Abstract
The temperature dependence of global photosynthesis and respiration determine land carbon sink strength. While the land sink currently mitigates ~30% of anthropogenic carbon emissions, it is unclear whether this ecosystem service will persist and, more specifically, what hard temperature limits, if any, regulate carbon uptake. Here, we use the largest continuous carbon flux monitoring network to construct the first observationally derived temperature response curves for global land carbon uptake. We show that the mean temperature of the warmest quarter (3-month period) passed the thermal maximum for photosynthesis during the past decade. At higher temperatures, respiration rates continue to rise in contrast to sharply declining rates of photosynthesis. Under business-as-usual emissions, this divergence elicits a near halving of the land sink strength by as early as 2040.
INTRODUCTION
The difference between gross primary productivity, carbon uptake by vegetation, and total ecosystem respiration, carbon loss to the atmosphere, comprises the metabolic component of the land carbon sink [net ecosystem productivity (NEP)]. To date, land ecosystems provide a climate regulation service by absorbing ~30% of anthropogenic emissions annually [mean ± 1 SD: 2.6 petagrams carbon (PgC) ± 0.8 year−1] (1). While temperature functions as a key driver of year-to-year changes in the land carbon sink (2), its temperature response is still poorly constrained at biome to global scales (3, 4), making the carbon consequences of anticipated warming uncertain.
Like all biological processes, metabolic rates for photosynthesis and respiration are temperature dependent; they accelerate with increasing temperature, reach a maximum rate, and decline thereafter. Yet, these carbon fluxes do not necessarily have the same temperature response, potentially resulting in sharp divergences in ecosystem carbon balance. For example, increasing respiration rates without corresponding increases in photosynthesis rates would decrease the efficacy of the terrestrial carbon sink. An observational constraint on the net difference in metabolic response across both gross fluxes is thus urgently needed to constrain projections of the future land carbon sink and, more specifically, isolate points of nonlinear and perhaps nonreversible change—tipping points (5). This is especially relevant given the highly divergent land carbon sink trajectories from Earth system models (4) that, nevertheless, agree on continued future increases in sink strength due to the CO2 fertilization effect (3).
Given in situ evidence that regions of the terrestrial biosphere are experiencing temperature thresholds at which they switch from a carbon sink to source (6–8), we asked the following questions: (i) What are the thermal maxima of photosynthesis
To address these questions, we used measurements from the largest continuous carbon monitoring network, FLUXNET (9), as an observational constraint to determine the temperature dependence of global rates of photosynthesis and respiration. Across ~1500 site years of daily data from all major biomes and plant functional types, we applied a 30-day moving window partial correlation analysis at each flux tower site to extract the temperature signal (a change in photosynthesis or respiration solely attributable to changes in temperature, i.e., the signal excludes other climatic effects such as water availability and sunlight) from daytime partitioned gross primary productivity [photosynthesis (P)] and total ecosystem respiration (R). We then normalized each site-level temperature dependence curve and applied macromolecular rate theory (MMRT) (10) in conjunction with Monte Carlo resampling to avoid length-of-record bias. The curves were subsequently aggregated to the biome level and then area-weighted to arrive at a global constraint of temperature dependence (see Materials and Methods). MMRT is a framework rooted in the principles of thermodynamics, which provides a mechanistic basis to extract the temperature dependence of rates across scales from individual enzyme kinetics to organismal and ecosystem metabolism (see Materials and Methods) (11). This framework is based on classical transition state theory from physical chemistry (12) and describes temperature rate dependence using three parameters, with emphasis on a maximum or optimal temperature value, Tmax, above which rates decline exponentially. The Arrhenius function is a special case of MMRT where the heat capacity term is zero and the temperature-rate relationship is exponential without a maximum (see Materials and Methods) (10). MMRT is applicable across a range of processes and levels of biological organization and has been successfully used to model the temperature dependence of enzyme kinetics (13), microbial growth (14), soil respiration (15), and leaf respiration (16). Here, we extend this analysis to include global land photosynthesis and net carbon fluxes, producing the first observationally derived curves for the temperature dependence of global carbon metabolism, using a single function grounded in thermodynamics.
