Correctly representing the balance between plant photosynthesis, growth and autotrophic respiration in the land surface model (LSM) component of Earth System Models (ESMs) is crucial to making accurate projections of global climate in the future. Forests cover nearly 4000 Mha of the worlds land surface and represent a significant sink of carbon from the atmosphere, sequestering 2.4±0.4 PgCyr-1, roughly 25 % of total annual anthropogenic carbon emissions . Most LSMs simulate growth and respiration as equal to instantaneous photosynthesis . Consequently, at any given time, the total rate of carbon utilisation by respiration and growth, referred to as plant carbon expenditure (PCE), is equal to the rate of carbon accumulation by photosynthesis, commonly referred to as gross primary productivity (GPP). However, in reality growth and respiration are not so strictly coupled to photosynthesis and plants regularly experience periods when the supply of carbon from photosynthesis does not equal the demands of growth and respiration . This asynchrony between supply and demand is facilitated by reserve pools of labile carbon known collectively as non-structural carbohydrates (NSCs). The NSC pool within a plant accumulates when photosynthesis exceeds carbon demand and is drawn upon to sustain growth and respiration when they are not supported by instantaneous photosynthetic assimilation . NSCs therefore act as a buffer, allowing key functional processes to be maintained, even when photosynthetic accumulation is low. This buffering is particularly important during periods of environmental stress, which can lead to reduced productivity over seasonal to multi-annual timescales. During prolonged periods of stress, carbon utilisation rates can diverge significantly from photosynthesis implying that plants rely heavily on their NSC reserves during these periods. Without simulating NSC storage, LSMs remain unable to capture this asynchrony between GPP and PCE and thus fail to correctly simulate forest-level respiration and growth fluxes.

The ability to sustain respiration and growth during periods of reduced productivity is an important process that can allow plants to survive and recover from extreme short-term climate events, such as drought . Under low water availability the transport of water from roots to other organs can be compromised by the cavitation of xylem tissue in the plant . Xylem damage can lead to a drop in hydraulic conductance, resulting in damage to plant tissue and increased risk of mortality . Plants combat this threat through control over the aperture of their stomata. Closing the stomata reduces water loss through transpiration and lowers the risk of xylem damage and hydraulic failure. The trade-off to this strategy is a reduction in productivity. The ability of a plant to employ this strategy is therefore reliant on its ability to store and utilise NSC. If carbon demand exceeds supply over long periods of drought, NSC reserves become exhausted, causing essential elements of plant function to fail, a process termed “carbon starvation”. Carbon starvation and hydraulic failure are tightly linked processes , not only because of their shared dependence on stomatal conductance but also due to the role that carbohydrates have in processes such as osmoregulation and, potentially, in refilling of embolised xylem . Carbon starvation may accelerate the effects of hydraulic failure and, in some cases, itself lead directly to mortality . Recent developments in modelling plant hydraulics provide more accurate predictions of stomatal behaviour during drought; however, these developments must also be accompanied by models of carbon storage in order to effectively simulate the trade-off between hydraulic damage and productivity loss. Until such developments are made, predictions of plant mortality and recovery in response to climate extremes such as drought will remain uncertain.

Accurately simulating forest mortality is vital to accurate predictions of climate. This is particularly true in tropical regions where terrestrial carbon storage is large and forests are frequently subjected to intense periods of environmental stress. Intense dry periods can reduce vegetation productivity and increase plant mortality in the tropics, over both short-term and multi-annual timescales . When combined with the effects of fire and land-use change, drought can cause regions such as the Amazon basin to shift from a net sink to a net source of carbon to the atmosphere . Loss of terrestrial carbon in the Amazon represents a significant feedback loop in the climate system , and large losses of biomass could cause drastic and irreversible changes to the climate. However, the nature of this “tipping point” is uncertain, and without accurate representation of forest resilience, including the balance between hydraulic failure and carbon starvation, predictions of large-scale forest die-back will remain unreliable. Drought is predicted to increase in both frequency and severity across the tropical rainforest biome in response to climate change . Accurately simulating drought responses is therefore a priority for the global modelling community , although many efforts to date have focused on simulating plant hydraulic properties and have largely ignored the development of a NSC pool in models.

