Biogeographic variation in evergreen conifer needle longevity and impacts on boreal forest carbon cycle projections

Significance

How evergreen tree needle longevity varies from south to north in the boreal biome is poorly quantified and therefore ignored in vegetation and earth system models. This is problematic, because needle longevity translates directly into needle turnover rate and profoundly affects carbon cycling in both nature and computer models. Herein we present data for five widespread boreal conifers, including pines and spruces, from >125 sites along a 2,000-km gradient. For each species, individuals in colder, more northern environments had longer needle life span, highlighting its importance to evergreen ecological success. Incorporating biogeography of needle longevity into a global model improved predictions of forest productivity and carbon cycling and identified specific problems for models that ignore such variability.

Abstract

Leaf life span is an important plant trait associated with interspecific variation in leaf, organismal, and ecosystem processes. We hypothesized that intraspecific variation in gymnosperm needle traits with latitude reflects both selection and acclimation for traits adaptive to the associated temperature and moisture gradient. This hypothesis was supported, because across 127 sites along a 2,160-km gradient in North America individuals of Picea glauca, Picea mariana, Pinus banksiana, and Abies balsamea had longer needle life span and lower tissue nitrogen concentration with decreasing mean annual temperature. Similar patterns were noted for Pinus sylvestris across a north–south gradient in Europe. These differences highlight needle longevity as an adaptive feature important to ecological success of boreal conifers across broad climatic ranges. Additionally, differences in leaf life span directly affect annual foliage turnover rate, which along with needle physiology partially regulates carbon cycling through effects on gross primary production and net canopy carbon export. However, most, if not all, global land surface models parameterize needle longevity of boreal evergreen forests as if it were a constant. We incorporated temperature-dependent needle longevity and %nitrogen, and biomass allocation, into a land surface model, Community Atmosphere Biosphere Land Exchange, to assess their impacts on carbon cycling processes. Incorporating realistic parameterization of these variables improved predictions of canopy leaf area index and gross primary production compared with observations from flux sites. Finally, increasingly low foliage turnover and biomass fraction toward the cold far north indicate that a surprisingly small fraction of new biomass is allocated to foliage under such conditions.

The boreal forest is an enormous terrestrial biome, characterized by long, cold winters, low rates of productivity and nutrient cycling (1–4), and low richness of the dominant tree species. All of these tree species have large geographic ranges (5), and ecological theory suggests that populations should vary both genotypically and phenotypically in response to such variation. In fact, it has been long recognized that terrestrial plants in general, and boreal conifers specifically, exhibit ecotypic population differentiation that reflects influence of site origin on a variety of chemical, morphological, physiological, and allometric traits (6–12). It is also well known that biogeographic variation in the environment influences the phenotypic traits plastically realized by species across their ranges. Both phenotypic and genotypic processes therefore contribute to observed variation in traits across environmental gradients. Hence, across the boreal biome, conditions that lead to slow growth, such as short growing season, cool summer temperatures, and low soil nutrient supply (1, 2, 13), should be reflected by shifts toward more conservative traits (such as long leaf life span and low nutrient concentrations) considered to be advantageous under such conditions (14, 15). However, although evidence that ecotypic variation occurs is abundant (7, 8, 10), comprehensive characterization of ecologically important intraspecific variation across biogeographic gradients is lacking (16).

Global land surface models that simulate carbon cycling of the world’s terrestrial biomes face distinct challenges for both species-poor biomes, such as the boreal forest, and species-rich biomes, such as tropical rain forests. For the latter, the challenge is to sufficiently well parameterize leaf and canopy properties given enormous species diversity within and among sites (e.g., >500 species for <1,000 individuals in a given several-hectare area, with large species turnover at 100-km scale). In contrast, for boreal forests, the task is seemingly easy: Only a small number of species dominate the entire boreal forest in each continent. Moreover, the same genera dominate the circumboreal region. However, although leaf and canopy properties vary with climate and geographic location, this variation is usually ignored in biome-scale or global models, because of a lack of systematic understanding of that variation.

