Climate-change-driven growth decline of European beech forests

Global environmental change is affecting ecosystems worldwide, with forests playing a key role in terrestrial water and carbon cycles and services to society. While much research has examined how climate change affects forest distribution and presence, less is known about species’ growth performance across their full ranges. Secondary growth is a strong proxy for total carbon sequestration, making tree-ring width and derived basal area increment (BAI) valuable indicators of vitality. Dendroecological studies have largely provided local to regional insights, but spatio-temporal analyses that cover entire species distributions are scarce, in part due to biases in existing international tree-ring databases that limit coverage of certain taxa, biomes, and trailing-edge populations. To overcome these limitations, the authors assembled a dense, species-specific network of over 780,000 ring-width measurements from 5800 European beech (Fagus sylvatica) trees across 324 sites spanning the species’ geographic and climatic range. Beech is ecologically and socio-economically critical in Europe; its resilience and plasticity under rapid climate change remain uncertain. Using this network, the study analyzes past growth and projects future variability under CMIP6 SSP climate scenarios via a generalized linear mixed-effects model (GLMM), comparing regions and mapping growth under local environmental stresses and disturbances. The goal is to identify recent growth changes and predict 21st century growth trajectories to inform assessments of resilience and guide forest adaptation.

The paper situates its work within several strands of literature: (1) a long-standing focus on predicting forest cover and species distributions under climate change, contrasted with a relative paucity of ecologically based predictions of species’ growth performance; (2) the importance of secondary growth as a proxy for carbon sequestration and tree health; (3) the prevalence of local-to-regional dendroecological analyses and the scarcity of species-wide, spatio-temporal growth studies due to biases and gaps in international dendrochronological datasets; and (4) prior findings on beech growth trends, including reports of recent growth declines linked to warming, drought, and extreme events, as well as studies showing increases or altitude-dependent responses. Methodological differences (e.g., handling age effects, using detrended chronologies or raw increments) have led to mixed conclusions, underscoring the need for a comprehensive, standardized approach across the species’ full range. The study aims to reconcile these findings via a large-scale, uniform BAI-based analysis leveraging a dense tree-ring network.

Tree-ring network: The authors compiled 324 sites of mature, closed-canopy European beech stands across Europe (5.8–28.4°E; 38.8–58.5°N; 1–1900 m a.s.l.), covering the species’ full geographic and climatic ranges (annual precipitation ~500–2000 mm; temperature ~3.8–13.5 °C). Approximately 5800 trees yielded ~780,000 ring-width measurements. Increment cores were dried and sanded following standard procedures. Ring widths were measured to 0.01 mm and crossdated (COFECHA or CooRecorder). Classical detrending was not applied; instead, ring-width series were converted to annual basal area increment (BAI, cm² year⁻¹) using measured diameter at breast height and the bai.out function in R package dplR, which accounts for geometric constraints and facilitates temporal comparisons of mean BAI unaffected by biological age trends.

Climate variables: High-resolution gridded climate data (CHELSA) provided monthly precipitation and maximum/minimum temperature, aggregated to seasonal means from 1901–2016 (previous-year summer through current-year autumn relative to the ring year). Prevailing moisture conditions were quantified by the De Martonne Aridity Index (AI = P/(10+T), with climate types from arid to extremely humid), computed for European grids for 1950–2016.

Predictive growth model (GLMM): A generalized linear mixed-effects model (R, lme4; maximum likelihood with adaptive Gauss-Hermite quadrature) was fitted for 1950–2016 to predict annual log(BAI) of tree j at site i in year t as a function of: AI (log-transformed), latitude (LAT), altitude (ALT), seasonal maximum and minimum temperatures (Tmax, Tmin), and seasonal precipitation (PP; log-transformed), using smoothing functions and including significant two- and three-way interactions among AI, LAT, ALT, Tmax, Tmin, and PP. Variables were standardized prior to modeling. Random effects included a random slope of previous-year basal area (BA) per tree and random intercept by tree code to account for size dependency and repeated measures. Model selection and validation followed AIC-based comparisons and likelihood ratio (χ²) tests against reduced and null models.

Spatial application and hindcast: The fitted GLMM was applied to each grid cell within the EUFORGEN beech distribution to compute annual BAI for 1950–2016. Mean BAI was compared between 1955–1985 and 1986–2016 (mid-1980s marking the onset of strong warming). Percentage growth changes were calculated for each grid cell; maps produced in R.

Future projections: CMIP6 multi-model ensemble means (21 models for temperature, 26 for precipitation) under SSP1-2.6 (optimistic) and SSP5-8.5 (pessimistic) were used. For each scenario and three periods (2020–2050, 2040–2070, 2060–2090), delta-method adjustments were computed as future minus historical (1985–2014) climatological differences and added to CHELSAcruts baseline to obtain projected seasonal climate inputs. Geographic variables and AI were held constant; seasonal climate variables were updated per scenario/period. The GLMM generated six growth-variation scenarios. Applicability domains (AD) were evaluated; predictions outside the historical climate range (1901–2016) were flagged as less reliable.