Despite their clear role in forest function, our current understanding of how NSCs are produced, stored and used remains poor . Absolute pool sizes are difficult to quantify , and it is not clear how NSC reserves are distributed and transported between different plant organs under stress . It is also not clear whether NSC storage is the passive result of asynchrony between supply and demand, as described above, or whether plants also have the capacity to actively regulate NSC stores at the expense of growth and respiration . This may go some way toward explaining the apparent absence of substrate-based modelling approaches within many LSMs. Some optimised modelling studies have been conducted that explore models of NSC storage and the substrate limitation of respiration and growth . These provide a theoretical framework to develop mechanistic models of NSC storage and utilisation that allow detailed simulations of plant function. However, there have been few attempts to develop such models in a manner that would be compatible with large-scale LSMs . This can largely be attributed to a scarcity of ecosystem level data (NSC content and distribution) that can be used to parameterise and evaluate models for a range of species and climates that covers all plant functional types (PFTs) used in LSMs . Site-level studies that explore how the components of plant carbon expenditure respond to environmental change e.g. provide useful insights into the role of NSCs within a plant and can guide model development. Nonetheless, given our current knowledge and data availability, it is necessary to develop a parameter-sparse model that can be calibrated against data sources that can be more effectively collected (e.g. growth and respiration data) and yet capture the essential characteristics of representing a NSC pool (e.g. decoupling photosynthesis from growth and respiration). Such an effort will not only constrain future climate projections but may also be used to stimulate further research that improves our empirical understanding of NSC storage and use.

In this study we present “Substrate Utilisation by Growth and Autotrophic Respiration” (SUGAR), a simplified model of substrate utilisation, designed to work within an LSM. The aim of the model is to allow the decoupling of PCE and GPP in order to provide a more accurate representation of respiration and growth fluxes, in particular in response to environmental stress. To demonstrate its behaviour and applicability to large-scale ecosystem modelling, we use SUGAR to simulate PCE fluxes over the Amazon basin, using GPP data from an ensemble of LSMs, constrained by global fluorescence measurements from the Greenhouse Gases Observing SATellite (GOSAT) as driving data. We assess the sensitivity of the model to initialised NSC content within a reasonable range of possible pool sizes and assess the changes the model makes to predictions of ecosystem carbon expenditure. We also test the model under stressed and non-stressed conditions by simulating the world's longest-running tropical rainforest throughfall exclusion (TFE) experiment and corresponding control forest in the Caxiuanã national forest, Brazil, over a 16-year period. Previous simulations of the TFE experiment by multiple LSMs have highlighted their inefficiency at capturing the effects of the artificial drought on forest function . It remains unclear to what extent the lack of NSC dynamics is responsible for the discrepancies between model predictions and observations in these previous studies. We examine the role NSC dynamics has on model predictions during the drought by post-processing the output of one of these LSMs, namely the Joint UK Land Environment Simulator (JULES). We compare the results from JULES and the new predictions from SUGAR to observations and a time series of net primary productivity (NPP) derived from data collected in .

Our “Substrate Utilisation by Growth and Autotrophic Respiration (SUGAR)” model simulates a single pool of carbohydrate at a grid box scale for each vegetation tile (Fig. ). Sugars and starches are not distinguished, meaning that all carbohydrate is readily available to support respiration and growth. Representing just a single pool in this way keeps the model simple and parameter sparse making integration into an LSM much easier. SUGAR is designed to sit below the photosynthesis component of a LSM. Assimilated carbon from photosynthesis (GPP) is collected by the NSC pool and the total carbon allocated to respiration and growth is then calculated and taken directly from the NSC pool. The pool is therefore always active and is constantly depleted by growth and respiration and replenished by photosynthesis. Both growth and respiration are assumed to be single-substrate enzyme reactions and depend on NSC content via the Michaelis–Menten equation. Respiration and growth both depend on temperature via the standard Q10 function . Carbohydrate content is not actively regulated by the plants in SUGAR, meaning that variations in NSC stores are the passive result of asynchrony between photosynthesis and PCE caused by variations in climate.

