Land carbon-concentration and carbon-climate feedbacks are significantly reduced by nitrogen and phosphorus limitation

Our future climate will depend on future emissions of greenhouse gases such as CO₂. However, the physical climate also depends on the global carbon cycle and how they interact with each other. Changes in the physical climate affect the uptake of CO₂ by the land and the ocean leading to changes in atmospheric CO₂ concentrations which in turn affect the physical climate. For example, an increase in temperature leads to an increase in soil heterotrophic respiration, reducing the land carbon uptake. Therefore, more CO₂ remains in the atmosphere which increases temperature even further. The sensitivity of the land carbon cycle to changes in atmospheric CO₂ concentrations and the physical climate can be characterised in terms of the carbon-concentration and carbon-climate feedback (Friedlingstein et al 2006, Boer and Arora 2009, Arora et al 2013).

Feedback analysis with earth system models (ESMs) has shown that the carbon-concentration feedback (β_L ) is positive from the land perspective at the global scale, meaning that an increase in CO₂ leads to an increase in carbon uptake by the land. On the other hand, the carbon-climate feedback (γ_L ) is thought to be negative at the global scale, because an increase in temperature generally decreases the capacity of the land to take up carbon. The quantification of the carbon-climate feedback assumes that changes in the physical climate can be expressed in terms of changes in the near-surface air temperature (Arora et al 2013).

Land carbon uptake and storage is not only dependent on environmental factors but is also controlled by the availability of nutrients such as nitrogen (N) and phosphorus (P), because they are able to constrain plant growth and photosynthesis (Vitousek and Howarth 1991, Vitousek et al 2010, Fleischer et al 2019). For example, a nutrient limitation can be diagnosed when additional N and/or P leads to an increase in plant growth. Due to the importance of nutrients, ESMs have begun to implement N limitation in land carbon cycle processes (Sokolov et al 2008, Thornton et al 2009). Nonetheless, only two ESMs used in the fifth Coupled Model Intercomparison Project (CMIP5) include N limitation and none include P limitation. Arora et al (2013) showed that the two N limited models in CMIP5 generally have a weaker carbon-concentration and carbon-climate feedback as compared to the models without N limitation.

In the most recent Coupled Model Intercomparison Project, CMIP6, the number of ESMs with N limitation has increased (i.e. six ESMs with N limitation as analysed in Arora et al 2020) and results demonstrate that as with CMIP5 the models that include a representation of the N cycle show a smaller feedback strength. However, the range of the feedbacks for both N limited and not N limited models in CMIP6 has also increased in comparison to CMIP5, although the range for the N limited models is generally smaller than the range for models that are not N limited. This demonstrates that large uncertainties remain in the quantification of the feedback strength and the impact of nutrients on land carbon uptake (Zhang et al 2014, Wieder et al 2015, Davies-Barnard et al 2020). Reducing these uncertainties is critically important for providing better estimates of future land carbon uptake and for developing suitable climate mitigation policies.

Although N limitation is now implemented in a number of ESMs, P limitation remains uncommon. In fact, there is only one ESM in CMIP6 that uses global N and P limitation (Wang et al 2010) in its standard configuration, which is the ESM version of the Australian Community Climate and Earth System Simulator (ACCESS), ACCESS-ESM1.5 (Ziehn et al 2020). ACCESS-ESM1.5 is a fully coupled ESM with components for atmosphere, land, ocean and sea-ice, including biogeochemical (i.e. carbon and nutrient cycle) modules for land and ocean. Here, we use ACCESS-ESM1.5 to analyse the impact of both N and P limitation on the land carbon uptake and consequently on the land carbon-concentration and land carbon-climate feedbacks. We aim to answer the following key questions:

• (a)  
  By how much does nutrient limitation reduce the land carbon uptake, and the carbon-concentration and carbon-climate feedback strengths?
• (b)  
  What are the relative roles of nitrogen limitation and phosphorus limitation to these reductions?
• (c)  
  Do the effects of nitrogen and phosphorus limitations on carbon-concentration and carbon-climate feedbacks vary latitudinally?
• (d)  
  How does nutrient limitation affect vegetation productivity and soil heterotrophic respiration and how does this impact both feedbacks?