RESULTS
In FLUXNET, the temperature response of global photosynthesis shows distinct maxima (
The normalized global temperature response of C3 photosynthesis (green), which exhibits
In contrast to photosynthesis, respiration rates increase across the range of ambient temperatures observed by FLUXNET (up to 38°C) (9), with no evidence of
With temperature dependencies of both gross fluxes, we use mass balance to derive the optimal temperature of net land carbon uptake
Integrated global temperature response curves for normalized photosynthesis (green dashed line), respiration (red dashed dotted line), and a mass balance estimate of the land sink (blue solid line) in relation to current climate (gray bar), where the mean across each curve sums to zero. Photosynthesis represents the integration of C3 and C4 curves (Fig. 1) weighted by global fraction of C3/C4 photosynthesis (37). The gray shaded bar represents observed mean annual temperature range from 1991 to 2015 (9, 22), and vertical dashed line indicates current annual mean temperature at FLUXNET tower sites.
This link between anticipated warming and declines in land carbon uptake is a function of differential responses of gross fluxes to temperature. At temperatures up to the inflection point for photosynthesis, respiration and photosynthesis are effectively “coupled”—both processes increase with increasing temperature, albeit at different rates. At temperatures above the inflection point for photosynthesis, however, these processes become increasingly “decoupled.” That is, the rate of increase slows for photosynthesis and, past
Currently, less than 10% of the terrestrial biosphere experiences temperatures past
(A) Current cumulative monthly dose of temperature above
DISCUSSION
Our findings demonstrate temperature limits for global photosynthesis rates and the terrestrial land sink as a whole. Despite two decades of FLUXNET observations and the warmest decade on record, we observed no evidence of acclimation in photosynthesis (see Materials and Methods and fig. S2). While it is possible that temperature adaptation could mitigate the size of this impact, given high daily, seasonal, and interannual variation in temperature, as opposed to uniform warming from experimental data, the likelihood of detecting acclimation is low. Furthermore, two decades is likely too short a period to see selection for genotypes with higher temperature tolerance, particularly in systems dominated by perennial plants (16–28). Given current proximity to
Beyond acclimation, and despite an increase of ~40-ppmv (parts per million by volume) CO2 over the 1991–2015 FLUXNET record, we also observed no notable alteration in the magnitude of photosynthesis across the data record (fig. S3). We note that, on the basis of the solubility of CO2 as a function of temperature and pressure, leaf water affinity for CO2 is nearly unchanged across the data record (30). We therefore contend that, in contrast to any CO2 fertilization effect (3), anticipated higher temperatures associated with elevated CO2 could degrade land carbon uptake and that failure to account for this results in a gross overestimation of climate change mitigation provided by terrestrial vegetation. We note that future work accounting for the timing of photosynthetic activity (31), CO2 concentrations, and the solubility of CO2 as a function of temperature (30) will be essential to accurately predict the role of CO2 fertilization in the land sink of carbon (32).
The temperature tipping point of the terrestrial biosphere lies not at the end of the century or beyond, but within the next 20 to 30 years (Figs. 2 and 3, A to D). Given the temperature limits of land carbon uptake presented here, without mitigating warming, we will cross the temperature threshold of the most productive biomes by midcentury, after which the land sink will degrade to only ~50% of current capacity if adaptation does not occur. While biomes will eventually shift spatially in response to warming, this process is unlikely to be a smooth migration, but rather a rapid disturbance-driven loss of present biomes (with additional emissions of carbon to the atmosphere), followed by a slower establishment of biomes more suited to the emerging climate. Furthermore, the establishment of new biomes is unlikely to be complete without human intervention and will be limited by edaphic factors, especially nutrient availability. This further suggests that we are rapidly entering temperature regimes where biosphere productivity will precipitously decline and calls into question the future viability of the land sink, along with Intended Nationally Determined Contributions (INDCs) within the Paris Climate Accord, as these rely heavily on land uptake of carbon to meet pledges (33). In contrast to Representative Concentration Pathway 8.5 (RCP8.5), warming associated with scenario RCP2.6 could allow for near-current levels of biosphere productivity, preserving the majority land carbon uptake (~10 to 30% loss). Failure to implement agreements that meet or exceed limits in the Paris Accord could quantitatively alter the large and persistent terrestrial carbon sink, on which we currently depend to mitigate anthropogenic emissions of CO2 and therefore global environmental change.