Detailed time series data of forest-level NSC stocks are extremely difficult to collect and are therefore scarce. This makes parameter evaluation difficult. Here we discuss the evaluation process for each parameter in SUGAR. Some of these parameters, for example the q10 parameter, have standard or commonly used values within the LSM literature, and the validity of their given values is beyond the scope of this paper. In these cases we present only a very brief justification. For the remaining parameters, we outline how with a few assumptions we can use the simplicity of SUGAR to evaluate these parameters without the need for detailed NSC data and instead use more commonly and readily measured variables. An overview of all parameters, their default values and the values used in this study is given in Table .

q10. The q10 parameter represents the factor by which respiration and growth increase with every 10 ∘C of warming. The exponential Q10 function is commonly used to describe the temperature dependence of plant metabolism in LSMs with the standard q10 value of 2.0 .

SUGAR alters the relationship between photosynthesis and carbon expenditure. This has implications for simulations of both extreme and more gradual changes in climatic and meteorological conditions. By decoupling PCE from GPP, SUGAR creates a buffering effect that decreases the seasonal variation in carbon expenditure, even in ecosystems where the variation in GPP is already low. As we increase the levels of stored substrate within our simulations, the variability in PCE declines, due to an increased ability to maintain respiration and growth when GPP is low, and replenishment of the NSC pool when GPP is high. This effect is most pronounced in the semi-arid regions of the southern Amazon where there is a strong seasonal cycle in GPP (Fig. ), corresponding to a strong seasonal pattern of precipitation. Semi-arid regions provide the largest contribution to the global carbon sink anomaly, in part due to this high variability in GPP . To represent this contribution, land surface models must capture the response of vegetation to the climate variability experienced in these regions now and in the future. SUGAR provides a mechanistic approach to achieve this by simulating respiration and growth as a separate function to GPP. Given the strong evidence from observations that NPP and respiration do not have the same seasonal and climatic responses as GPP , accurately predicting future variability in atmospheric CO2 concentrations will be reliant on a sub-model such as SUGAR that can allow this decoupling to occur. Research demonstrating the importance of highly seasonal arid regions highlights the necessity of substrate-based approaches in large-scale ecosystem models and should motivate the community to focus on improving our understanding of NSCs and how to model them.

The sensitivity of the biosphere to climate change has large impacts on the future climate. For example, large losses of tropical forest carbon may represent a tipping point in the climate system that could have highly adverse and irreversible consequences for the global climate . However, both the nature and likelihood of such a tipping point is uncertain. Feedbacks between the climate and the carbon cycle mean that small perturbations in the state of the biosphere can make significant changes to the future state of the climate . Small changes in the sensitivity of a tropical forest to climate change may be the difference between the continued absorption of CO2 by ecosystems such as the Amazon and the severe die-back scenarios predicted by some models . Therefore, the difference between a forest that is able to buffer the effects of even a short drought or reduction in productivity and a forest that is not, may be significant at a global context in the future, even if it appears small in the present day. Non-conservative propagation of perturbations in the state of vegetated ecosystems contributes to large uncertainty in climate models , which greatly reduces our ability to constrain future climate possibilities and tipping points within the carbon cycle. Accurately representing the response of forest biomass, particularly in the tropics, to changes in climate is crucial to reducing this uncertainty and is a major goal of the climate and land surface modelling community. The buffering effect demonstrated in SUGAR may have an indirect yet large impact on the predictions of future climate by LSMs and provide a more realistic representation of forest sensitivity to climate.