2.1. Nutrient limitation in ACCESS-ESM1.5

2.2. Feedback analysis

The feedbacks are derived from standardised '1pctCO2' experiments as part of the CMIP6 DECK (Diagnostic, Evaluation and Characterization of Klima) (Eyring et al 2016) and the Coupled Climate-Carbon Cycle Model Intercomparison Project (C4MIP) (Jones et al 2016), where prescribed atmospheric CO₂ concentrations are increased by 1% per year from pre-industrial levels (corresponding to year 1850) over a 140 year period (figure 1(a)). Simulations are performed with a static land cover map representing 1850 conditions and therefore do not consider any land use and land cover change.

Zoom In Zoom Out Reset image size

Our feedback analysis is largely based on the framework used in Friedlingstein et al (2006) and Arora et al (2020) with the land carbon uptake defined as:

$$\begin{equation} \Delta C_L = \beta_L \Delta c + \gamma_L \Delta T \end{equation} \tag{ 1 }$$

where Δc is the change in atmospheric CO₂ concentrations, ΔC_L stands for the land carbon uptake (or change in global carbon stocks) and ΔT for the changes in near-surface temperature. The feedback parameters, β_L and γ_L , can be calculated using the results from any two combinations of fully coupled (COU), biogeochemically coupled (BGC) and radiatively coupled (RAD) simulations. In the BGC simulation only the biogeochemical processes for land and ocean respond to the increase in CO₂, whereas in the RAD simulation only the radiative transfer calculations are impacted by the CO₂ increase. The COU simulation combines both of these effects.

Under the assumption of a linear system all combinations would results in the same values for β_L and γ_L . However, in practice this is not the case due to non-linearities in the models. On the land, this non-linearity can be associated with the different response of vegetation productivity and soil heterotrophic respiration to an increase in temperature in the presence or absence of increasing CO₂ (Zickfeld et al 2011). The non-linear interactions in ACCESS and other CMIP6 models will be further investigated in a separate study.

Here we use the COU and BGC simulations to calculate the feedback parameters by successively solving equation (1) for β_L and γ_L :

$$\begin{equation} \beta_L = \frac{1}{\Delta c} \left(\frac{\Delta C_L^{\,BGC} \Delta T^{\,COU} - \Delta C_L^{\,COU} \Delta T^{\,BGC}}{\Delta T^{\,COU}- \Delta T^{\,BGC} }\right) \end{equation} \tag{ 2 }$$

$$\begin{equation} \gamma_L = \frac{\Delta C_L^{\,COU} - \Delta C_L^{\,BGC}}{\Delta T^{\,COU} - \Delta T^{\,BGC}}. \end{equation} \tag{ 3 }$$

We consider three different configurations of ACCESS to quantify the impact of nutrient limitation on the feedbacks: (1) no nutrient limitation referred to as C, (2) nitrogen limitation referred to as CN and (3) nitrogen and phosphorus limitation referred to as CNP. Therefore, our analysis is based on three sets of simulations (COU-C and BGC-C, COU-CN and BGC-CN, COU-CNP and BGC-CNP). We also compare our results for land carbon uptake and the feedbacks for the C and CN configuration against the CMIP6 model mean as presented in Arora et al (2020). The feedback analysis in Arora et al (2020) is based on results from five ESM's using a C configuration and five ESM's using a CN configuration. ACCESS results are also included in this analysis, however only in the CNP configuration, which is our standard configuration.

2.3. Surrogate model

In order to provide a more detailed analysis of the modelled response of β_L and γ_L for the three different ACCESS configurations we propose a simple model surrogate driven by net primary production (NPP) from the ACCESS simulations. The land carbon uptake (ΔC_L ) can be expressed as the sum of changes in carbon stocks over a specified period of time or as the integrated net ecosystem productivity (NEP), which is also defined as the difference of integrated NPP and integrated soil heterotrophic respiration (R_h ):

$$\begin{equation} \Delta C_L = \int NEP = \int NPP - \int R_h. \end{equation} \tag{ 4 }$$

To derive the model surrogate we rewrite the total land carbon uptake ΔC_L in equation (4) as the sum of the changes in three carbon pools (a vegetation biomass pool C_v , a fast soil C_fs and a slow soil pool C_ss ):

$$\begin{equation} \Delta C_L = \Delta C_v + \Delta C_{fs} + \Delta C_{ss}. \end{equation} \tag{ 5 }$$

Formulating this as a balance equation for each carbon pool leads to a total of eight parameters that can be optimised by minimising the mismatch between the surrogate carbon pools and the simulated ACCESS carbon pools. The eight parameters represent fractions of carbon allocated to the pools, rate constants and temperature coefficients and from these R_h can be derived for all three configurations.

