AMAZALERT Delivery Report Report on 4 focus areas of improvement in models. Title Modelling Amazonian carbon budgets and vegetation dynamics Work Package Number 2 Delivery number 2.3 First author Bart Kruijt (ALTERRA) Patrick Meir Michelle Johnston Anja Rammig Sophie Fauset Co-authors Tim Baker David Galbraith Celso von Randow Hans Verbeeck Date of completion 25 March 2013 Name leading Work Package Leader Bart Kruijt Approved by the Leading Work YES Package Leader To complete by the Coordinator Approved by the Coordinator YES Date of approval by the Coordinator 4 April 2013 AMAZALERT, No 282664 Deliverable 2.3 Modelling Amazonian carbon budgets and vegetation dynamics Bart Kruijt1), Patrick Meir2), Michelle Johnston3), Anja Rammig4), Sophie Fauset3), Tim Baker3), David Galbraith3), Celso von Randow5) and Hans Verbeeck6) (To be submitted as book chapter to LBA Springer book) 1) ALTERRA, Wageningen University and research centre, Wageningen, Netherlands 2) University of Edinburgh, Edinburgh, Scotland, UK 3) University of Leeds, Leeds, UK 4) Potsdamm Institute for Climate research (PIK), Potsdam, Germany 5) Brazilian national Space research Institute (INPE), Cachoeira Paulista, SP, Brazil 6) University of gent, Gent, Belgium Executive Summary Modelling the Amazon terrestrial carbon budget and vegetation dynamics is, despite the large amount of research on dynamic vegetation models and land-atmosphere exchange, still at an early stage in comparison to the development of global climate models, to which DGVMs are ultimately coupled to generate Earth system simulations. The principal uncertainties deriving from model analyses have focussed on a lack of knowledge on the response of tropical vegetation to increases in temperature and atmospheric CO concentrations, and to a lesser extent, upon reduced water availability. The extent to 2 which water availability is important is also model dependent (Rammig et al. 2010, Galbraith et al. 2010, Huntingford et al, 2013). The size of the predicted effect of moisture stress on Amazon vegetation is dependent on both the forest response and the future risk of drought, which varies among climate models (Meir and Woodward 2010). However, if the risk of drought is kept constant, DGVM sensitivity to moisture stress appears frequently to be smaller than, for example, sensitivity to warming (Galbraith et al. 2010), even though recent empirical evidence demonstrates that drought can have large impacts on the forest carbon cycle (Phillips et al. 2009; da Costa et al. 2010); hence this area may need revisiting in the future. These environmental factors directly affect primary productivity and respiration, the main components of the carbon budget of a forest ecosystem.. At longer time scales, the indirect effects of carbon allocation (eg root-shoot –leaf partitioning) and demography (recruitment, longevity, mortality) might be of equal importance to determine vegetation biomass and carbon storage. Again however, there is limited information specific to the Amazon, especially upon how forests vary with climate and soils, although this picture is likely to change as present day allocation patterns are investigated in more detail (e.g. Malhi et al. 2009). The uncertainty on moisture dependence partly stems from lack of spatial information on water supply parameters: soil hydraulic properties, including hydraulic lift; rooting depth and root dynamics. Apart from that, knowledge is only now being expanded on the variability in water stress response among species and groups of species. The latter affects, apart from ecosystem water use, the demography within ecosystems: drought-limited establishment and drought-induced mortality differ among these groups and vulnerability to increased mortality during drought has been shown to be size (or perhaps age) dependent (Engelbrecht et al. 2007; da Costa et al. 2010; Phillips et al. 2010. The effects of water stress on plant respiration are still poorly understood. Although the majority of evidence from non- tropical studies indicates a decline in plant respiration following short term reductions in moisture availability (Atkin and Macherel 2009), there are indications that it is enhanced under long term water stress (Metcalfe et al. 2010, da Costa et al. in press), and following seasonal drought in Amazonian rain forest (Miranda et al. 2005). 2 AMAZALERT, No 282664 Deliverable 2.3 To better represent water stress, systematic efforts are needed to map soil hydraulic properties and root distribution across the Amazon (Fisher et al. 2008). Plant response to soil water and atmospheric demand need to be inventoried for a range of ecological groups, and if the data support categorisation, for example of the sort observed in mortality trends (da Costa et al. 2010), distinct response modes need to be parameterised, informed by data obtained at canopy, tree and leaf scales. Models need to be made suitable to simulate water supply, root dynamics, allocation and use of carbon resources, and stomatal response by a focal set of ecological groups; data-model comparison should then be used to parsimoniously identify the most important processes. The uncertainties in the temperature response of photosynthesis and respiration mainly stem from lack of information on the shape this response, specific of the Amazon (or indeed, any tropical forest region), in particular with respect to minimum, optimum and maximum temperatures (Atkin et al. 2008 GCB, Smith and Dukes 2013). It is largely unknown how these response curves in turn are likely to acclimate, , to long-term changes in average temperatures or moisture availability. Most models, in their current form, rely on parameterisations from earlier work in temperate zone vegetation. There is some concern that they combine sets of parameters from different experimental data sets, where different model formulations were used, potentially leading to erroneously simulated temperature response irrespective of climate zone. Temperature also affects stomatal conductance and VPD, which interact to affect net photosynthesis (Lloyd and Farquhar 2008), so care must be taken to distinguish temperature response of overall leaf or canopy productivity from temperature response of individual model parameters. To reduce uncertainty in the temperature response of Amazon vegetation productivity, in-situ temperature responses of the model parameters involved in simulating photosynthetic capacity and respiration need to be determined. Canopy-scale data, such as from eddy correlation can be used but especially here it should be taken into account that these show overall response, including stomatal response and soil respiration. At the leaf scale, detailed response properties of photosynthesis and respiration parameters can be determined, as can the impacts on gas exchange capacity of long-term experimental manipulation of leaf temperature. Here, of course, there are important issues relating to the need to represent diversity across the whole ecosystem and indeed the wider Amazon basin. Linking warming responses in vegetation to those below ground presents a bigger experimental challenge, although translocation studies within rainforests at different elevations provide some insight (Zimmermann et al. 2010), as may future ecosystem-scale warming experiments (Wood et al. 2012) The main modelling issue here is that models should be equipped to allow acclimation of temperature optima. Uncertainty in CO response is strongly linked to our understanding of its dependence on nutrient 2 availability, and over time, nutrient dynamics. As photosynthesis and water use become more efficient under higher CO , the rate of increase in productivity depends on nutrient availability and the allocation 2 of carbon to long and short-term residence pools. Given the relative abundance of nitrogen in Amazonian soils, this question primarily concerns phosphorus availability (refs) which is limited and dynamically dependent on soil properties and vegetation activity The challenge is to introduce a dynamic model for phosphorus availability in Amazon soils. Apart from the need for comprehensive mapping of nutrient availability across the Amazon (Quesada et al. 2010), , specific investigations into soil nutrient dynamics and plant-soil interactions are required to parameterise such model improvements. As much of the issues in modelling the Amazon carbon cycle and vegetation functioning revolves around the question whether or not important shifts in vegetation type, or degradation can be expected, it is important that simulation models can represent such transitions, either through soil and vegetation processes alone, or the interaction of fire and ecological process. In their current state, most models only represent two or three tropical vegetation types (Plant Functional Types, or PFTs), and consequently there is a risk that rapid and substantial shifts from, e.g. evergreen to seasonal forests or savannas are simulated prematurely, partly because of a lack of intermediate functional types which exist in the real world. Solutions to this that modelling efforts should aim at, include the introduction of 3 AMAZALERT, No 282664 Deliverable 2.3 more types or statistical approaches with more types co-existing in the same model grid points, but perhaps more fruitfully models could represent the abundance and range of plant traits, or ranges of species groups, rather than discrete PFTs. The carbon storage capacity of an ecosystem depends both on the short-term budget of carbon, as discussed above, and on the subsequent fate of sequestered carbon (or: NPP). Carbon that is allocated to sort-lived or easily decomposable tissues will disappear from the ecosystem much more rapidly than long-lived carbon components. Therefore, understanding allocation is equally important as photosynthesis and respiration. Allocation is, however, much less well-understood than the primary production processes. Several approaches exist to model it, varying from fixed partitioning, via resource limitation dependency, to finding evolutionary optimality. If compared to data sets from the Amazon, which are very scarce, these models usually show poor agreement, although the evolutionary class seems to perform best. Data show that the strongest trade-off in allocation seems to be between aboveground wood an (fine) roots. What is needed for modelling response to climate change, is models in which allocation can also dynamically change and adjust (or adapt) to changed climate. Such change will affect the carbon retained in the ecosystem. Plant mortality and recruitment can be seen as ultimate allocation (to litter and to nxt generations, respectively), however, the factors driving these processes are very different from those affecting allocation in live plants. Mortality may be associated with (or negatively correlated with) the investment of carbon into wood and roots, i.e. in resilience to enhance longevity at minimum ‘cost’. Partly this is related to the same environmental factors, such as drought, but for mortality also windthrow is important. Being rather random and indiscriminate, wind can have strong disturbing effects. These demographic processes are among the least well represented in DGVMs, even though their effect of ecosystem carbon is of the same magnitude as primary production an allocation. One of the problems for models is that they work with groups of plants or whole ecosystems, and at best deal with separate groups or cohort within vegetation patches. More individual-oriented models would offer advances in demographic effect. Fire, in turn is an ultimate form of disturbance, and can act as the defining switch for vegetation change from closed forest to degraded vegetation. The occurrence of fires can be self-reinforcing, as initial understory fires and localised lead to dry-down, opening up and increase of edge length. All these enhance the probability of ignition of fire. Observation techniques from remote sensing to detect, for example, understorey fire are improving, and several fairly detailed models exist for fire susceptibility. However coupling these new techniques to DGVMs or global climate models still proves a challenge, and modelling the spread of fire is also still an important issue to resolve. A promising way ahead is, apart from coupling fire models to DGVMs and coupled earth system models (including climate), is to link fire models to land-use change models. The capacity to reliably represent Amazon forest biomass and vegetation in models has definitely improved greatly over the last decade. Also, our insight into which factors are most important to sensitively evaluate climate change impact has improved, and it can be expected that these insights will soon lead to much more reliable DGVMs. Lack of data, both process-oriented experiments and ecosystem-scale validation data, are stil sparse, however, so that undcertainties will remain substantial in the foreseeable future. Nevertheless, we are confident that given experimental effort and modell development, within a few years we will be able to assess the effects of climate change on the Amazon for the upcoming century with reasonable confidence. 4 AMAZALERT, No 282664 Deliverable 2.3 Introduction The forests of the Amazon region are under threat of both climate change and land-use change, with risks of accelerated degradation involving positive feedbacks through moisture, CO and temperature 2 (Davidson, 2012). The notion of accelerated degradation under 21st century climate change was brought into focus first by White et al. (1999) and Cox et al (2000), followed by a series of studies showing inter-model variability in this climate sensitivity (e.g., Friedlingstein, 2006, Nobre and Borma, 2009, Sampaio et al, 2007). Simulated changes in biomass depend on a number of factors. First, the emissions scenario behind the simulation, second the climate model, third, the vegetation model used and finally the inclusion of land-surface feedbacks. The early studies relied on one single climate model (HADCM3), which was severe in its forecasts and one surface model (MOSES-TRIFFID) which was very sensitive (Huntingford et al, 2008, Galbraith et al, 2010). The Recent studies appear to indicate a more modest sensitivity (Good et al, 2013; Cox et al, 2013; Huntingford et al, 2013). In this review we focus on the vegetation component, which is substantial. Some recent studies have highlighted the strong sensitivity to CO2 and temperature in vegetation models used to make these predictions (e.g. Lapola et al. 2009, Galbraith et al. 2010, Huntingford et al. 2013). The degree of sensitivity, especially with respect to temperature, differs among models, in particular the Dynamic Global Vegetation models (DGVMs), and the way different processes are represented in them. Moreover, although a subject of discussion (Poulter et al. 2010), recent work suggests that the uncertainty associated with the physiologically-driven ecosystem-scale responses in the models is higher than the uncertainty associated with future climate projections (Huntingford et al. 2013). Increased atmospheric CO 2 concentration potentially reduces water stress, but also may lead to changes in vegetation structure and competition. The effects of CO on tropical vegetation, however, remain largely unmeasured, especially 2 because limitations by nutrients and temperature are poorly understood and also because little is known about how enhanced productivity might affect growth patterns and demography (allocation, recruitment, ageing and mortality). The effects of changing temperature on the balance of primary productivity (photosynthesis), respiration and decomposition are poorly understood for the tropics, with most available information coming from temperate vegetation and agricultural crops. Several efforts have been made to develop and compare appropriate models to simulate the carbon budgets, seasonal variability, and climate sensitivity of the region. A recent effort coming into publication soon, within the scope of the Large-Scale Biosphere-Atmosphere experiment in Amazonia (LBA) is the LBA-MIP (Model Intercomparison Project, Gonçalves et al., 2013, Von Randow et al, 2013, submitted). Other studies include a World Bank funded initiative (Vergara and Sholz, 2010), the Moore foundation funded Amazon-Andes Initiative (Moorcroft et al in prep.), and AMAZALERT (Kruijt et al, in progress). A recent collection of studies on the climate sensitivity of the Amazon can be found in New Phytologist journal (Meir et al, 2010). The publications in this special issue focused on drought sensitivity, but also pointed out other, sometimes larger sensitivities in the relevant models, such as the simulated responses to CO and temperature (Rammig et al, 2010, Jupp et al, 2010, Galbraith et al, 2 2010, also see Poulter et al, 2010). Recent reviews covering the importance of a better understanding of dynamics dependence of ecosystem productivity on environmental factors and climate change include Booth et al. (2012) and Smith and Dukes (2013). Dufresne et al (2002) noted that the Cox et al (2000) predictions of amazon dieback depend strongly on allocation of the extra carbon gained by increasing CO to vegetation versus soil pools, affecting 2 respiration rates, explaining some of the differences between the HADCM3 simulations (Cox et al, 2000) and IPSL simulations (Friedlingstein, 20060. In general, DGVMs are to date not very successful in reproducing the observed spatial variability and biomass over the Amazon basin (see AMAZALERT Mid- term report and Delbart et al, 2011, Castanho et al, 2013). A more detailed representation of the vegetation dynamics might enable us to simulate processes as mortality , disturbance and recruitment in a more realistic way. This could possibly lead to better simulations of spatial variability in biomass. 5 AMAZALERT, No 282664 Deliverable 2.3 This review aims to give an overview of the most important issues concerning the modelling of carbon budgets and vegetation dynamics of the Amazon forests, and tropical forests in general. Subsequently, we briefly cover water relations, temperature dependence, CO and nutrient dependence and patterns in 2 growth and mortality. The review should also assist to set the agenda for model improvement and data needs, to adequately equip global and regional dynamics vegetation models in assessing climate sensitivity of the region’s vegetation. Key issues in DGVMs for the Amazon Soil-plant water relations In the Amazon, flux tower data, data from manipulative experiments and remote sensing indices show unexpected responses of vegetation properties and carbon fluxes to dry periods. Where soils are deep and the dry season moderate, productivity appears to be hardly affected by drought, and even to peak during the dry season (Saleska et al, 2003; Fisher. 2006. Originally considered largely a-seasonal in terms of productivity, Malhi et al, (1998) showed seasonality in NEE at a Central Amazon site near Manaus. This variability in apparent carbon uptake was correlated with soil moisture, especially where during the dry season where uptake was reduced; the moisture constraint at this site imposed by the specific soil characteristics was considered further by Fisher et al. (2008) and shown to be related to soil properties as well as climate. Malhi’s (1998) results, however, appear to be somewhat anomalous, as for example Araujo et al (2002) found little seasonal variation in CO fluxes. Following the early 2 measurements near Manaus it was suggested that Amazon rain forest carbon uptake tends to be higher rather than reduced and vegetation tends to ‘green-up’ during the dry season or late dry season. This was supported by eddy correlation flux measurements (higher NEE) and by the analysis of satellite reflectance data (MODIS EVI) (Saleska et al. 2003, Saleska et al,2007, Huete et al, 2006). The latter large scale analysis, however, was later criticised because EVI values can be affected by changes in (dry season) canopy structure (Anderson et al, 2010) and by imperfect correction for clouds and aerosols (Samanta et al., 2010, 2012). The phenomenon, where observed at individual flux sites, has been explained by vegetation being relatively tolerant of normal dry season conditions because of deep root soil water access and soil moisture storage combined with higher insolation and reduced litter decomposition and respiration during the dry season (Fisher et al. 2008, Saleska et al. 2003, Bonal et al, 2008). These studies demonstrate variability in the responses by GPP and respiration processes to seasonal rainfall that is not represented in large-scale and global models. In recent years several large scale models have been altered to improve the simulated seasonality of carbon fluxes for flux tower sites by introducing improved equations (e.g. Baker et al. 2008) or by optimising model parameters (e.g. Verbeeck et al. 2011). The response to severe or extended drought has been studied further, using two rainfall exclusion experiments in Amazonia (Brando et al, 2008; Davidson et al, 2008; Da Costa et al, 2010), and also observations of natural forest growth and mortality following the extreme natural drought of 2005 in the region (Marengo et al. 2008, Phillips et al. 2009). Under two multi-year, large scale experimental drought treatments , photosynthesis, transpiration and/or biomass increment responded sharply to the onset of artificial soil moisture reduction, while tree mortality increased substantially only after about three years of the drought treatment (Fisher et al. 2006 Brando et al. 2008, da Costa et al, 2010). The severe natural drought of 2005 in Amazonia included atmospheric as well as a soil drought effect; it also impacted mortality substantially, but over a shorter timescale of one year, causing an estimated overturn of the preceding regional carbon sink, as calculated using data from the region-wide inventory plot network, Rainfor (Phillips et al. 2009). In both types of studies (experimental and observational), bigger trees were affected disproportionally (da Costa et al. 2010, Phillips et al, 2010). Hence, current understanding suggests that the response by rainforest ecosystem productivity to seasonal drought varies over time and space, dependent partly on soil conditions, time scale and drought severity, as well as upon differences among species in vulnerability to drought (Van der Molen et al, 2011, Fisher et al. 6 AMAZALERT, No 282664 Deliverable 2.3 2006, da Costa et al. 2010). The impacts of drought on tree mortality are treated in a subsequent section; here we address the direct effects of water stress on GPP and transpiration. Reduced water availability affects photosynthesis through its effect on stomatal aperture, which modulates not only water loss but also CO uptake. At longer (leaf ontogeny) time scales, water stress 2 can also affect leaf structure and, for example, mesophyll conductance, directly affecting photosynthetic capacity (Egea et al, 2012). In simulation models, the effects of water stress can be approached from two sides: from the demand imposed by the atmosphere and the leaves on the plant’s hydraulic systems and the soil, and from the supply of water from the soil through the rooting system to the leaves and atmosphere. In models, demand and supply have to be matched somehow. The atmospheric demand is can be represented as a radiation-dependent potential evapotranspiration with implicit boundary-layer (Priestly-Taylor or equivalent, Monteith, 1995) or by only considering exchange with the lowermost layer of the atmosphere (through a Penman-Monteith equation or explicit vapour gradient-diffusion). In both cases, the demand is modulated by the surface (or stomatal) conductance, through which (water and CO ) demand is made to match the supply. Most modelling effort in the past 2 has gone into finding efficient representations of stomatal response to water stress and CO demand. 2 Broadly speaking, the approaches have been to: (i) consider stomata to respond in a linear multiplicative model to a range of independent environmental factors, where each response is determined by a set of parameters (eg Jarvis 1976; Stewart 1988); (ii) to maintain a fixed ratio of conductance to photosynthesis and humidity (or internal to atmospheric [CO ], Ball et al. 1987; 2 Leuning, 1995); (iii) to maintain a fixed, or optimal ratio of CO uptake to water loss (water use 2 efficiency, Cowan (1977); Medlyn et al., 2011), or to leaf water potential (the SPA model, Williams et al, 1998). The issues at the demand side are mainly related to the behaviour of the stomata. Different species or groups of species take different strategies in economising the rate of water loss per unit of CO gained. 2 This distinction is recently made between ‘isohydric’ and ‘anisohydric’ strategies, referring to those that tend to conserve water potential vs those that do not (and tend to conserve labile carbon), respectively (Fisher, 2006; van der Molen, 2012).To assess the productivity and survival of these different groups the surface conductance submodels of these need to be parameterised. It can be expected that more ‘conservative’ species tend to reduce conductance more quickly in the face of drought stress. These kind of necessary improvements in models are often parameter intensive. As DGVMs are typically run for many grid points, the number of site-specific parameters needs to be minimised. For this reason it is attractive to explore model formulations that rely on few parameters and, instead, assume interdependency or optimisation of photosynthesis and transpiration. A promising candidate is the class of models that optimise marginal carbon gain per unit water lost (Cowan, 1977, Groenendijk, 2011,Medlyn, 2011). The LPJ class of DGVMs already applies this approach in a simplified way. In this algorithm, stomatal conductance is reduced iteratively from an unstressed maximum, until water demand matches the supply from the soil. One of the major uncertainties is, however, how stomata respond to limited water supply, usually quantified as relative soil moisture availability or as water tension. Many models impose an empirical reduction function of soil moisture on surface conductance (e.g. ORCHIDEE, Verbeeck et al. 2011). Usually this function is highly non-linear, leading to reductions only at low values of soil water content. Other models explicitly model the hydraulic resistances of the plant and the soil matrix as an intermediate, affecting the stomata through the leaf water potential (SPA, Williams et al. 1998; Fisher et al. 2006, Medvigy et al. 2009, Christoffersen et al. in prep). These different approaches mainly affect the rate of soil drying and the reduction of transpiration as drought progresses. Whichever way, with extended drought periods soil moisture eventually gets depleted and then the critical issue is how roots access and potentially expanding the supply of available water. Adequate understanding and representation of root growth strategies is essential, especially specifying the maximum soil volume and depth that roots can access (Nepstad et al. 1994) and whether they can cope with low-oxygen groundwater conditions. What appears to be important here is information on soil hydraulic properties such as the pressure-volume relationship and hydraulic conductivity; such datasets are scarce for Amazonia. Belk et al. (2007) and Fisher et al (2008) illustrated the importance of how these parameters cause differences in the rate at which hydraulic resistance declines with soil moisture 7 AMAZALERT, No 282664 Deliverable 2.3 content, comparing sites that differed in dry season soil moisture stress. Although specific mechanisms such as hydraulic lift have been investigated in some modelling analyses (Baker et al. 2008), soil depth, the presence of ground water and potential for capillary rise are likely to be important integrating parameters for large-scale models (Christofferson et al. in prep). These aspects are poorly represented in most larger-scale models such as DGVMs, and they can also cause systematic biases in simulations. In terms of experimental work, to advance the issue of drought sensitivity in the Amazon, there is a need for more comprehensive basin-wide information on soil hydraulic properties, measured for a sufficient number of soil types. This would enable modellers to consistently project simulation across the basin. The strategies that roots of different ecological groups take to explore and enhance the supply of water are also important, though challenging to study. The latter could, however, be interpreted as part of the whole plant response to water stress, leading to different rates of stomatal response to drought in different ecological groups. Thus, stomatal response curves as a function of the soil water balance and atmospheric demand (e.g. VPD) for these groups could provide the empirical parameters that in fact are needed by most models. Such responses can be studied the leaf scale (using porometers or chambers), tree scale (using sap flow sensors) or whole-ecosystem scale (using eddy covariance data, with the provision that soil evaporation needs to be eliminated first). Creating and studying artificial drought, such as in the ‘Esecaflor’ plots of the Caxiuana reserve (Fisher et a. 2007, Meir and Woodward 2010, da Costa et al, 2010) and ‘Secaflor’ in the Tapajos area (Nepstad et al, 2002, Brando et al. 2008, Markewitz et al. 2010) greatly helps such analysis. Temperature dependence of GPP. The temperature dependence of photosynthesis in dynamic global vegetation models typically takes the shape of an optimum function, bell-shaped or composed of two exponential functions. Sensitive parameters typically include a minimum temperature, a maximum, an activation energy and an optimum temperature. Some models assume a wide, flat optimum range, others assume sharp optima. A range of typical temperature functions is shown by Galbraith (2010), who also shows that in many cases, model results on 21st century changes in productivity with climate depend more strongly on temperature than on moisture. Several studies have highlighted the strong sensitivity of DGVM- simulated GPP (and NEP) under climate change to assumed shapes of temperature dependence (Booth et al, 2012; Huntingford et al, 2013; Vermeulen, submitted) The temperature responses as represented in most current DGVMs are based on fixed parameter settings, determined about two decades ago in laboratory experiments on a few species only (eg Von Caemmerer et al. 1994, Bernacchi et al. 2001, Medlyn et al. 2002). Alternatively, optima have been tuned such that they represent average growing-season temperatures of temperate ecosystems, and have remained fixed or even hard-coded in the photosynthesis modules of models. An important question here, of course, is how plastic are such optima, and the associated second-order parameters such as minimum and maximum temperatures of the temperature response (Smith and Dukes, 2013)? In theory, as Lloyd and Farquhar (2008) argue, enzyme kinetics speed up exponentially with rising temperatures until a high-temperature cell lysis or enzyme denaturation point is reached. So, for enzyme-dominated processes such as carboxylation by RuBisCo it is unlikely to expect a temperature optimum in ambient conditions. The net effect of several complex processes interacting together, however, such as those involving membrane transport (photon absorption, electron transport, osmosis and active uptake) is more likely optimised to ambient temperatures and to be plastic (Lloyd and Farquhar, 2008) resulting in temperature-related plasticity for the overall process of photosynthesis The temperature response of photosynthesis depends on several parameters, all of which are temperature-dependent, both in reality and in most vegetation models. The much-used Farquhar equations depend on a maximum carboxylation rate, V and a maximum electron transport rate, J , cmax max both of which can vary strongly within and between ecosystems. There are also more intrinsic temperature-dependent parameters, such as the CO -compensation point in the absence of 2 8 AMAZALERT, No 282664 Deliverable 2.3 mitochondrial respiration, Г*, and the Michealis-Menten constants for carboxylation and oxidation, k c and k . The strong temperature dependence of these intrinsic parameters is hard to quantify o independently, and thus they are usually assumed and invariable with respect to other variation than temperature, with values taken from the literature (e.g. Bernacchi et al. 2001, Sharkey and Schrader, 2006). The formulations differ between sources, however, and affect the shape of the temperature dependencies of V and J as well, when these are fitted to data using the Farquhar equations. cmax max Although this may seem a technical issue, the consequence of this is also that it is important to distinguish in model-data evaluations whether the temperature dependence of maximum photosynthetic rate (A ) is considered or that of the underlying V and J parameters. Where the latter may show max cmax max sharp, peaked responses, the effective A might be rather insensitive to temperature as the variation max with temperature of the underlying parameters compensate.. Also, it is essential that parameter sets are consistent in models, i.e., all should refer to the same set of pre-assumed model parameters. These issues of relatedness of parameters are often overlooked (e.g. in Smith and Dukes, 2013). As a consequence of the assumed optimum functions, temperature dependence can be strong in DGVMs, and therefore it is very important that the associated parameters of temperature response curves are made realistic. Also, apart from any immediate temperature responses, it is important to assess how fast, if at all, these optima acclimate to changing ambient temperatures. Even then, we could distinguish acclimation of existing leaves and ontogenetic acclimation, i.e. where acclimation can only occur during the formation of new leaves (Smith and Dukes, 2013). At the time scale of interest in climate impacts on vegetation, genetic adaptation is not likely to be relevant, but competition between species or groups with different temperature responses is. To achieve such improvement in parameterisation of temperature responses, experimental data are needed. Long-term ecosystem flux data, with naturally co-varying temperatures and GPP estimates can be used to some extent, but the lack of distinction in these data between species or species groups, between respiration and photosynthesis, and the inevitable covariance between temperature and other environmental variables, limit their application. There is a need for photosynthetic temperature parameters determined at leaf-level, both for the immediate response and the long-term response. In a vast tropical rain forest domain like the Amazon it seems almost impossible to achieve any meaningful and representative quantification of this, but with limited work at a few sites it should at least be possible to test the implicit assumption that temperature responses and optima are fairly conserved across species and regions given similar climate. If variability within ecosystems in this is high about a mean value of limited change, then overall temperature sensitivity of ecosystem productivity will be limited, but in the long run, species composition may change as a consequence of different temperature dependences. Some work at plant and leaf scale is already under way. Along the Peruvian slopes of the Andes into Amazonia, gradient studies shed light on the plasticity of plant- and ecosystem GPP under different ambient temperature regimes (e.g. van de Weg et al. 2012, Girardin et al. 2010 Malhi et al. in review and unpublished). In central Amazonia Doughty et al (2008, 2010) has studied both immediate temperature responses and acclimation to elevated temperatures. It is, however, possible that in some of this analysis temperature response and response to correlated VPD change were hard to distinguish. It should be noted, that in analysing temperature or moisture response in models, correlated temperature and VPD also present a challenge. Concluding, more leaf-scale data of photosynthetic temperature response are needed. Studies should quantify the response of photosynthetic capacity (Vcmax, Jmax) and stomatal conductance rather than (maximum) net photosynthesis. Longer-term warming (or cooling) experiments should shed light on acclimation to extended periods of temperature change. Care should be taken that several species groups are considered. To capture the main ecosystem response, emphasis should be on leaves in the upper canopy of ecosystems or those exposed to direct radiation. 