Despite long recognition of ecotypic variation in boreal conifers, certain broad-scale aspects of physiological macroecology remain poorly quantified. One trait known to differ intraspecifically in evergreen conifers with climate and latitude is the needle life span (NL) (synonymous with needle longevity) (17, 18). In general, individuals growing in colder locations tend to have longer NL (17–19), but important patterns and consequences are unknown, such as the following. (i) What is the shape of the NL relation to geographic climate variation? (ii) Are proportional shifts in NL for a given climate gradient similar among species? (iii) Do other leaf traits often associated interspecifically with leaf life span (e.g., refs. 15 and 20) covary with NL intraspecifically in a similar fashion? (iv) What are the consequences of geographic patterns of needle traits to ecosystem and regional carbon cycling? To improve our understanding of these issues, we measured needle traits of the four dominant evergreen conifers from Minnesota to northern Canada, compiled similar data for the dominant evergreen species in Eurasia (Pinus sylvestris), characterized the relations of needle traits with climate variables for all five species, and incorporated such findings into a land surface model [Community Atmosphere Biosphere Land Exchange (CABLE)] that simulated biome-wide carbon and nitrogen cycling.

We sampled foliage from naturally grown trees (range of height 2.5–5 m) at 127 sites ranging from Minnesota to Ontario, Manitoba, Saskatchewan, Alberta, and the Northwest Territories, in central Canada. We sampled sunlit upper canopy branches of trees ranging from 2.5 to 5 m in height, to standardize tree size and age across sites, as well as the light environment of the sampled branches. The latter was of paramount concern, because needle longevity in the crown of a conifer increases systematically with shading (by as much as 50–60% for spruce and fir). The four species were present (and sampled) at 83 (Pinus banksiana), 50 (Picea mariana), 45 (Picea glauca), and 21 (Abies balsamea) of the sites, respectively. We also use a literature compilation for Pinus sylvestris across northern Fenno-Scandinavia to extend our analysis beyond North America. Because several climatic factors covary with latitude and are hypothesized to be the mechanism behind the latitudinal pattern, we focus our analyses and interpretation on climate, not latitude. Additionally, it is difficult to ascertain which of the related climate metrics (including annual and seasonal measures of temperature and precipitation) are responsible for the observed latitudinal patterns (Methods and Supporting Information). Because mean annual temperature (MAT) was identified in the model selection process as the climate metric that explained the most variance in NL, we use it in the results presented herein. However, we discuss the ways in which multiple aspects of climate, and associated soil resource availability, can combine to influence these results, below and in Supporting Information.

Results

For all five species, NL increases with increasing latitude (Fig. S1) and decreases with MAT (Fig. 1). The relationship is strongest for both Picea species and P. sylvestris and weakest in P. banksiana. The biogeographic range in NL is particularly pronounced in the Picea species—ranging from ∼6 y at their southern margins (∼46°N) to 15 y at the most northern locations sampled (∼62–63°N), which are not far from their northern limit in that part of central Canada. Abies has almost as strong a shift in NL per unit shift in MAT as the Picea species but has a narrower range, and thus NL varies from 5 y at its southern limit (∼46°N) to 9 y at its northern (∼55°N) limit. P. sylvestris shows a similar proportional shift (more than doubling NL from 3.5 to 7.0 y from 58 to 69°N) as the North America Picea species. In contrast, P. banksiana, which has as large a geographic range as the Picea species, has a much weaker dependence of NL on MAT—it varies from 4.0 y in Minnesota to 5.5 in northern Canada. It is not clear why P. banksiana has such a noisy and modest relation of NL to MAT, compared with the other NA species or its European congener. It is possible that further to the south in its range it is relegated by competition to particularly poor sites that support slow growth and therefore longer NL than would otherwise occur, but data to test this hypothesis are not available. Regardless, there is a strong signal that NL decreases with MAT for these five species that are among the most dominant species in much of North American and Eurasian boreal forests.

Fig. 1.

Mean NL (years) for four North American and one European evergreen conifer species in relation to MAT along a large latitudinal gradient: A. balsamea (logNL = 0.860–0.041*MAT; P = 0.0083, R² = 0.34, n = 19), P. glauca (logNL = 0.911–0.044*MAT; P < 0.0001, R² = 0.65, n = 46), P. banksiana (logNL = 0.650−0.014*MAT; P = 0.0052, R² = 0.10, n = 80), P. mariana (logNL = 0.932–0.056*MAT; P < 0.0001, R² = 0.66, n = 51), and P. sylvestris (logNL = 0.792–0.040*MAT; P < 0.0001, R² = 0.65, n = 79).

Needle %N increases significantly with MAT in P. banksiana, P. glauca, and P. sylvestris, with nonsignificant relations in the other two species. For data pooled across all five species, there are significant (P < 0.0001) relationships between %N and both NL (Fig. S2) and MAT. This indicates that boreal evergreen forests dominated by individuals with long NL (owing to prevalence of long NL species such as Picea, or to adaptive response to NL to cold temperatures, or both) also have low needle %N.

In addition to illuminating intraspecific ecological strategies, the biogeographic patterns documented here have important implications for ecosystem-scale function. We used the CABLE land surface model (21, 22) to explore the impact of biogeographic variation in needle traits on evergreen boreal forest C cycling (Methods and Table 1). This exploration is important because most, perhaps all, widely recognized global land surface models consider all boreal forest as a single plant functional type (PFT) with a single set of trait values, and land surface models in general are inadequate with respect to biomass allocation (23, 24). To our knowledge, leading models such as Biome-BioGeochemical Cycles (Biome-BCG), Community Land Model version 4 (CLM-4), Lund–Potsdam–Jena Dynamic Global Model (LPJ) (and its offspring), and others do not include a temperate dependence of needle turnover rate (or biomass allocation) within the boreal evergreen PFT. Because CABLE does not recognize species differences, we used a generalized algorithm to modify NL with MAT in a fashion that averages responses across the species available (Methods).

Table 1.

Versions used to assess CABLE model performance for boreal evergreen conifer forests

Version	Allocation	Needle life span	%N
Default	Constant	Constant(3.76)	Constant(1.23)
cALLcNLc%N	Constant	Constant(6.02)	Constant(1.23)
cALLfNLc%N	Constant	NL(MAT)	Constant(1.23)
fALLcNLc%N	ALL(MAT)	Constant(6.02)	Constant(1.23)
fALLfNLc%N	ALL(MAT)	NL(MAT)	Constant(1.23)
cALLfNLf%N	Constant	NL(MAT)	%N(MAT)
fALLfNLf%N	ALL(MAT)	NL(MAT)	%N(MAT)

The default version of CABLE represents the prior version (22, 33). Allocation (ALL) was either the default constant or a continuous function of mean annual temperature (MAT) (25). Needle life span (NL) was either constant at the prior default value (3.76), constant at the mean value (6.02) for the observed data, or modeled as a continuous function of MAT. Needle %N was either constant at the prior default value (1.23) or a continuous function of NL, as adjusted using a canopy depth extinction coefficient.

Structure, parameterizations, and functional routines of CABLE are roughly similar to many other global land surface models, such as those mentioned above. Of relevance to this study, its parameterizations of PFTs and their key traits are similar to many other global land surface models. Thus, the examination herein of how changes in parameters and functions alter model performance should be generally reflective and not exclusive to CABLE.

Temperature sensitivity of biomass allocation is likely an important but poorly understood regulator of C cycling in boreal forests (25), and models at present do not predict allocation well, even ignoring biogeographic variation (23, 24). Standing biomass distribution to leaves, stems, and roots is a joint function of turnover rate and initial allocation of new biomass. Given this, and that our goal includes assessing ecosystem impacts of changes in foliage turnover rate, we jointly examine impacts of temperature-sensitive biomass distribution (26) on CABLE predictions. Our modification of the biomass distribution in CABLE is based on analyses that show that foliage and roots represent a greater and smaller fraction, respectively, of total biomass in forests in increasingly cold environments (26), which is consistent with independent data showing proportionally greater carbon allocation below ground for Picea in colder, drier habitats (25). Incorporating temperature-dependent biomass partitioning yields the result that allocation of new biomass (NPP) to needles must be markedly low in the cold far north (Fig. 2), given that NL increases in colder, higher latitudes (Fig. 1), but the fraction of standing biomass located in needles is low at low MAT (26) (Fig. S3).

Fig. 2.

Representation of the fraction of new total biomass allocated to foliage in relation to MAT for evergreen conifer forests, for four above-ground biomass classes (50–300 Mg/ha). These curves result from the MAT-dependency of needle life span (Fig. 1) and fractional standing biomass distribution to foliage (Fig. S1). The vertical dashed lines represent the range of data for which they were developed. The allocation fraction is mathematically derived as a function of life spans of leaf, wood, and root; biomass fraction of leaf, wood, and root; and their relations with MAT and total biomass. The idea of calculating allocation fraction in this fashion is from a semianalytic solution to accelerate biogeochemical model spin-up (39).

The CABLE model as previously published (21) (called default hereafter) assumed NL of 3.76 y, root life span of 18 y (this includes and averages across fine roots, coarse roots, and woody roots), and wood life span of 70 y, needle %N of 1.23%, and a constant ratio of allocation of new biomass to needles, stems, and roots for evergreen needle leaf forest (21). To assess impacts of incorporating realistic biogeographic variation of biomass allocation (ALL), NL, and needle %N into the model, we compare these default results to permutations (Table 1) that incorporate these variables either as constants (cALLcNLc%N) or as functions of MAT (fALLfNLf%N), in various combinations. The constant values for ALL and %N are from the default CABLE, the constant value of NL is either the default value or the average across all sites for our field data, and the functions that vary with MAT are from our observations (Figs. 1 and 2 and Figs. S2 and S3), although for %N this is indirect, because we predict %N from temperature-dependent NL, to retain their covariance. We map these predictions spatially (Figs. 3 and 4) and also compare predictions for four north–south climate zones to values of canopy leaf area index (LAI) and gross primary production (GPP) from flux sites published by Luyssaert et al. (3) (Fig. 5). In the figures, moving from top to bottom (Figs. 3 and 4) or left to right (Fig. 5) generally moves from the default model with constant input parameter values, to new versions with one or two modified parameters, to the model where values of allocation, NL, and needle %N are all biogeographically varied.

Fig. 3.

Patterns of simulated GPP across the boreal forest zone in vegetation under seven model versions, representing different model permutations with panels corresponding to modifications described in Table 1.

Fig. 4.

Patterns of simulated LAI across the boreal forest zone in vegetation under seven model versions, representing different model permutations with panels corresponding to modifications described in Table 1.

Fig. 5.

Comparison of GPP (A) and LAI (B) predictions from the CABLE land surface to observed FLUXNET data from Luyssaert et al. (3). The comparisons are shown for the predictions under seven versions, representing different model permutations (Table 1).

The default version of CABLE successfully predicts the north–south direction of LAI and GPP (Figs. 3–5) and slightly overpredicts GPP while underpredicting LAI, except in the coldest zone (Fig. 5). If a more realistic mean NL (6.02 y) as estimated from our data is used as a constant NL value everywhere (cALLcNLc%N), the model still gets the direction of LAI and GPP correct, but overpredicts GPP and LAI. This directly results from model logic—with a constant allocation routine and a constant NL, greater NL results in greater LAI.