A more detailed description including all surrogate model equations can be found in the supplementary material (available online at stacks.iop.org/ERL/16/074043/mmedia). Assuming a reasonable fit of the model surrogate, the calculation of β_L and γ_L using the carbon pool changes based on the model surrogate should be close to those calculated from ACCESS.

The model surrogate then allows us to quantify how nutrient limitation affects vegetation productivity and carbon loss from soil heterotrophic respiration (as a function of the eight surrogate parameters) and how this relationship impacts the feedback parameters β_L and γ_L . This type of analysis is not possible based on the ACCESS results alone, because in ACCESS we cannot easily apply nutrient limitation to only one specific process. With the surrogate we can, for example, separate NPP and R_h and analyse possible combinations of NPP (in C, CN or CNP configuration) with R_h (in C, CN or CNP configuration).

3.1. Land carbon uptake and feedbacks

3.1.1. Global scale

The change in prescribed atmospheric CO₂ concentrations and the corresponding change in global mean surface temperature for ACCESS are shown in figures 1(a) and (b). The temperature change in the COU simulation over the 140 years is about 4.5^∘C for all configurations. For the BGC simulation the temperature change varies between 0.15 ^∘C and 0.39 ^∘C. The small increase in temperature for the BGC simulations can be attributed to changes in latent and sensible heat fluxes at the land surface due to changes in vegetation structure, coverage and composition. ACCESS does not have a dynamic vegetation model, but uses a prognostic leaf area index and climatic changes are therefore mainly due to increased radiation and reduced stomatal conductance in response to increasing atmospheric CO₂ (Law et al 2017, Ziehn et al 2017).

Annual NEP and land carbon uptake are presented in figures 1(c) and (d). NEP is a relatively small flux, representing the difference of two large fluxes NPP and R_h . Depending on which of the larger fluxes dominates, NEP is either positive or negative. Variations in the climate lead to a large interannual variability in NEP, whereas the trend in NEP is determined by a change in the climate. For the COU simulations, we see an initial increase in NEP mainly due to an increase in NPP in response to the increase in CO₂ concentrations (also known as CO₂ fertilisation effect). Nutrients limit the increase in NPP in the CN and CNP configuration and we therefore see the largest increase in NEP for the C configuration. The increase in temperature also leads to a relative increase in R_h . However, this effect is delayed due to large carbon pool turnover times. We therefore see a peak in NEP between years 20 and 30 and then a strong decline for all three configurations. The NEP response for the BGC simulations is similar to the COU simulations at the beginning due to the increase in CO₂ and consequently in NPP. In contrast, temperature does not increase significantly in the BGC simulations, which means that there is no temperature related increase in R_h . Therefore, NEP and land carbon uptake are larger for the BGC simulation for all three configurations compared to the COU simulations. In the COU simulation the land becomes a carbon source for all configurations after about 120 years mainly due to the relative increase in R_h . In the CNP configuration the differences in carbon uptake between COU and BGC are significantly smaller than in the C and CN configuration, which suggests that in the CNP configuration the temperature increase has an overall smaller (negative) impact on the land carbon uptake.

Table 1 also lists the global land carbon uptake over the 140 years including a comparison with the CMIP6 mean for models in C and CN configuration. In the COU-C simulation the land carbon uptake in ACCESS is 393 Pg C and 882 Pg C in the BGC-C simulation. This again suggests a large (negative) temperature feedback in ACCESS, which will be discussed later. When adding nitrogen limitation (CN configuration), we observe a reduction in land carbon uptake in ACCESS of about 30% in comparison to the C simulations in both, COU-CN (280 Pg C) and BGC-CN (645 Pg C).

Table 1. Land carbon uptake (Pg C) in COU and BGC for C, CN, CNP configuration. CMIP6 values expressed as mean ± 1 standard deviation are based on five models in C and five models in CN configuration. Numbers in brackets indicate the reduction in land carbon uptake with reference to the C and CN configuration respectively.