MATERIALS AND METHODS
Macromolecular rate theory
MMRT is based on classic transition state theory that describes the temperature dependence of chemical reaction kinetics using statistical thermodynamics (6). A central tenant of MMRT is the explicit recognition of the change in heat capacity for enzyme-catalyzed rates,
For MMRT, the heat capacity term is incorporated into the Eyring equation to give Eq. 1 below. The resultant signature of MMRT is a curved plot of rate versus temperature in the absence of denaturation.
Equation 1 (based on transition state theory) lies at the heart of MMRT. It describes the rate (k) in terms of the Boltzmann, Planck, and universal gas constants (kB, h, and R, respectively), absolute temperature (T), the transmission coefficient (κ), the change in enthalpy
We have previously demonstrated that MMRT accurately describes the temperature dependence of biological rates at increasing levels of complexity [microbial growth rates (14), soil respiration (15), and plant respiration (16)]. This provides justification for using MMRT in for analysis of the temperature dependence of ecosystem fluxes in the FLUXNET dataset.
FLUXNET data and processing
As an observational dataset for the temperature dependence of land carbon fluxes, mean daily estimates of carbon fluxes and micrometeorological variables were retrieved from the FLUXNET 2015 synthesis dataset for all tier 1 and tier 2 sites, along with uncertainty estimates based on gap filling and SD of fluxes (35). The daytime partitioning algorithm was selected for estimates of gross primary productivity (P) and total ecosystem respiration (R) to minimize bias associated with constraining the temperature response of partitioned fluxes through nighttime temperatures (36). Concurrent estimates of air temperature, latent and sensible heat, and downwelling short-wave radiation flux were also extracted, along with metrics for gapfilling of flux and meteorological data.
Combined, latent and sensible heat were used to calculate evaporative fraction (EF), the inverse of the Bowen ratio, and a robust index of relative water availability to the biosphere (Eq. 3) (37)
EF is an effective metric to assess water availability as it captures the signal from a multitude of potential water pools (e.g., soil moisture and precipitation) through evapotranspirative fluxes from the biosphere and scales well globally. We also evaluated the signal of vapor pressure deficit (VPD) as an alternate metric for water stress but found no statistically significant alteration of response (fig. S4). Given large variation in productivity and climate across biomes, we normalized all carbon fluxes along with micrometeorological variables within site to avoid signals based on biogeography. To maintain in situ derivation of temperature dependence, carbon flux and temperature data that were fully gap-filled were excluded from the analysis.
Temperature signal
Both photosynthesis and respiration are known to be controlled by a number of enviroclimatic variables, namely, sunlight, water, and temperature. To isolate the temperature signal, we used a 30-day moving-window partial correlation analysis on daily estimates of daytime partitioned gross primary productivity and total ecosystem respiration, with EF, downwelling short-wave radiation, and air temperature from 0° to 38°C (from biologically relevant temperatures for metabolic activity to the upper limit of the FLUXNET record) as explanatory variables at the individual site level, and filtered for significant relationships at P < 0.1. The result was the proportion of variation in gross fluxes that were solely attributable to each enviroclimatic variable. We then normalized and fit the temperature response of both fluxes by site to the first derivative of MMRT (11) to investigate changes in metabolism as a function of temperature (Eqs. 1 and 2). As MMRT was fitted at the site level and then bootstrapped, we filtered the FLUXNET synthesis dataset for towers that had a > 10 statistically attributed data points to ensure that data were sufficient to constrain temperature response curvature. Temperature projections from MMRT were limited to the ambient temperature window of observations from the FLUXNET record (0° to 38°C, the upper limit of FLUXNET observations).
Tmax or tipping point determination
Two important points exist within temperature dependence curves for biosphere metabolism: the inflection point (Tinf) and the thermal maximum (Tmax). The inflection point of temperature-dependent rates represents temperatures where an increase in rate (k) is maximal relative to temperature (T) and denotes where rates change from convex to concave.