Incorporating our general temperature-dependent relationship for NL in the model instead of a constant value (cALLfNLc%N) also overpredicts GPP and LAI (Fig. 5) but additionally also fails to capture the directional signatures for GPP and LAI—with GPP stable and LAI decreasing with MAT. This is because with a constant allocation routine a realistically longer NL in the north unrealistically leads to higher LAI (and consequently higher GPP than would otherwise occur), despite a shorter growing season and colder temperatures.

Incorporating MAT-dependent biomass allocation (and with NL constant, 6.02 y, fALLcNLc%N) also overpredicts GPP and LAI, but gets their directional variation correct (they are all higher toward the warmer south). The permutation (fALLfNLc%N) that includes temperature-dependent NL as well as temperature-dependent biomass allocation makes predictions that are similar. It is likely that the four new CABLE versions presented above all overpredict GPP in part because the constant %N (1.23%) used in the model is higher than the empirically observed average (∼0.9%), and LAI and %N jointly influence productivity in the CABLE model, just as observed empirically for both maximum instantaneous GPP and annual NPP (4). The default CABLE predictions may approximate GPP (Fig. 5) accurately because unrealistically high %N offsets unrealistically low LAI.

The model permutation (cALLfNLf%N) in which both temperature-dependent NL and NL-dependent %N are incorporated in the model (but with the default allocation scheme) does improve predictions on average compared with the model with temperature-dependent NL but with the default constant %N, but the directionality of LAI and GPP are not captured. The model (fALLfNLf%N) that incorporates both of the temperature-dependent functions (for NL and biomass partitioning) and the NL-dependent %N seems to do well in matching not only the direction of LAI and GPP but also the values observed empirically, although least well in the coldest zone (Fig. 5).

Discussion

Given the strong and consistent inverse relationship between leaf life span, nitrogen, and leaf carbon dioxide exchange rates documented interspecifically, including for conifers (14, 15, 20), it is highly likely that evergreen conifers in cold, northern zones of the boreal forest have low photosynthesis and respiration rates compared with either conspecifics or unrelated taxa in warmer, southern habitats. A combination of direct and indirect effects of low temperatures, including low nutrient availability and a short growing season, likely contribute to selection for increasingly conservative foliage traits at the highest latitudes (25, 27). Thus, patterns of NL and %N are consistent with the notion that under cold, nutrient-poor conditions and a short growing season, which limit realized carbon gain, a successful ecological strategy involves trading off potential for high carbon gain by having leaves with low maximum photosynthetic capacity but associated low respiratory costs and having canopies that require only 6–12% replacement each year (14).

Although we present results solely in relation to MAT, it is unclear whether it is growing-season temperatures (or precipitation), average annual temperatures (or precipitation), or relative moisture and nutrient supply that are mechanistically responsible for the low growth and associated long needle life span in high-latitude, cold locations (Supporting Information). Similarly, it is likely that low temperatures and dry conditions both contribute to high carbon allocation below ground in colder forests (25, 26). For the study sites on either continent, MAT was correlated strongly with mean annual precipitation (MAP), as well as with seasonal climate metrics and with the ratio between MAP and potential evapotranspiration. MAT explained as much or more of the variance in needle traits than any other climate metric and was selected statistically in model selection, hence its use herein. Nonetheless, it is likely that the documented patterns of needle trait variation with MAT are a response to seasonal as well as annual aspects of the thermal and moisture regimes along the latitudinal gradients.

Boreal forests play key roles in global carbon cycling, so developing accurate ecosystem understanding is critical for improved carbon cycle modeling. The region we studied here represents 29% of global forests and was found to have the strongest positive carbon feedback to future warming using earth system models (28). Empirical observations have found that needle life span and needle %N (this paper) and the fraction of total stand biomass distributed to foliage (26) all vary with MAT within this biome, but such patterns have not been accounted for by any of the earth system models that were used for estimating the sensitivity of terrestrial carbon balance to warming.