	ACCESS-ESM1.5		CMIP6 mean
Configuration	COU	BGC	COU	BGC
C	393	882	754 ± 268	1012 ± 335
CN	280 (−29%)	645 (−27%)	601 ± 103 (−20%)	738 ± 120 (−27%)
CNP	213 (−24%)	306 (−53%)	—	—

In the CNP configuration the carbon uptake in ACCESS is further reduced to 213 Pg C in COU and 306 Pg C in BGC. Whereas in COU the addition of P leads to a smaller but overall similar reduction in carbon uptake as the addition of N (about 25%), in the BGC simulation the reduction in carbon uptake is even more pronounced (about 50%) when adding P limitation. This is because when adding P limitation the carbon uptake is reduced over the whole latitudinal range, whereas when adding N limitation the carbon uptake is only reduced in the extra tropics, most strongly pronounced around 60^∘ N. In the COU simulation reduction in carbon uptake due to N and P limitation occurs in the same latitudinal band, mainly around 60^∘ N. Latitudinal differences will be further analysed in the next section.

The land carbon-concentration (β_L ) and carbon-climate (γ_L ) feedbacks are shown in table 2. In the C configuration we calculate β_L as about 1.1 Pg C ppm⁻¹ and γ_L as about −123 Pg C ^∘C⁻¹. We observe a reduction of about 30% in the feedback strength for β_L in the CN configuration (0.8 Pg C ppm⁻¹) in comparison to the C configuration. For γ_L we also notice a reduction of 30%. When adding P limitation (CNP configuration), both feedbacks in ACCESS are reduced further to about 0.4 Pg C ppm⁻¹ for β_L and −22 Pg C ^∘C⁻¹ for γ_L . This corresponds to a reduction of more than 50% and more than 70% respectively in comparison to the CN configuration.

Table 2. Carbon-concentration (β_L Pg C ppm⁻¹) and carbon-climate feedback (γ_L Pg C ^∘C⁻¹) in COU and BGC for C, CN, CNP configuration. CMIP6 values expressed as mean ± 1 standard deviation are based on five models in C and five models in CN configuration. Numbers in brackets indicate the reduction in the feedbacks with reference to the C and CN configuration respectively.

	ACCESS-ESM1.5		CMIP6
Configuration	βL	γL	βL	γL
C	1.11	−122.77	1.20 ± 0.4	−63.80 ± 63.1
CN	0.77 (−31%)	−85.40 (−30%)	0.86 ± 0.14 (−28%)	−31.14 ± 21.9 (−51%)
CNP	0.37 (−52%)	−22.34 (−74%)	—	—

Overall, nutrient limitation (N and P together) in ACCESS reduces the global land carbon uptake by almost 50% (COU simulation), whereas the CO₂ sensitivity (feedback β_L ) and surface air temperature sensitivity (feedback γ_L ) are reduced by about 65% and 80% respectively.

3.1.2. Latitudinal scale

The latitudinal distribution of β_L is shown in figure 2(a). In the C configuration the carbon-concentration feedback is largest in the mid-northern latitudes (around 50^∘ N), which also corresponds with the region of the largest land carbon uptake in both BGC and COU (figures 2(c) and (d) respectively). In fact, the latitudinal distribution of the land carbon uptake in COU is strongly correlated with β_L for all three configuration. This is because the calculation of β_L (equation (2)) mainly depends on the change in the land carbon pools for BGC if the temperature change in BGC is small (which it is in all three cases), scaled by the change in atmospheric CO₂ concentrations, which is the same in all three configurations. For the CN configuration β_L and $\Delta C_L^{\,BGC}$ show only small changes in the tropics in comparison to the C configuration, which indicates that tropical regions are not N limited in our model. In contrast, the difference between the C and CN configuration in the mid to high northern latitudes (30–70^∘ N) for β_L and $\Delta C_L^{\,BGC}$ is significant and N limitation obviously has a large impact here. The addition of P (CNP configuration) on the other hand seems to have a noticeable impact across all latitudes in ACCESS, however the reduction in β_L and $\Delta C_L^{\,BGC}$ in comparison to the CN configuration is largest in the tropics and the mid-northern latitudes.