9 AMAZALERT, No 282664 Deliverable 2.3 Temperature dependence of respiration Although it is an integral part of ecosystem carbon budgets, in this section we offer only a brief outline of the issues concerning respiration in ecosystems. More elaborate analyses of the issues can be found elsewhere (e.g. Atkin et al. (2003, 2008), Meir et al. 2008, Chambers et al.2004 and Smith and Dukes (2013). With ‘respiration’ here we refer to all biological processes in ecosystems that lead to production of CO 2 derived through the oxidative reduction of organic material. The two distinct respiration pathways, autotrophic (plant) respiration and heterotrophic respiration (by consumers and decomposers of all kinds) are best modelled separately as environmental responses can be quite distinct. Few to no process-based models, for application at large scale, have been used to comprehensively represent respiration in forest ecosystems (though see Atkin et al. 2008). Although several DGVMs distinguish autotrophic and heterotrophic respiration, in general a few parameters are used to determine environmental response: a base respiration rate, roughly related to the amount of substrate or respiring organisms, whichever is limiting; and parameters describing dependence on temperature (exponential) or moisture (an optimum curve or linear function). DVGMs often couple the (base) rate of plant respiration to productivity (photosynthesis capacity, net primary productivity, or root activity), on the grounds that the necessary enzymes in these processes turn over and need re-synthesising quickly. This coupling of parameters in models typically operates at longer time scales (months-years), but sometimes also at daily time scales, where photosynthetic capacity also acclimates at these rates to the environment (LPJ, Sitch et al, 2003). Respiration of all types (autotrophic as well as heterotrophic) in models typically responds to temperature in an exponential way, with no optimum. The fact that respiration aws formulated this way did to a large extent contribute to the predicted Amazon die-back in the HADCM3 simulations of Cox et al, (2000). The physiological basis of such exponential relationships is fairly weak. In reality, there is good reason to assume that, in the absence of other limitations such as those imposed by moisture, respiration, like most enzymatic processes, will increase with temperature monotonously up to cellular dysfunction (eg lysis) or potentially enzymatic denaturation. The empirical basis for specifying this kind of relationship has hitherto been limited, though this is beginning to change with new empirical datasets testing temperature responses over large temperature ranges (O’Sullivan et al. 2013 PCE in press). The main uncertainties here concern the exponential coefficients and how they might alter over large temperature ranges, the maximum and/or critical point beyond which respiration declines, and base respiration values. For autotrophic respiration the base respiration mainly depends on growth, transport and maintenance requirements, and in the absence of stress responses (eg drought) will be generally correlated with productivity and productive capacity. Whether and why the temperature coefficient of autotrophic respiration varies much from the Q =2 standard value ( Atkin and Tjoelker 2003, Davidson 10 et al. 2006) is actively debated. There are also indications that moisture stress (i.e., plant water potential) has the potential to affect either base respiration or temperature coefficients ((Atkin and Macherel 2009, Metcalfe et al. 2010)). For respiration in soil, the processes are more diverse and less well-understood partly because of the complexity of soil composition and structure. First of all, while models usually split autotrophic (root- derived) respiration from heterotrophic respiration, most available data refer to bulk soil respiration, although there is an increasing amount of information on root exclusion experiments (Subke et al. 2006). While root respiration and its exponential coefficients are closely linked with productivity, heterotrophic soil respiration is related to the activity of a multitude of soil organisms, and depends on accessible and decomposable soil organic matter, soil moisture, oxygen and nutrients, as well as temperature. Traditionally, decomposition is modelled for a few ‘fractions’ of soil carbon, defined by their readiness to decompose, while this description sometimes relates poorly to observable organic matter fractions (Buurman and Roscoe, 2011. The dynamics of SOM breakdown are complicated at short and longer time scales by feed-backs such as occlusion of SOM in aggregates (eg. Zimmermann et al. 2012; Stockmann et al, 2013) and this may be affected by the activity and diversity of soil 10