The growing season was predicted to start earlier during warm-dry years and later during cool-wet years compared to mean climatology (Fig. 4c, d). Results suggested a widespread shift toward earlier cessation of the growing season during warm-dry years in the second half of the 21st century, particularly at sites with low climatic water balance and high-emission scenarios. This was due to increasing drought-driven growth deficits and increased frequency of zero growth rates simulated in summer and autumn during warm-dry years (Fig. 4a, b; Supplementary Fig. 10). The summer growth deficit was partly alleviated during cool-wet years compared to the mean climatology (Fig. 5, Supplementary Fig. 12).
The simulations highlighted a trade-off between the extension of the growing season and summer drought stress toward the end of the 21 century and their net effects on annual growth increments of temperate tree species in Central, Eastern, and Southeastern Europe. Current cambial activity in the region is limited mainly by low temperatures as indicated by prevailing cold-driven growth deficits. All forecasts agreed on the alleviation of cold limitation as climate change progresses, though with divergent growth trajectories based on rates of growing season extension and summer drought stress. Tree growth might become increasingly limited by the reduced summer growth rate due to amplified drought, mainly in exceptionally warm-dry years. These negative impacts might be offset or even outweighed by the extension of the growing season due to earlier cambial reactivation and higher growth rates in spring and autumn. Consequently, the simulations forecast a gradual transition from the unimodal growth pattern into a long growing season with right-skewed or even bimodal intra-annual growth dynamics at dry sites. We observed systematic variation in the net effects of shifting kinetics and phenology along the gradient of water availability, primarily reflecting the pace and seasonality of warming throughout the 21 century and the occurrence of stochastic warm-dry climatic extremes.
Simulated annual growth rates showed a high similarity between low-emission (SSP1-2.6, SSP2-4.5) and high-emission (SSP3-7.0, SSP5-8.5) scenarios of mean climatology until 2040-2059. Mean predicted tree-ring width indices during this period remained stable or slightly increased compared to the 1961-2020 baseline period. They were similar to values from 2005-2010, i.e., relatively wetter years compared to the most recent decade (2010s) characterized by amplified drought stress. This suggests that temperate forests might keep benefiting from prevailingly increasing growth trends for the next few decades, if climatic extremes, including drought spells, are absent or occur only rarely.
Forecasts based on mean climatology from high and low-emission scenarios diverged during the second half of the 21 century. While low-emission scenarios predicted annual growth rates fluctuating close to the baseline mean, the high-emission scenarios associated with a rapid temperature increase produced unprecedented trends in simulated tree-ring widths. Notably, both conifers and broadleaves at the dry edge of climatic space captured in our study are expected to substantially reduce growth compared to the baseline under the SSP5-8.5 and SSP3-7.0 scenarios. The forecast reductions in tree-ring widths are larger than the mean growth declines simulated for the same species using a similar framework in dry forests in Northeastern Spain. This suggests that forests from temperate Europe might be more vulnerable to drought stress compared to the Mediterranean due to lower resilience after drought events.
In contrast to growth declines simulated at dry sites, humid edges of coniferous species distribution, mainly of Picea sp., are expected to profit from high-emission scenarios of climate warming. By the end of the century, the mean annual growth of these mostly high-elevation stands might accelerate for almost 50 % of the baseline mean if climate follows the SSP5-8.5 scenario and mean climatology. This is in line with currently observed positive growth trends and an increase in the productivity of cold ecosystems across the globe. Such intensification of the growth rate might challenge mountain forestry in Central-Southeast Europe by accelerating stand dynamics and tree turnover, due to either a negative feedback between tree growth rate and lifespan or a higher sensitivity of fast-growing trees to disturbances. Consequently, faster growth can, counterintuitively, lead to shortened carbon residence times in mountain forests. Indeed, high-emission SSP scenarios might cause continuously declining forest growth at dry sites and benefit humid forests with fast turnover. The diverging growth trajectories across the landscape might pose a significant challenge to future forest and landscape management.