Zoom In Zoom Out Reset image size

The carbon-climate feedback is mainly determined by the difference between $\Delta C_L^{\,COU}$ and $\Delta C_L^{\,BGC}$ (see equation(3)) and scaled by the change in temperature in the COU simulation (again, assuming the temperature change in the BGC simulation is small). Therefore, if the carbon uptake in the COU simulation is smaller than in the BGC simulation γ_L is negative, which is the case for all configurations for ACCESS at the global scale (see table 2). However, at the latitudinal scale (figure 2(b)) we can identify latitudinal bands with a positive γ_L . For the C configuration these are the latitudes above 60^∘ N and for the CN configurations the latitudes around 60^∘ N. As with β_L the differences between the CN and C configurations are small for the tropics, but larger for the extra-tropics. The negative feedback is stronger for C in comparison to CN for latitudes between 30 and 60^∘ N. This might be due to the fact that in the CN configuration more N becomes available for plants when temperature increases which in turn decreases the difference between $\Delta C_L^{\,COU}$ and $\Delta C_L^{\,BGC}$. In the latitudes above 60^∘ N this even leads to a positive γ_L with the carbon uptake in COU larger than in BGC due to the positive effect of temperature on productivity and nitrogen availability. In the CNP configuration γ_L is smaller almost everywhere when compared with γ_L from the CN configuration. The largest reduction in γ_L can be observed in the tropics due to the large reduction in carbon uptake in BGC. Further north (above 20^∘ N) γ_L becomes slightly positive in the CNP configuration with larger positive values above 45^∘ N, indicating that significantly more P becomes available to the plants in our simulations in those latitudinal bands when temperature increases. The carbon uptake in COU is similar in all three configurations, apart from latitudes above 45^∘ N, which might suggest that overall the climate (temperature) is the limiting factor in those regions.

3.2. Contribution analysis

To better understand the change in the modelled response of β_L and γ_L for the different ACCESS configurations (table 2), we use the three-pool model surrogate driven by NPP from the ACCESS simulations. The optimal parameter values and surrogate performance for all three configurations are presented in the supplementary material. The performance of the model surrogate is adequate ($R^2~\gt 80\%$ for all three configurations) when compared against simulated pool sizes with ACCESS. This is confirmed by comparing β_L and γ_L for all configurations from the surrogate model and from the ACCESS model (figure 3, cross-hatched bar compared with bar of matching colour). The feedbacks obtained from the surrogate are generally close to the those obtained from ACCESS as we would expect (errors $\leqslant 5\%$). Only γ_L for the CNP configuration shows a larger discrepancy between ACCESS and the surrogate (around 30%), due to a larger mismatch in the soil carbon pool sizes for the BGC case (see supplementary material). However, this difference is still small compared to differences in γ_L for the three different configurations. Thus, in order to investigate the response mechanisms between the various configurations this mismatch is not of great concern.

Zoom In Zoom Out Reset image size

With the model surrogate we assess the various combinations of NPP (C, CN, CNP configuration from ACCESS) and R_h (C, CN, CNP parameterization in the surrogate) and the impact this has on the feedbacks (figure 3, crossed hatched bar compared against other colours). Our results show that a change in the R_h parameterization does not significantly impact the feedbacks. Overall this means that the feedbacks are independent of the underlying nutrient configuration of R_h with the feedback strength in our model entirely determined by the NPP configuration.

The land carbon uptake in ACCESS without nutrient limitation (C configuration) is only about half of the corresponding CMIP6 mean uptake (see table 1). However, the range of land carbon uptake is large in the CMIP6 models due to differences in process representation in the land components of those models. These differences are caused by the type of processes included in the land component such as dynamic vegetation or fire, but also in the way processes are parameterised. Results are closer together in the BGC-C simulation (882 Pg C in ACCESS vs. 1012 Pg C for CMIP6).

For β_L in the C configuration there is good agreement between the ACCESS results and the CMIP6 model mean (β_L of about 1.1 and 1.2 Pg C ppm⁻¹ respectively), although the range of the five CMIP6 models is large (0.6–1.8 Pg C ppm⁻¹). On the other hand, γ_L in ACCESS is almost twice as large as the CMIP6 model mean (−124 vs. −64 Pg C ^∘C⁻¹). However, the range of γ_L for the models in the C configuration in CMIP6 is again very large with the two extremes at −163 and +16 Pg C ^∘C⁻¹ and therefore the ACCESS results still fall within this range.

Instead of only comparing the absolute values for land carbon uptake and feedbacks between ACCESS and the CMIP6 models in the C and CN configuration we can also compare the relative change (reduction). These estimates indicate a universal reduction in land carbon uptake (COU and BGC) and feedback strength (β_L and γ_L ) due to N limitation of about 30% for ACCESS. Results based on the CMIP6 model mean (five models in C configuration and five models in CN configuration, tables 1 and 2) are more diverse. The reduction in land carbon uptake in COU is about 20%, whereas in BGC it is closer to 30%. Consequently, the reduction in the feedbacks for the CMIP6 models is about 30% for β_L and 50% for γ_L . However, one of the CMIP6 models in the C configuration represents a strong outlier and removing this model from the calculation of the CMIP6 model mean leads to a more universal reduction of about 20% for both land carbon uptake and feedbacks.

