Forest production efficiency increases with growth temperature

The study investigates how forest production efficiency (FPE)—the fraction of gross primary production (GPP) allocated to net production (NPP) or biomass production (BP)—varies with stand age and climate. The central questions are whether FPE systematically relates to mean annual temperature (MAT), total annual precipitation (TAP), and absolute latitude (|lat|), and how stand age modulates FPE. This is important for understanding forest carbon sequestration and improving ecosystem model projections, as many models assume respiration increases more steeply with temperature than photosynthesis, implying lower FPE in warm climates. The authors assemble a global dataset to test these relationships and compare empirical patterns with outputs from state-of-the-art Dynamic Global Vegetation Models (DGVMs).

Prior work has debated whether NPP is a constant fraction of GPP (Waring’s hypothesis), with mixed evidence and recognition of variability across sites and ages. Previous studies reported age- or size-related declines in GPP, NPP, and efficiency metrics (CUE or BPE). Respiratory thermal acclimation literature indicates maintenance respiration adjusts to growth temperature, weakening the instantaneous temperature dependence. Studies have suggested nutrient availability and belowground carbon allocation regulate forest carbon balance. An emergent-constraint analysis (He et al.) inferred higher CUE at high latitudes and declines with MAT, aligning with model assumptions that respiration increases more than GPP with temperature, but its validity depends on models correctly linking site-specific to global CUE. The present study revisits these topics using an expanded empirical dataset and explicit multivariate controls.

Data: The authors compiled 244 records (197 BPE, 47 CUE) from over 300 peer-reviewed studies at >100 forest sites worldwide (1995–2015 predominance), covering boreal, temperate, and tropical forests and multiple forest types (DBF, EBF, ENF, mixed). Environmental variables included stand age (n=204; ~5–500 years), MAT (n=230; −6.5 to 27.1 °C), TAP (n=232; ~125–3500 mm yr−1), latitude/longitude, elevation, LAI, treatments, and disturbances.

Definitions and measurements: GPP is annual gross carbon fixation by autotrophs. NPP is GPP minus autotrophic respiration and other losses; BP is biomass production. Unaccounted “occult” fluxes (e.g., NSC changes, root exudates, mycorrhizal subsidies, BVOCs) can cause CUE to exceed BPE. GPP, NPP, and BP estimates were grouped by method: biometric (n=13 for GPP; n=200 for NPP/BP), micrometeorological eddy covariance with partitioning (n=98; n=4), model-based (n=53; n=24), and chamber-based scaling (n=73; n=9). For analysis, modelling-derived CUE values and any estimates relying on fixed fractions were excluded.

Uncertainty: Following Luyssaert et al., gross uncertainties decreased with |lat| and were reduced by method-specific factors (e.g., 0.3 for micromet and biometric where applicable; 0.6 for models; 0.8 for scaling). Uncertainties for CUE/BPE propagated from component uncertainties.

Statistical analysis: Because CUE and BPE did not significantly differ at co-located sites, both were combined as FPE. Only records with complete FPE, MAT, age, TAP, and |lat| were retained (142 observations). Mixed-effects multiple linear regression was fit using lmer (R lme4), with fixed effects of MAT, age, TAP, and |lat|, and a random intercept for GPP method class (micromet vs. scaling/biometric): FPE = β0 + β1·MAT + β2·age + β3·TAP + β4·|lat| + ηGPPmethod + ε. Multicollinearity was acceptable (VIF 1.1–3.8). Residuals were tested for normality (Anderson–Darling); models lacking the GPP-method random effect failed normality and were rejected. Model selection used ANOVA comparisons across reduced models; the full additive model could not be reduced at 5% significance. A log-transformed multiplicative variant confirmed directions of effects and had slightly better AIC/BIC but similar adjusted R2 (~0.32 vs. 0.31).

Model–data comparison: Outputs from eight DGVMs in TRENDY v7 (ISAM, JULES, LPJ-GUESS, CABLE-POP, ORCHIDEE, ORCHIDEE-CNP, JSBACH, SDGVM) were analyzed for 1995–2015 means. Because models lack stand age, a modified regression with MAT, TAP, and |lat| and a random intercept by model was attempted; residual normality conditions were not met and no consistent emergent relationship was found. Slopes of modelled CUE vs. MAT were estimated for each model and compared with the empirical slope.

• FPE increased significantly with growth temperature (MAT), total annual precipitation (TAP), and absolute latitude (|lat|), and decreased with stand age.
• Mixed-effects model (n=142) parameter estimates (Table 1): MAT slope ≈ +0.0060 per °C (p=0.016); age slope ≈ −0.00038 per year (p=0.0013); TAP slope ≈ 6.8×10⁻⁵ per mm yr−1 (p=0.0014); |lat| slope ≈ +0.0039 per degree (p=0.016). Intercept ≈ 0.19 (n.s.). Random intercepts indicated micrometeorological GPP estimates yielded systematically higher GPP (thus lower FPE) than scaling/biometric. The model explained ~30–31% of variance (Pearson’s and Spearman’s R² ≈ 0.31).
• The positive MAT effect contradicts the common model assumption that respiration increases more steeply with temperature than photosynthesis; it implies regulation via acclimation and/or whole-plant allocation strategies that reduce carbon losses at higher growth temperatures.
• Age effect: clear decline in FPE with stand age (slope ≈ −0.0004 yr−1), consistent with prior evidence of reduced production efficiency in older stands.
• Range: Statistically fitted BPE/CUE mostly between ~0.27 and 0.58; 92% of values within physiological bounds 0.2–0.65; values >0.65 occur in some young stands; none <0.2 observed.
• DGVM comparison: All eight TRENDY v7 models showed negative CUE–MAT slopes (e.g., −0.0025 °C−1 for LPJ-GUESS to −0.0098 °C−1 for SDGVM; ensemble mean ≈ −0.005 °C−1), opposite to data-driven slope (+0.006 °C−1). Models matched temperate-region CUE but diverged strongly in boreal and tropical regions and showed discontinuities by biome.
• Precipitation and latitude: Newly documented positive effects on FPE likely reflect increased water availability and radiation regime (longer summer days, diffuse light) enhancing GPP relative to respiration.
• CUE vs. BPE: Where both were available, CUE exceeded BPE in 7 of 13 cases; otherwise, differences were statistically indistinguishable, indicating variable magnitude of unaccounted carbon flows among forests.

The empirical multivariate analysis demonstrates that FPE increases with MAT, TAP, and |lat| and declines with age, addressing the long-standing question of how climate and stand development regulate forest production efficiency. The positive temperature effect runs counter to instantaneous enzyme-kinetic expectations and prevailing model structures but aligns with respiratory thermal acclimation and potential genetic adaptation, which reduce maintenance respiration at higher growth temperatures. Additional mechanisms include adjustments in enzyme quantities/activity, tissue composition, and allocation that minimize respiratory costs. Low FPE in cold climates may reflect higher belowground carbon investment to acquire nutrients in cold soils and costs of frost damage repair.

The positive latitude effect likely captures radiation regime influences (longer daylength, greater diffuse radiation), which can raise GPP efficiency without proportionally raising respiration. Increased precipitation likely enhances stomatal conductance and photosynthesis more than respiration, raising FPE.

Age-related declines likely arise from hydraulic limitations reducing GPP, greater maintenance costs due to increased sapwood volume, declining soil fertility and shifts in allocation and defense with ontogeny. Management (e.g., thinning, rejuvenation) may mitigate declines in efficiency.

Model–data discrepancies indicate that standard respiration formulations (fixed basal rates with fixed Q10/Arrhenius responses) cannot reproduce the observed positive MAT–FPE relationship, leading to implausibly low CUE in warm climates unless ad hoc biome-specific parameter discontinuities are introduced. Incorporating acclimation and whole-plant regulatory feedbacks is necessary for realistic projections of forest carbon balance under warming.

This study provides a global, data-driven demonstration that forest production efficiency increases with growth temperature, precipitation, and absolute latitude and declines with stand age. The results challenge common assumptions embedded in many land ecosystem models, which predict decreasing CUE with temperature, and imply that such models likely overestimate carbon losses in a warming climate. Production efficiency emerges as a relatively conservative yet environment- and age-sensitive quantity that can be used to constrain and guide model development.

Future directions include: integrating thermal acclimation and adaptive respiratory regulation into vegetation models; explicitly representing whole-plant allocation strategies and nutrient acquisition costs across climates; disentangling radiation regime effects from temperature; better quantifying unaccounted carbon fluxes (NSC dynamics, root exudation, mycorrhizal subsidies, BVOCs); and expanding empirical datasets, particularly in understudied tropical and boreal regions and across disturbance and management gradients.

• Limited number of sites with direct NPP measurements (n=31) prevented fitting an independent CUE-only model; CUE and BPE were combined as FPE, potentially masking differences due to unaccounted carbon flows.
• Heterogeneity and uncertainties in GPP/NPP/BP estimation methods; micrometeorological GPP includes ground vegetation, while most biometric estimates pertain to trees only; method biases handled via random effects but may persist.
• Uncertainties estimated via expert-judgment approach and method-specific factors; true errors may differ by site and method.
• Incomplete reporting on disturbances (e.g., fire n=6; other biotic/abiotic events largely unknown), treatments, and site histories; potential confounding effects not fully resolved.
• DGVM comparison lacked stand age as a predictor (models do not represent age explicitly), limiting direct comparability and precluding identical regression structures.
• Spatial coverage uneven (fewer tropical and boreal records relative to temperate), which may influence generalized relationships.
• Assumption that biometric NPP approximates BP due to omission of occult fluxes could bias comparisons between CUE and BPE.