Predicted annual growth rates during future climatic extremes revealed more severe growth loss at dry sites under warm-dry spells compared to mean climatology. Notably, simulated tree-ring widths were between 41% and 70% narrower compared to baseline mean during warm-dry years of the 2080-2099 period at sites with a current climatic water balance of less than 200 mm per year. These results are alarming considering previous empirical observations of forest dieback triggered by reduced growth during dry spells in Palearctic temperate forests. Integrating tree-ring formation models, such as VS-Lite, with models simulating stand processes is essential for understanding how projected growth reduction contributes to the risk of tree die-off. Although cool-wet years may alleviate drought stress and increase growth at dry sites during the 2080-2099 period, forecasts for both warm-dry and cool-wet extremes led to reduced growth at humid sites. This illustrates, that humid sites, i.e., mostly mountain forests, might show future sensitivity to both summer droughts and cool growing seasons.
Annual growth rates represent a product of growing season duration and rate of wood formation throughout the year. Increasing air temperature usually stimulates the growing season duration but indirectly limits summer growth kinetics through the moisture availability. Consequently, the forecast growth trends primarily reflected the capacity of the growing season extension to offset reduced summer growth rates. According to simulations for the 1961-2020 baseline period, the mean duration of the growing season currently varies between approximately three months at humid sites and six months at dry sites. Moreover, most of the simulated growth deficits were due to cold rather than drought stress, and the simulated growth cessation mostly occurred due to cold rather than drought-limitation (Supplementary Fig. 10). This shows that climate warming might promote an extension of the growing season to some extent under warmer but not drier climates. Accordingly, simulating the earlier start of the growing season outweighed the aggravating summer drought stress at most sites regardless of the SSP scenario considering the mean climatology of the 2040-2059 period. The mechanistic representation of growth phenology in the VS-Lite model might be a reason for slightly better growth trends predicted till the 2050s in our study compared with statistical forecasts based on seasonal climatic means.
From 2040-2059 onwards, the net effect of the growing season extension and drought-driven reduction of growth kinetics depended on the rate of climate change. While the simulated intra-annual growth patterns for mean climatology under the SSP1-2.6 scenario were mostly stable during the second half of the 21 century, they showed systematic shifts under high-emission forecasts. Scenario SSP5-8.5 predicted growing seasons to be on average 1.44 months longer at the end of the century compared to the baseline period, mainly due to earlier spring growth onset (Fig. 4c, d). However, simulations suggest that such a pace of growing season extension and associated acceleration of growth in spring and autumn will not be sufficient to compensate for steep summer reductions of growth kinetics at dry sites. Accordingly, the intra-annual growth dynamics predicted for the mean climatology at the end of the century in low and middle elevations were characterized by a long growing season but low growth rates, with a local minimum in summer and potentially two growth peaks in spring and autumn. The phenomenon of right-skewed or bimodal growth over the year is characteristic for seasonally dry environments and might represent an adaptation to temporarily withstand summer drought that is typical of Mediterranean climates. However, autumn growth reactivation leading to bimodality is known to be facultative, i.e., it might occur irregularly in time and across space depending on precipitation seasonality and growth plasticity of a given species. Notably, the autumn growth peak at dry sites was diminished in simulations for warm-dry years at the end of the 21 century and high-emission scenarios. As a consequence, predicted dates for the end of the growing season were ambiguous along the humidity gradient due to the frequent fading of the autumn growth peak at dry sites. This highlights the limited ability of phenological shifts and accelerated autumn growth to offset drought stress in a long-term.
The VS-Lite model simulated annual and monthly growth increments with a high coherence with empirical datasets of tree-ring width chronologies, dendrometer data, and NDVI series of canopy-forming species in the baseline period. This confirms its capability to approximate climatic drivers of stem growth, cambial phenology, and canopy greenness across large tree-ring networks. However, how tightly the growth will agree with the model in the forecasting period might depend on species-specific growth plasticity, i.e., the physiological ability to shift from a unimodal growth pattern with a short growing season common in the baseline period into longer growth with multiple peaks predicted for the end of the century. Although most species considered in our study have temperate or boreal ranges with the prevalence of unimodal growth, observed growth patterns have demonstrated their capability to reduce growth in summer and accelerate in autumn under seasonally dry climates. For instance, growth bimodality has been reported during recent dry years using dendrometers or xylogenesis monitoring primarily for Pinus sylvestris and Picea abies. Moreover, all major conifers including P. sylvestris, P. abies, and Abies alba frequently form intra-annual density fluctuations within the latewood, which are deemed to evidence an acceleration of autumn growth in response to precipitation after a dry summer. Available dendrometer data from Central Europe show common growth multimodality for ring-porous broadleaves including Quercus robur, Quercus petraea, and Fraxinus excelsior although the growth of diffuse-porous Fagus sylvatica is rather unimodal due to photoperiod constraints of its phenology. Accordingly, a significant proportion of temperate species, particularly widespread conifers and oaks but not diffuse-porous broadleaves, should be capable of tracking predicted bimodal growth pattern. The growth seasonality might, therefore, become an important competitive trait under warmer and drier climate.