Overall, the reduction in carbon uptake and feedback strength due to N limitation between ACCESS and the CMIP6 model mean shows good agreement. This is remarkable given the large differences in the representation of the N cycle and its impact on vegetation amongst the CMIP6 models, which is demonstrated by the large range of absolute land carbon uptake by those models (table 1).

The latitudinal analysis of the land carbon uptake and feedbacks has revealed that under climate change (COU simulations) mainly the extra-tropical northern latitudes become nutrient limited in ACCESS. For N, this is generally in agreement with meta-data analysis and observations (Vitousek and Howarth 1991, Terrer et al 2020). Phosphorus is assumed to be the key limiting nutrient in the tropics (Vitousek 1984) and the fact that the extra-tropics (mainly centred around 50^∘ N, see figure 3(d) are also P limited in our simulations is somewhat surprising. For example, a simulation of the free-air CO₂ enrichment experiment AmazonFACE with an ensemble of 14 land surface models showed that P availability reduced the biomass carbon growth in the Amazon region by about 50% compared to estimates from models in a C and CN configuration. However, the CO₂ enrichment experiments were conducted over only 15 years and therefore do not take a change in temperature into account. In general, results from CO₂ enrichment experiments are more comparable to our BGC results, where ACCESS in the CNP configuration does indeed show a strong reduction in carbon uptake and feedback strength in the tropics compared to the C and CN configuration.

The comparison of COU and BGC results demonstrates that in our model the climate (i.e. temperature) might be limiting the carbon uptake in the tropics. In the Amazon region this is mainly because temperature in the COU simulation exceeds the optimal temperature for photosynthesis and productivity therefore declines. A similar behaviour for this region was found in Ziehn et al (2020). A more detailed analysis of geographical differences (including the Amazon and other tropical regions) in land carbon uptake and feedbacks for the three different configurations in ACCESS will be provided in separate study.

Our result based on the model surrogate indicate that at the global scale, soil carbon is generally driven by vegetation productivity. This is in agreement with Todd-Brown et al (2013), who demonstrated that differences in simulated soil carbon across ESMs in CMIP5 were caused by NPP. Current ESMs do not explicitly simulate microbial dynamics due to its high computational requirements and lack of nonlinear response to carbon input. It is an arguable question whether the relationship of soil carbon and NPP is approximately linear over large spatio-temporal scales. At the global scale, as with other ESMs, soil carbon in ACCESS is largely driven by NPP, partly due to the smoothing effect over large regions.

Our analysis is based on simulations with a fully coupled ESM using a prescribed 1% yr⁻¹ CO₂ increase over 140 years for three different configurations: C (no nutrients), CN (nitrogen limited) and CNP (nitrogen and phosphorus limited). We find that nitrogen limitation reduces the land carbon uptake and the feedback strength in ACCESS for both the carbon-climate and carbon-concentration feedback by about 30%. Overall, this agrees well with the multi-model mean of CMIP6 models in the C and CN configuration, which shows a similar reduction. The addition of phosphorus limitation in ACCESS reduces the carbon-climate and carbon-concentration feedback further by about 50% and 75%, respectively. The total combined effect of nitrogen and phosphorus limitation in ACCESS leads to a reduction in the carbon-climate and carbon-concentration feedback by about 65% and 85%, respectively. Through the use of a surrogate model, we demonstrated that this reduction can be attributed mainly to a reduction in NPP. As a consequence, future land carbon uptake may not increase as much as projected by models without accounting for nutrient limitation, particularly in the temperate regions due to nitrogen limitation and across all latitudes due to phosphorous limitation. On the other hand, nutrient limitation also reduces the carbon-climate feedback via a reduction in carbon input to soil which in turn reduces soil heterotrophic respiration and additional surface warming. However, the reduction in surface warming and consequently the reduction in carbon emissions (due to the reduced carbon-climate feedback) are much smaller than the reduction of land carbon uptake due to nutrient limitation. Therefore, land-based carbon mitigation may not be as effective as projected by models without accounting for nutrient limitation. Ultimately this means, that a greater reduction in CO₂ emission is required in order to limit global warming to below 2 ^∘C.

Any data that support the findings of this study are included within the article.