Using CABLE, we showed that implementing one or two of these three model changes (Table 1) as shown in Figs. 3–5 will either minimally alter or worsen the model predictions (compared with FLUXNET observations) compared with simulations without any of those changes as shown in the default output. Only with all three changes implemented in the model (fALLfNLf%N, Table 1) does the performance of CABLE become quite similar to the default CABLE for GPP and improved for LAI (Fig. 5). Therefore, implementing ecological realism into CABLE, and perhaps other global land surface models, may not lead to better agreements with observations or independent estimates of GPP. However, given that the default CABLE model uses unrealistically low and unrealistically geographically stable values for a key leaf parameter (turnover rate), it is possible, perhaps likely, that the default CABLE (and perhaps other models) may agree with empirical observations because other aspects of the model have compensating errors that were not previously identified. Such errors may be invisible in models gradually developed, parameterized, and compared with observations, because unintentional tuning can occur that causes models to match each other and observations under conditions for which they were developed but diverge under novel conditions that additionally challenge the models (29). We observed something similar here using a sensitivity analysis to 3 °C warming.

The simulated sensitivity of annual GPP to a warming of 3 °C is quite different for each of the seven versions (Table 1) of CABLE (Fig. S4), even when their predictions for current conditions are similar. The sensitivity of the simulated annual GPP by the default CABLE to a 3 °C warming is quite small for each of four zones. The sensitivity of the simulated annual GPP to a 3 °C warming is negative if NL is constant but realistic (cALLcNLc%N) or is realistically varied with MAT (cALLfNLc%N). In contrast, sensitivity to warming is positive if variable biomass allocation is used (Fig. S4). However, it has been observed that growth and productivity of boreal trees and forests increase with warming, but more further to the north (30–32), which is only mirrored in CABLE predictions if NL, needle %N, and biomass allocation are all biogeographically varied (fALLfNLf%N).

One take-home message of the CABLE modeling work (Figs. 3–5 and Fig. S4) is that incorporating more realistic parameter settings or functional equations into a model may initially make predictions worse rather than better, but such exploration is still critically needed, because it has the potential to help identify other ways in which a model could be, and should be, improved. In this particular case, modifying CABLE in three ways—with biogeographically varying allocation, NL, and %N—improved predictions in some ways (compared with the original default version of the model), such as LAI and sensitivity of GPP to temperature. This will not necessarily happen however, upon making models more realistic, because the complexity of many land surface models could result in performance declines with improved parameterization. We view such a process—of one step backward and two steps forward—as positive rather than negative, because it provides an opportunity to “get under the hood” of the model and find whatever else may need improving.

In summary, our observations of intraspecific needle trait variation support the idea that some important vegetation attributes likely vary in repeatable patterns along climate gradients, and that identifying such patterns can help to advance our general understanding of biogeographic macroecology. That such patterns can assist the development of enhanced global models is further reason to encourage continued exploration in this domain.

Methods

Needle traits for P. banksiana, P. mariana, P. glauca, and A. balsamea were sampled from individual 2.5- to 5-m-tall trees from 15- to 30-y-old forests, after full leaf expansion and before senescence in the summers of 2005 and 2006. Sampling of canopy leaf traits was made at 127 sites encompassing an area from 43.2° to 67.0°N, and 84.4° to 117.7°W. Dominant upland tree species (including five deciduous angiosperms in addition to the four conifers) were sampled in zones of ∼0.5 ° latitude (∼111 km) delineated before travel. The first suitable site observed with at least three species was sampled and supplemental sites were added as feasible until as many of the available target species as possible were sampled from each zone. All target species present were sampled at each site. At least two individuals were sampled from each site. Sampling occurred within forest stands and >25 m from disturbed areas such as roads. Only healthy/vigorous trees were sampled. Lateral branches with southern exposure from the lower portion of the top 25% of the crown were cut using telescoping clippers. From those branches, prior year (1-y-old) needles were sampled and needle life span was evaluated. Branches with male cones, cone scars (in P. bansksiana), disease, or insect damage were avoided. These species produce needles once per year in spring. Thus, needle life span was calculated by counting all living annual cohorts (from terminal scar to terminal scar) up to and including the youngest cohort with >50% but <90% of needles remaining (18, 33). This was done on the canopy branch or highest accessible branch by hand if annual cohorts extended beyond what was present on an individual branch. Life span of the oldest cohort retaining >50% of the original live foliage was estimated to the nearest 10th of a cohort (see refs. 18 and 33). Because cohorts younger than this one tended to retain nearly all of their foliage, and cohorts older than this tended to retain few, this technique provides a reasonable estimate of both mean cohort and mean needle longevity.