• Model performance: The full GLMM including moisture (AI), seasonal climate (precipitation and maximum/minimum temperatures), geography (latitude, altitude), interactions, and random effects had the lowest AIC (617,187.1) and explained ~86% of growth variability. Precipitation correlated positively with growth; seasonal temperatures had strong, season-dependent effects and were overall stronger predictors than precipitation. Interactions among AI, latitude, and altitude significantly modulated climate sensitivities.
• Past growth (1955–1985 vs 1986–2016): A widespread growth decline occurred across most of beech’s range, with a clear latitudinal gradient: declines strongest in southern Europe (up to −20%), and increases in northern areas (Sweden, Norway) up to +20%. High growth at low altitudes in NW and central Europe declined in spatial extent in the recent period; low growth persisted at high altitudes in the Alps and Carpathians and at northern/southwestern edges (Sweden, Spain). Overall growth magnitude decreased in the recent period.
• Future projections (relative to 1986–2016 baseline): • SSP1-2.6: By 2020–2050, southern Europe shows reductions up to ~30%, decreasing northward to ~10% to ~0% in central Europe. Growth increases of ~25% projected for mountainous central Europe and ~35% in southern Scandinavia. Patterns persist and intensify modestly through 2040–2070 and 2060–2090, with more accentuated declines in the Balkans and southern Europe and stronger polarization versus Alpine and northern regions. • SSP5-8.5: Substantial declines dominate. For 2020–2050, central Europe exhibits −20% to −30% declines, with some elevated sites affected; southern Europe exceeds −50% (especially 2040–2070). Positive trends remain north of ~55°N and in mountainous central Europe. The southeast-to-northwest gradient, modulated by altitude, intensifies through 2040–2070 and 2060–2090.
• Ecological and economic implications: Projected productivity losses, particularly in southern and parts of central Europe under SSP5-8.5, indicate reduced carbon sink strength of beech forests and heightened risks of hydraulic failure and dieback during “hotter droughts.” Some high-latitude and high-altitude sites may experience growth enhancement even under severe warming.

The GLMM reveals strong geographic variation and a regional growth optimum for beech in mountainous central Europe under current climate, shaped by north–south and northwest–southeast gradients and altitude. Beech tends to be more productive at lower elevations in NW Europe, reflecting phenological and xylogenesis shifts with altitude and latitude. The observed and projected patterns indicate that recent warming and drying have increasingly constrained growth, with a pervasive decline since the 1980s except at northern margins and higher elevations.

The study reconciles mixed findings from prior regional studies by using raw BAI (avoiding detrending artifacts) and a consistent large-scale approach. While drought sensitivity is well documented across Europe, results suggest that elevated temperatures are becoming a primary limiting factor across much of the range, unless accompanied by substantial precipitation increases. Warming increases vapor pressure deficits, induces stomatal closure, elevates water demand, and raises risks of hydraulic failure. Associated processes (defoliation during droughts, extended canopy duration with metabolic constraints, and higher fine-root turnover) further enhance temperature sensitivity. Consequently, without compensatory precipitation increases (as partially seen under SSP1-2.6), growth is projected to decline broadly; the southern distribution edge is particularly vulnerable. These trends challenge expectations of increased terrestrial carbon stocks under climate change where beech dominates and underscore the need for adaptation in forest management.

Uncertainties include the fidelity of future climate projections (especially precipitation) in CMIP6 ensembles, the influence of extreme events (late frosts, heat waves, fires), edaphic factors (nutrient status), and interspecific competition. Applicability domain analyses highlight reduced confidence where future climate exceeds historical variability. Despite these limitations, the consistent large-scale decline patterns and strong model performance provide robust evidence for significant growth reductions under warming scenarios.

Analyzing a continent-wide tree-ring network with a GLMM, the study documents a pervasive decline in European beech growth from 1955 to 2016, with −20% in southern regions and +10–20% increases at northern and high-elevation sites. Future warming exceeding ~1.5 °C is projected to drive widespread growth decreases (−20% to −40%), potentially reaching −50% under hotter, drier conditions (SSP5-8.5), while some high-latitude and mountainous areas may see increases. Given the link between growth decline and elevated mortality risk, and beech’s dominant role in European forests, these results imply a weakening carbon sink and significant ecological and economic impacts. The authors call for immediate consideration of these projections in long-term silvicultural planning and suggest future research to disentangle large-scale drivers (atmospheric circulation, continentality, photoperiod), incorporate extreme events, soil fertility, and biotic interactions, and to refine projections with improved climate and process-based models.

• Climate projection uncertainty: CMIP6 ensemble projections, particularly for precipitation, include multiple sources of error; future climate variability may exceed historical ranges (applicability domain), reducing prediction reliability for some pixels/time periods.
• Omitted processes and factors: Extreme events (late spring frosts, heat waves, fires), soil properties and nutrient availability (N, P, K), and interspecific competition were not explicitly modeled but can significantly influence growth and modulate climate sensitivities.
• Assumptions in projections: Geographic variables and aridity index were held constant for future scenarios; potential future changes in site conditions (e.g., land use, stand structure, CO2 fertilization effects) are not represented.
• Data representativeness: Although the network is dense and range-wide, it focuses on mature, closed-canopy stands and may under-represent managed or disturbed stands; measurement and crossdating uncertainties are possible though minimized by standard protocols.