As well as a buffering of carbon expenditure, SUGAR also enables a transition of the primary driver of growth and respiration. With little or no carbohydrate, carbon expenditure in SUGAR is driven predominantly by the rate of photosynthesis (Fig. ). Carbon is used by the ecosystem as soon as it is assimilated, meaning that the rate of expenditure is highly correlated with the rate of photosynthesis. This is often described as “source-driven carbon dynamics” meaning that photosynthesis is the key driving flux in determining the carbon balance of the ecosystem. “Source-driven carbon dynamics” are at the centre of many LSMs, including JULES. As more carbohydrate is added to the ecosystem in SUGAR, temperature becomes the predominant driver of PCE via the Q10 function (Eq. , Fig. ). As more carbon is stored, growth and respiration become less carbon limited and more controlled by the Q10 function within SUGAR. This shift can be seen as a transition towards “sink-driven carbon dynamics”. Under the theory of sink-driven carbon dynamics, environmental variables such as temperature and water availability exert a direct control over carbon expenditure that can be larger than that of photosynthesis . Processes such as end-product inhibition , in which photosynthesis is inhibited by an excess of assimilate in the leaves, mean that growth and respiration may even exert indirect control over the rate of photosynthesis. The result is that “sink” fluxes (i.e respiration and growth) driven by environmental variables are the predominant determinants of ecosystem carbon balance. Since the NSC pool in SUGAR does not exert any control over photosynthesis (e.g. via end-product inhibition), the behaviour of SUGAR here cannot be described as truly sink driven. However, SUGAR provides a framework that allows processes such as end-product inhibition to be implemented and thus provides the opportunity to represent both sink- and source-driven dynamics in LSMs. This allows a greater representation of how the limiting factors of growth and respiration interact with and respond to a changing climate.

Using the Caxiuanã control simulations we demonstrate that SUGAR and JULES predict very similar long-term NPP accumulation in the natural climate conditions of a tropical moist forest. However, there are larger differences between SUGAR and JULES on an annual timescale, due to the buffering of the natural variability in GPP by SUGAR. These results further highlight the importance of substrate-based modelling to better capture the responses to natural variation, even under current climate conditions and without extreme events . In the TFE plot, SUGAR makes significant improvements to the prediction of ecosystem carbon fluxes, particularly for accumulated NPP. This improvement is caused by a combination of two processes that occur in SUGAR and that are not present in JULES. The first process is the utilisation of the NSC pool during the early stages of the experiment. SUGAR expends a mean 5.53 Mgha-1 more carbon than is assimilated through photosynthesis in the first 3 years of drought (2002–2004) and a further 5.80 Mgha-1 in 2005. This allows an increase in both NPP and respiration relative to JULES and is consistent with the analysis in , which suggests the TFE plot was expending 7±4.5 MgCha-1yr-1 more than it was accumulating in 2005, implying that NSC stores were being depleted in response to the drought. The second process is the down regulation of respiration in response to the depleting NSC pool. In the JULES simulations, photosynthesis declines much faster than respiration and, since growth is equal to GPP–Ra in JULES, this means that NPP drops significantly as GPP declines in response to the drought. The result of this effect is that in 2 years (2005 and 2007) the predicted annual mean NPP by JULES is negative. Negative NPP is generally considered to be unrealistic, particularly over the timescale of a year , and since JULES does not contain a labile carbon pool to support the deficit, missing carbon is taken from the structural pool. The physical interpretation of this is that trees in JULES respire away their structural carbon and shrink. While there is some evidence of recycling and remobilisation of structural compounds, the magnitude of structural carbon being allocated to respiration (via the resulting negative NPP) in these JULES simulations is not realistic. In SUGAR, respiration declines due to the depletion of the NSC pool. This down-regulation of Ra means that a larger proportion of instantaneous GPP is available for NPP, resulting in larger predictions of NPP in SUGAR than JULES, despite similar estimates of total PCE. While this latter process aids the prediction of NPP in SUGAR, it should be noted that observations from Caxiuanã actually indicate an increase in TFE plot respiration between 2005 and 2011 . SUGAR is currently unable to capture this increase and this is likely due to the simplicity of the assumptions made within the model. For example, we have assumed that plant growth is directly dependent on carbohydrate availability and temperature only. Water stress may reduce plant growth in SUGAR but only indirectly by inhibiting photosynthesis and causing a decrease in available carbon. However, in reality plant growth can be affected directly by decreasing water availability through the inhibition of cambial expansion . This decline in growth may even occur before declines in photosynthesis, which can cause a build-up of NSC and eventually result in an increase in respiration . We are not suggesting that this specific process explains the observed increases in respiration on the drought plot at Caxiuanaã, but such interactions between NSC utilisation and the environment are likely to have been important during the TFE experiment. Neither SUGAR nor JULES are able to capture these processes currently. However, by implementing SUGAR within JULES we create a basis upon which we can start to represent these interactions and continue to improve predictions of forest responses to drought.

The ability of SUGAR to accurately capture PCE responses to drought in these simulations is also somewhat limited by the GPP used to run it. Photosynthesis in JULES has a high sensitivity to reductions in soil moisture e.g.,. In the Caxiuanã simulations JULES predicts an average decline in annual GPP of 4.42 MgCha-1yr-1 from 2001 to 2005 in the TFE plot. Combining the observed PCE rates in the TFE plot with the predicted GPP by JULES would imply that the forest is using an average of 10.96 MgCha-1yr-1 carbon more than it is assimilating in the first 4 years. This would then imply that the forest has access to at least 43.86 MgC ha-1 of NSC, ca. 22 % of estimated forest biomass. Such a high NSC content is unlikely for tropical forests, which are more likely to have reserves close to ∼10 % . The other more likely explanation is that JULES is overestimating the decline in photosynthesis in response to the drought. To test this, we artificially reduced drought stress in JULES by 50 % and repeated the Caxiuanã simulations (Figs.  and ). This improved predictions of PCE in both models, supporting the hypothesis that JULES overestimates the sensitivity of photosynthesis to drought at this site. The recent work to improve stomatal responses to drought stress has the potential to significantly improve GPP predictions in LSMs such as JULES. However, there is a clear link between hydraulics and labile carbon storage, given stomatal closure comes at the cost of a reduction in carbon assimilation. The ability of a plant to store and use labile carbon is crucial to its ability to survive and recover from drought-induced stomatal closure . Without including at least simple representations of NSC storage, the potential of this recent work to improve the representation of stomatal behaviour in response to drought in LSMs is unlikely to be realised.

SUGAR is a purposefully simple model of NSC storage and is missing some key processes known to be important in defining the complexities of NSC storage and use within a plant. A more complex NSC model might, for example, distinguish between starch and sugar pools or represent multiple pools for each plant organ and actively control the input or output of NSC into pools . However, such models would likely require representation of substrate transport between pools and the scaling of NSC data to the level of trees and forests. Recent advancements in measurement protocols may allow these datasets to be reliably collected , however, previously the level of uncertainty on such figures has been up to 400 % . As a result, comprehensive NSC datasets, measured through time in response to climatic variations and across enough biomes to allow all model PFTs to be evaluated, are currently not available. Therefore, this is not a currently viable way to constrain model output. SUGAR is designed to break the direct link between PCE and GPP found in many LSMs and to provide more mechanistic predictions of growth and respiration. It can be parameterised, initialised and evaluated with data that is commonly collected across the globe – biomass, GPP, and temperature; CUE (to find α); and respiration and NPP (for evaluation). It also requires an input of initialised NSC fraction (fNSC), which is not easily measured for an ecosystem, although values of fNSC can be constrained within sensible bounds . It may also be possible to use SUGAR as a tool to further constrain observed values of NSC content by conducting sensitivity studies of fNSC. Given the existing level of knowledge, it is more robust and realistic to use a simple model such as SUGAR, which can be evaluated against more easily available observations such as Ra, PCE, NPP and GPP. As the accuracy and spatial extent of NSC data grows, models such as SUGAR can act as a simple skeleton that allows new processes to be implemented into LSMs to more accurately represent the complexity of plant carbon storage and use.

We have developed a simple model of NSC storage that is designed to be integrated into an LSM. The model makes significant changes to the variability of growth and respiration predictions in both extreme and more stable climatic conditions. This has large implications for simulations of future climate given the importance of predicting the variability of atmospheric CO2 concentrations. The model also allows a more mechanistic representation of the limiting factors of carbon expenditure, which may become increasingly important as the climate changes in the future. Due to the simplicity of the model, it is easily parameterised using pre-existing data and does not require complex datasets of NSC storage that are currently unavailable. This makes the model attractive since it can be easily integrated into LSMs without introducing unreasonable uncertainty in parameter values. The magnitude of the change demonstrates the importance of representing carbon storage in LSMs, and we hope this will motivate both the modelling and empirical communities to further develop our understanding and model representation of NSC dynamics.