Needles were scanned for leaf area and killed and dried in a 65 °C drying oven within 12 h of their collection. Leaves or needles from individual trees were ground into a composite sample and analyzed for analyzed for %N at the University of California, Davis Stable Isotope Facility (Department of Plant Sciences) using an Elementar Vario EL Cube or Micro Cube elemental analyzer (Elementar Analysensysteme GmbH) interfaced to a PDZ Europa 20-20 isotope ratio mass spectrometer (Sercon Ltd.).

CABLE is the Australian community land surface model (22, 34) and has been implemented into the Australian community earth system model (ACCESS) (34). CABLE simulates the exchange of momentum, heat, water, and CO₂ exchange between land surface and lower atmosphere at hourly time steps, plant growth, litter fall, and biogeochemical cycling of carbon, nitrogen, and phosphorus at daily time steps. As in most global land surface models, 13 plant functional types are used to approximate the variation of vegetated surface on Earth and each plant functional type is defined by the unique set of parameters values that are estimated from calibration or published literature (see ref. 21). CABLE performs well compared with other global land surface models at site level (see ref. 35) and can also reproduce major global biogeochemical pools and fluxes quite well (see ref. 21). The model has recently been used to study the effects of correlations among key leaf traits on the simulated gross primary production (36).

For this study, our goal was to explore how different assumptions about needle life span, needle N concentration, and the fraction of NPP allocation to foliage affect the simulated GPP, NPP, and canopy LAI of evergreen needle leaf forests in North America, Europe, and Asia. For some of the CABLE scenarios we generated temperature-dependent NL (logNL = 0.8246–0.0320*MAT°C) based on regression of NL in relation to MAT that included all available data for all five species (n = 845). In CABLE we also generated foliage N (mg/g) as a function of NL (logN = 0.1356–0.2252*logNL) based on regressing N vs. NL, including all available data for all five species (n = 391). To account for the decline in needle N with canopy depth within the canopy, as observed for evergreen conifer forests (e.g., refs. 37 and 38), we assumed that needle N decreases exponentially with the cumulative canopy LAI (see ref. 39). This adjustment has a similar impact on whole-canopy N as adjusting N by weighting the canopy by proportion in different needle age classes and adjusting for differences in %N with needle age. We drove CABLE by reusing the three-hourly 1° by 1° meteorological forcings for 1990 from the Global Soil Water Project II (40, 41) year after year until the simulated annual GPP and NPP reached their steady-state values. All of the simulated results as reported here are steady-state values. Additionally, to analyze the sensitivities of simulated annual GPP, NPP, and canopy LAI to a 3 °C warming, we recalculated the parameter values (fraction of NPP allocated to needle, needle life span, and %N) that vary with MAT using MAT+3 if applicable for every land point within the domain of this study for each of the seven scenarios, and ran CABLE by adding 3 °C to the surface air temperature across every time step. CABLE was run by reusing the GSWPII meteorological forcing for 1990 until steady-state values of GPP, NPP, and canopy LAI were obtained. The sensitivities of annual GPP, NPP, and canopy LAI to 3 °C warming were then calculated using the results from the simulations here and previously without 3 °C warming.