countries and continents, which potentially obscures climate-longevity relationships in unexpected ways.

We suspect that the harvesting regime of managed forests, which is spatially dependent, confounds conclusions made between climate and tree longevity. To account for its effect, one could ideally map global forest management regimes, however data availability is limited. In Europe, where such data are available, evidence shows a strong spatial dependence in the location of primary forests⁸ and different management regimes^9,11 (Fig. 1). Since we observe a profound, variable role of forest management at continent scale, there is a strong risk that differences in management regimes between continents affect climate-longevity relationships at the scale investigated by Gao et al.². We therefore highlight a potentially profound, yet unquantified, role of forest management on spatially-inferred climate-longevity relationships by Gao et al.². Specifically, there is a risk that it challenges the conclusion that arid-adapted trees had significantly higher longevity, in relation to their conservative growth strategies. Furthermore, trees in arid environments are generally less productive than trees in wetter environments²⁹, making them less feasible and less sustainable in forestry^30,31, and this could have promoted their longevities.

The number of years measured from a sample is dependent on the research question, the size of the tree, the workforce available, and the tree coring equipment or sample retrieval approach. In Gao et al.² it was assumed that the pith offset was prioritised when feasible during sampling for the data collected in the ITRDB. We demonstrate that based on the same selection criteria, a substantial issue from unquantified pith offsets in the data remains. To minimize this effect, an example of optimal sampling includes extracting three cores per tree and measuring ring widths from the most pristine core that reached closest to the pith³². Even with this ideal sampling design, cores still had a pith offset, and tree age was calculated using not only the length of the core but also the estimated number of missing rings. Therefore, even optimally sampled tree ring measurements reported in the ITRDB can have pith offsets and, on top of this, they may not have been measured to find tree age, making them even more prone to misinterpretation.

We retrieved ITRDB data (last accessed: 16-12-2025) and applied the selection criteria described by Gao et al.². Our replicated dataset represented 74% of the sites in their study³³. To test series for potential completeness, we used the same underlying approach that Gao et al.^2,34 implemented to standardize ring width data, which was a fitted exponential growth model that was supplemented by a fitted linear model if the exponential model did not converge^2,34. This provides an approximate pith offset estimation, and it may not always represent trees with suppressed growth during their early years, for example those in primary closed-canopy forests^35,36. Nonetheless, if trees with suppressed growth during early years are underrepresented, then it only means that the pith offset severity is underestimated, because it indicates that more missing years are needed to reach the minimum growth threshold. While 45% of series unlikely had pith offsets, the remaining 55% were prone to missing rings, and most of these were likely to have pith offsets lasting 30 years or longer (95%) which may even reach 100 years or longer (86%) (Fig. 2). This suggests substantial underestimation of pith offsets in Gao et al.², severely challenging the validity of their growth rates at specific ages (25 and 50 years), which are notably sensitive to missing years. When we replicated their analysis, there was 52% of tree replicate loss for 25 years when the pith offset was capped at either 30 years or 100 years, and 50% tree replicate loss for growth rate estimates at 50 years, when the pith offset was capped at 100 years. These losses affected 99% of the sites we tested and had a biased effect on different site-level juvenile growth rate estimates. While this unlikely affects relationships between mean growth rates and longevity (Fig 3 [a,d] in Gao et al.²), it strongly challenges the reliability of relationships between juvenile growth rates and longevity (Fig 3 [b,c,e,f] in Gao et al.²). For example, since age-related trends were removed from growth rates by calculating Ring Width Index (RWI), any change in year can misrepresent the growth at 25 or 50 years, and there is a substantial range of potential error even within a pith offset of only 30 years, in trees estimated to have no pith offset (i.e., trees that can be tested for pith offset effects; Fig. 2c). Therefore, there is a risk that one cannot reliably conclude from the available evidence in Gao et al.² that climate stress during the juvenile period promotes tree longevity. Likewise, it may not be possible to infer from the data and analyses in Gao et al.² that trees typically maximise their lifespan by slower juvenile growth. Finally, based on the dataset available, it remains uncertain how reliable the conclusion by Gao et al.² is, that early-life drought exposure contributes to an outstanding longevity in trees adapted to arid environments. While our pith offset estimates are approximate because measured pith offsets are not available for this dataset, it is not available for Gao et al.² either, meaning that their early-life growth conclusions are subject to the same risk of not being supported nor refuted based on their presented data and analyses. Without pith offset data, it is only possible to speculate what effects the pith offsets may have on conclusions.

When visualising potential pith offsets, we additionally identified that different tree ring width series covered varying time periods that did not always overlap (Fig. 2a), meaning that they represented different climatic times. Consequently, the different age classes defined by Gao et al.² (mature, old, ancient) were differently represented by the time period for key climatic variables; Mean Annual Precipitation and Mean Annual Temperature (1970–2000). Demonstrating this, 20% of mature tree series ended before 1970, while notably larger percentages of old (31%) and ancient (43%) tree series had ended before 1970. While this can confound spatial climatic interpretations related to site longevity estimates, our tests confirmed that results by Gao et al. are robust to this issue.

To guide future use of ITRDB for longevity estimates in old-growth forests, we recommend the following:

• The dataset must be screened to identify old growth forests. When relating climate to tree longevity, it is very important to either have higher caution with managed forests or incorporate their management regime into data analyses.

• Accommodate analyses for the uncertainties in pith offsets from wood samples in the ITRDB. For example, age estimates at early years (e.g., 25 & 50) are especially prone to the consequences of pith offset errors.

• If comparing longevity estimates to climate, it is important to consider temporally restraining the dataset, or adjusting analyses, in case of mis-representing the temporally variable nature of climate. Furthermore, a space-for-time approach can be misleading depending on the response³⁷, requiring closer examination of space-for-time approaches to tree growth rates and longevity.

In conclusion, Gao et al.² prepared an expansive analysis, showcasing the powerful insights that can be made from ITRDB. Here, we highlight that the diversity of data in the ITRDB challenges the reliability of conclusions surrounding climate-longevity relationships and juvenile growth rate estimates. This highlights the need for extensive care when interpreting climate-longevity relationships and juvenile growth rate estimates, and future data analyses. Nonetheless, our scientific comment encourages the exploration of further analyses that responsibly expand the horizons of analysing data from ITRDB.