Our study is based on a new regional dataset of 2013 tree-ring width chronologies. Although the majority of these sites are within the core of the current climatic niche of the investigated genera in Europe, the coldest (e.g., Scandinavian coniferous stands) and driest (e.g., Western Mediterranean oak stands) continental margins are not represented (Fig. 1b), which restricts the generalization of our findings to the temperate zone of Central, Eastern, and Southeastern Europe. Correlations between observed and simulated chronologies peaked for conifers and Quercus sp. but were the lowest for diffuse-porous broadleaves like Fagus sp. This might reflect the spatial distribution of individual sites in our study region, where conifers and Quercus sp. often occupy climatic margins of forest distribution, including treelines and dry lowlands, while Fagus sp. forms forests in medium elevations. Moreover, better performance for coniferous sites might be a legacy of the VS-Lite and Vaganov-Shashkin models originally designed for simulating the growth of boreal conifers.
Numerical growth forecasts assume temporal stationarity of climate-growth interactions calibrated in the baseline period. The stationarity of the VS-Lite model was shown to be high for major Palearctic conifer species and this was confirmed for our diverse dataset using an independent trial (Supplementary Fig. 4). Our predictions capture robust long-term growth trends expected till the end of the 21 century but ignore year-to-year variation beyond the mean and extreme climatologies for each bi-decadal period. By restricting our growth forecasts to mean, cool-wet, and warm-dry years we aimed to reduce the effects of inter-annual uncertainty of climatic models driven by the stochastic nature of the climatic system.
The simulated phenological shifts might have been affected by the monthly temporal resolution of the VS-Lite model. For instance, autumn integral growth rates were slightly overestimated compared with dendrometer records (Fig. 2b). Wood formation models operating on daily temporal resolution might be better suited to simulate summer growth quiescence and short-term autumn reactivation under seasonally dry conditions. However, the mean rate of growing season extension predicted by the end of the 21 century under high-emission scenarios, 0.39 days per year ( ≈ 1.44 months over 110 years), is similar to currently observed phenological shifts in Central Europe, supporting the good performance of VS-Lite for estimating growth phenology. Further studies should test improvements in phenological mechanisms of the VS-Lite model, including effects of winter conditions on growth phenology through chilling-forcing interactions, risks of late-frost damage after cambial reactivation, or non-climatic drivers of growth phenology. Moreover, VS-Lite model might benefit from incorporating additional drivers of wood formation like vapor pressure deficit which has recently been successfully implemented in the daily Vaganov-Shashkin model, although its predictive power for growth at a monthly scale still needs to be tested. Finally, the empirical model of tree-ring growth used in our study captured non-linear but fairly continuous shifts in phenology and intra-annual growth patterns in a response to changing climate. However, sites where climate change will trigger stochastic events including increased mortality might experience more dramatic and abrupt shifts of both intra-annual and inter-annual growth patterns. Accordingly, our forecasts of intra-annual growth at the tree-ring level may be less reliable at sites affected by stand-replacing forest disturbances in the future.
Shifting growth seasonality from summer to spring and autumn as predicted by our model for high-emission scenarios might have cascading impacts on forest functioning. Empirical evidence suggests that the prolongation of the growing season stimulates spring carbon sequestration but respiration outweighs increased assimilation in autumn. In consequence, a significant extension of the growing season in temperate forests of the Northern Hemisphere together with reduced summer growth rates as predicted in our study might offset the phase and alter the amplitude of the annual cycle of atmospheric CO concentrations. Similar to the carbon sequestration, the seasonality of other processes linked to plant phenology and crucial for society including the water cycle, soil development, and landscape albedo might to some extent adjust to the forecast shifts in wood phenology. Accordingly, forecasts of growth phenology and kinetics provided by our study contribute to understanding the broader implications of future shifts in the seasonality of temperate forests.