Representativeness and uncertainty of FLUXNET data
Two major challenges exist for global constraint of biosphere metabolism using the FLUXNET dataset: (i) Observations are unequally distributed across the vegetated surface, with differing lengths of record, and (ii) because of the ambient and mixed-species nature of tower observations, the data are inherently noisy. The FLUXNET dataset, however, has been shown to be statistically representative of plant functional types and Koeppen-Geiger biomes globally, suggesting that statistical analyses leveraging FLUXNET observations are robust (38).
To address challenges (i) and (ii), we designed a bootstrap method that sampled entire tower records rather than observations (to avoid length of record bias) to fit global curves for the temperature dependence of land carbon fluxes. We then calculated the bootstrap-to-bootstrap variation in Tmax within Koeppen-Geiger biomes and across latitudes (fig. S2) to capture uncertainty stemming from sparse data sampling and heterogeneity of ecosystems sampled.
Bootstrapping and C3/C4 dichotomy
Early investigations into the temperature response of photosynthesis at global scales demonstrated a clear bimodal distribution that was well explained by C3 and C4 heuristics (39). All FLUXNET sites were therefore dichotomized into these two groups on the basis of climate criteria (40). The temperature response by group was then bootstrapped across the FLUXNET synthesis dataset 10,000 times, such that an entire site’s temperature curve was sampled rather than specific observations, thereby decreasing the length-of-record bias and unequal distribution across bioclimatic space of some long-lived sites.
Respiration temperature signal
The exponential Arrhenius-like response of total ecosystem respiration is largely a function of ambient temperature observations, where FLUXNET effectively samples temperatures far below
Evaluation of CO2 fertilization and VPD
To evaluate the effect of CO2 on the temperature dependence of gross primary production and the ability of CO2 fertilization to counter temperature-induced declines in land carbon uptake, we conducted a second moving-window partial correlation analysis that included ambient CO2 from FLUXNET towers with all other components remaining identical. While a ~1% increase in NEP per year has been correlated with elevated CO2 elsewhere in the literature (32), 90% confidence intervals of the temperature dependence signal with and without CO2 included showed no statistically significant difference (fig. S3). An identical model run was completed, replacing VPD with EF to evaluate potential changes in temperature response based on atmospheric demand rather than water flux, again demonstrating no significant differences (fig. S3).
Spatial gridding
To generate spatial grids of temperature response, FLUXNET sites were aggregated on the basis of Koeppen-Geiger climate classification regions where the FLUXNET synthesis dataset retained a coverage of >5 sites. We collapsed classes lacking sufficient replication to the next level of climatological organization.
Cumulative dose of temperature and fraction of vegetated surface in decline
Of interest for our work was the cumulative amount of mean monthly temperatures beyond
Decrease in biosphere productivity RCP8.5
As biosphere productivity varies spatially, we incorporated upscaled FLUXNET data to evaluate the impacts of exceedance of Tmax on total terrestrial biosphere productivity (38). To avoid biases stemming from interannual variability, we calculated mean biosphere productivity between 2003 and 2013 and evaluated temperature exceedance on the basis of our
Acclimation
To search for evidence of acclimation at the ecosystem-to-global scale, we first isolated FLUXNET tower sites that spanned both the first and second decades of the 2015 synthesis dataset. We then evaluated temperature dependence curves across both decades to look for upward shifts in
Data availability
FLUXNET data access depends on the tier of data used. Tier 1 data are open and free for scientific and educational purposes, and their use follows the fair use policy accessed at https://fluxnet.fluxdata.org/data/data-policy/. Tier 2 data are from producers who are currently unable to share their data in an open manner and require an approved proposal for data access. Data access proposal information can be found at https://fluxnet.fluxdata.org/. A list of FLUXNET sites from tiers 1 and 2 used in this analysis can be found in table S3. Downscaled WorldClim CMIP5 climate data used to evaluate future climate and therefore climate space beyond
SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/7/3/eaay1052/DC1
This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.
REFERENCES AND NOTES
- Copyright © 2021 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC).