1 Supporting Information, Figure S1. MuSICA model simulations of PLC in piñon pine using different vulnerability curves to predict loss of xylem conductivity with decreasing xylem water potential. Piñon pine-MuSICA simulations 70 Less vulnerable More vulnerable 60 PLC (%) 50 40 30 20 10 0 0 200 400 600 800 1000 Days since 1/1/2007 1200 1400 1600 1800 2 Supporting Information, Figure S2. Modeled versus measured predawn leaf water potential via the Sperry model. 0 2 Modeled Predawn Pressure (MPa) r 0.82piñon pine r2 0.82 juniper r2 0.97 -2 -4 -6 -8 -8 -6 -4 Measured Predawn Pressure (MPa) -2 0 3 Supporting Information, Figure S3. Example of the prediction of soil water potential using ED(X) for an ambient plot at 15cm soil depth. 2 Soil water potential (MPa) 0 -2 -4 -6 -8 ED(X) simulated Observed -10 0 100 200 Day of year 2008 300 400 4 Supporting Information, Table S1. A summary of how empirical variables were utilized or simulated by models. The model input parameters are typically static (e.g. soil texture), model driver parameters typically change over time (e.g. micrometeorological data), and model output represents simulated variables. n/a is not applicable. Variable FINNSIM Sperry TREES MuSICA ED(X) Mortality n/a n/a n/a n/a Output LAI Input Input Input Input Output Density Input n/a n/a Input Input Height Input n/a Input Input Input Cover Input n/a n/a n/a Output NSC Output n/a Output Output Output SLA Input n/a Input Input Input 1 2 Driver/output Driver In/Output Output N/A pd, md E3 Driver Output In/Output2 Output Output 1 Respiration Driver/output n/a Output In/Output n/a Vcmax n/a n/a Input Input Input Vulnerability Input Input Input Input Input Driver n/a Output Output Output soil SWC n/a n/a Output Output n/a Soil depth n/a n/a Input Input Input Soil temp. n/a n/a Input Output n/a Soil texture n/a Input Input Input Input Allometry Input Input Input Input Input Air temp. Driver n/a Driver Driver Input VPD n/a n/a Driver Driver Input PAR n/a n/a Driver Driver Input Wind speed n/a n/a Driver Driver Input Atm Press. n/a n/a Driver Driver Input Hyd Cond Output Output Output Output Input 1 pd input, md output; maintenance R input, refilling R output. 2 pd, md, and E at saturated K input; values at other times output 3 Used for evaluation of all models CLM(ED) Output Output Output Output Output Output Input n/a Output Output Input n/a Output Output Input Output Input Input Driver Driver Driver Driver Driver n/a 5 Supporting Information, Notes S1. Vulnerability to Cavitation Variation in vulnerability to cavitation. To examine the importance of vulnerability curves on PLC simulations, we compared the original MuSICA PLC simulations to those with an alternative set of curves. The degree of modeled PLC depended on the shape of the hydraulic vulnerability curves. Neither set of vulnerability curves resulted in prediction of 100% PLC, thus we conclude that our original interpretation, that PLC was elevated but did not reach 100% in the trees that died, is robust to some variation in vulnerability curves. We note that we did not have vulnerability curves for the most vulnerable tissues, fine roots. Vulnerability curves describing the loss of xylem hydraulic conductivity as a function of xylem pressure were constructed using the air-injection method (Sperry & Saliendra, 1994). The air-injection method is based on the assumption that the xylem tension required to pull air into a conduit and cause embolism is equal in magnitude to the positive air pressure required to push air into the conduit when the xylem water is at atmospheric pressure. Branch segments were collected in the field, and air emboli were removed by soaking the samples in perfusion solution under vacuum for 48 h (Domec et al., 2009). The vulnerability curves were generated by first pressurizing the air chamber to 0.05 MPa to avoid water extrusion from needle scars when axial flow was induced, and allowing the system to equilibrate for 3 min. Water flow through the branch was initiated and maximum hydraulic conductivity was measured. A pressure of 0.50 MPa was then applied and held constant for 2 min. After equilibration, the air chamber pressure was reduced to 0.05 MPa, and conductivity re-measured. This process was repeated for pressures ranging from 0.5 to 6.00 MPa, or until the conductivity of the segment was negligible. For piñon pine, new vulnerability curves determined on branches collected from the site gave comparable results to the ones from Plaut et al. (2012), with a 50 percent loss of branch xylem conductivity reached at a xylem pressure ranging from -2.9 MPa to -3.5 MPa. Curves measured on juniper trees, however, showed that the branches were less resistant to cavitationinduced embolism, with a 50% loss of conductivity reached at -7.9 MPa versus -11.3 MPa in Plaut et al. (2012). In both species the slopes of these new vulnerability curves (the slope is indicative of the rate at which embolism spreads as the xylem pressure decreases) were steeper than the slopes of the curves described in Plaut et al. (2012). In piñon pine the slope increased from 0.2 (20%) PLC MPa-1 to 0.27 (27%) PLC MPa-1, and in juniper from 0.1 (10%) PLC MPa-1 to 0.16 (16%) PLC MPa-1. Using equation (1), the less vulnerable curve parameters are: b = 3.992, c = 2.6. The more vulnerable curves are: b = 3.43, c = 1.65. Both sets of curves were generated from trees on-site but in different years (2008 for the more vulnerable curves, 2011 for the less vulnerable curves). Additional references provided in Notes S5. 6 Supporting Information, Notes S2a. Estimates of maintenance respiration and allometric calculations. We estimated annual foliar and sapwood maintenance respiration for all plots for conditions prior to the precipitation manipulation using published estimates of tree respiration rates per unit of biomass, biomass estimated from measured basal area, tree density, and published allometric equations (Grier et al., 1992). Foliar maintenance respiration was estimated as 40 nmolC mol C1 -1 s at 15°C (60% of the rate for Pinus contorta foliage biomass reported in Ryan, 1995) and sapwood maintenance respiration estimated as 0.8 nmolC mol C-1 s-1 at 15°C (rate for Pinus contorta sapwood biomass, Ryan, 1990). Total plot surveys of juniper and piñon basal area and tree density were used to estimate the diameter of the tree of average basal area by species for each plot. Biomass was estimated for the tree of average basal area using equations in Grier et al. (1992), with measured sapwood area to basal area ratios used to estimate sapwood biomass for stem wood. Branches <2.5 cm were assumed to be 100% sapwood; branches 2.5 - 7.6 cm were assumed to be 50% sapwood for juniper (and assumed to have the same sapwood ratio as the sapwood to basal area ratio for piñon), and branches > 7.6 cm were assumed to have the same sapwood ratio as the sapwood to basal area ratio for both species. Leaf area for the plot was estimated using a locally-developed, species-specific allometric equation with basal area as the predictor (NG McDowell, unpublished data), and leaf mass from leaf mass per area for piñon and juniper given in Grier et al. (1992). Annual, plot-level estimates were generated using respiration estimates for the species-specific tree of average basal area, tree density, plot area, mean annual temperature for the site reported for 1985-2011 for the nearby Mountainair, NM RAWS weather station (http://www.raws.dri.edu/), and a Q10 of 2. Additional references provided in Notes S5. 7 Supporting Information, Notes S2b. Measurement of non-structural carbohydrates (NSC) Leaf samples were collected starting in 2007, and twig samples were added to the collection protocol in 2009. Samples were collected from target trees of both species on all plots. Wholetree samples were collected concurrently off-plot to evaluate whole-tree carbohydrate distribution and in attempt to scale on-plot leaf and twig data to whole-tree carbohydrate content (LT Dickman et al., unpublished results). Nonstructural carbohydrates are defined here as free, low molecular weight sugars (glucose, fructose, and sucrose) plus starch. All samples were collected on dry ice and stored at 70°C. Samples were microwaved at 800 watts for 5 min. to stop enzymatic activity, then dried at 65°C for 48hrs. Leaf tissues were ball-milled to a fine powder (High Throughput Homogenizer, VWR). Woody tissues were milled to 40 mesh prior to ball-milling (Wiley Mini Mill, Thomas Scientific). Samples were analyzed following the protocol described by Hoch et al. (2002), with minor modifications. Approximately 12mg of fine ground plant material was extracted in a 2mL deep-well plate with 1.6 mL distilled water for 60 min. in a 100°C water bath (Isotemp 105, Fisher Scientific). After removal of a 700µL aliquot for starch analysis, the remaining extract was centrifuged (Allegra X-15R, Beckman Coulter) for 45min at 4450x g rpm, and 20µL of untreated supernatant was used for the determination of free sugars (glucose and fructose). The 20µL aliquot was incubated in a microplate shaker (BioShaker M.BR-022UP, TAITEC) for 45min. with phosphoglucose isomerase (from Baker’s yeast – Type III, Sigma-Aldrich), glucose hexokinase and glucose-6-P dehydrogenase (Glucose Assay Reagent, Sigma Aldrich), to convert fructose to glucose and glucose to gluconate-6-phosphate. The concentration of free glucose in a sample was determined photometrically in a 96-well microplate spectrophotometer (Cary 50 UVVis), relative to glucose standards of known concentration, by the increase in optical density at 340nm resulting from the reduction of NAD+ to NADH as glucose-6-P is oxidized. To hydrolyse sucrose to glucose and fructose, a 100µL aliquot of centrifuged supernatant was incubated in a microplate shaker for 40mins with 50µL invertase (Grade VII, from Baker’s yeast, Sigma-Aldrich) buffered to pH 4.6 with 0.4 M,NaOAc (Sigma Aldrich). A 20µL aliquot of invertase-treated sample was used for determination of total glucose as described above. Sucrose was calculated as low molecular weight sugars minus free sugars. To break down total NSC to glucose (Pazur & Kleppe, 1962), 700µL of extract was transferred to a new deep-well plate, prior to centrifugation, for overnight incubation in a 48°C water bath with amyloglucosidase (from Aspergillus niger, Sigma Aldrich) buffered to pH 4.5 with 0.1 M,NaOAc (Sigma Aldrich). Following incubation, the plate was centrifuged for 60min at 4450x g rpm, and a 20µL aliquot of supernatant was used for determination of total glucose as described above. Starch was calculated as total NSC minus low molecular weight sugars. All NSC values are expressed as percent dry matter. Additional references provided in Notes S5. 8 Supporting Information, Notes S3, Model-specific developments and application FINNSIM: FINNish SIMulation was used to simulate whole-tree hydraulic conductance and cavitation (based on (Hölttä et al., 2009), including feedback loops with phloem transport (based on Hölttä et al., 2006) to allow investigation of carbon-water interdependencies. We modified the model from Hölttä et al. (2009) to include whole tree level carbon balance calculations, phloem transport and the resulting distribution of carbon within a tree, and a simple estimation of the carbon cost of refilling. Whole tree-level carbon balance was calculated as a balance between photosynthesis, temperature driven respiration rate, sugar to starch conversion dynamics (Nikinmaa et al., 2012), and sub-models for embolism refilling and potassium-aided phloem transport, which was required for continued photosynthesis at low xylem water potentials in junipers. The model was used to predict tree carbohydrate concentrations (soluble sugars and starch), xylem water potential, phloem turgor gradients, and PLC. Data sources for FINNSIM included vulnerability curves (Plaut et al., 2012), temperature dependency of respiration, xylem and phloem cross-sectional areas estimated by visual inspection from dead trees on the drought plots (AJ Boutz et al., unpublished data), and the xylem and phloem conductances from relations between flow rates and measured/estimated pressure gradients (S Sevanto et al., unpublished data). Soil water potential, transpiration, and photosynthesis rates were used from MuSICA output. All other required parameters and model structures (described in Hölttä et al., 2006, 2009) were taken from allometric equations or on-site measurements (Table S1). Sperry: The model of Sperry et al. (1998) is a resistance-network model of the soilplant-atmosphere continuum. Hydraulic resistances were obtained from vulnerability-tocavitation curves for xylem (Plaut et al., 2012) and unsaturated conductivity curves for soil. The model predicted the relationship between steady-state transpiration rate (E) and xylem water potential () up to the maximum possible values (Ecrit and crit) at total hydraulic failure i.e., zero hydraulic conductance. Actual E and hydraulic conductance were predicted from measured or estimated . For this study, the seasonal maximum tree hydraulic conductance was calibrated to yield the best fit of modeled to measured pre-dawn , and the model filled in the missing predawns across all growing season days (Figure S2). The predawn time course was used with the mid-day (interpolated between measurement days) to model the daily time course of midday E, Ecrit, and loss of hydraulic conductance. Re-calibration was necessary every growing season, and occasionally for post-monsoonal periods, because of substantial recovery in plant hydraulic conductance indicated by observations (Plaut et al., 2012; R Pangle et al., unpublished data). Six trees per species were individually modeled, 3 from drought plots and 3 from ambient plots. TREES: The Terrestrial Regional Ecosystem Exchange Simulator (TREES) (Samanta et al., 2007; Loranty et al., 2010; Mackay et al., 2012; Roberts 2012) is a dynamic model of plant water and carbon flows. A unique methodological improvement in TREES is a full coupling of the Sperry et al. (1998) model of plant water balance and cavitation with stomatal conductance (GS), photosynthesis (A), and E driven by energy supply and vapor demand. Thus, TREES explicitly incorporates A and dynamic plant hydraulic conductance into a unified numerical solution. It also predicts PLC, NSC, and growth efficiency. The model was calibrated using predrought gas exchange, transpiration, water potentials, and vulnerability curve measurements (Plaut et al., 2012, Limousin et al., 2013). TREES was set up to re-adjust the plant hydraulic conductance after substantial soil recharge events to account for refilling, but the model was manually re-calibrated to measured predawn water potentials at specific events (as done for the Sperry model). The rooting zone was partitioned into a shallow layer (30 cm) comprised of two 9 root modules, and a deep layer with one root module extending a further 30 cm. Soil water balance was updated in each half-hour time step for each layer using precipitation inputs, drainage, and rhizosphere fluxes. Water potentials, hydraulic conductances, and fluxes were calculated based on the updated soil moisture, cavitation status, and transpiration demand. TREES was used to simulate the same trees described for the Sperry model. MuSICA: The MuSICA model is a multilayer, multi-leaf process-based biosphereatmosphere gas exchange model that simulates the exchanges of mass (water, CO2) and energy in the soil-vegetation-atmosphere continuum (Ogée et al., 2003). The version of the model used in this study includes a more detailed description of root water uptake and plant water storage dynamics, as well as soil water hydraulic redistribution and root cavitation (Domec et al., 2012) and plant NSC storage dynamics (Ogée et al., 2009). In this study the soil was divided into seven layers of 10cm depth each. Stand density, biomass, leaf area, and soil properties were taken from Pangle et al. (2012) and Plaut et al. (2012). Maximum rooting depth and root distribution for both species were determined by fitting modeled xylem water potential to measured predawn water potential from spring 2007 only. Both species were modeled at the same time and thus competed for the same soil water. ED(X): The Ecosystem Demography (ED) model tracks cohorts of trees based on their sizes (Moorcroft et al., 2001). ED(X) simulates tree mortality of cohorts based on the assumption of carbon starvation (Fisher et al., 2010). To better present the seasonal cycles of carbon storage, instead of using GPP directly for growth, it is first all fed into the NSC pool, which is then used by respiration and then growth of new tissue determined by carbon sink strength. The sink rate is simulated to be dependent on a targeted storage specified by the user. In the study, the target carbon storage was set to be 20% of leaf biomass based on the observational data. Another new development in this study was simulation of soil water potential and calculation of water supply to each leaf layer based on tree hydraulic conductivity and the water potential gradient from leaf to soil. Soil water potential was simulated as a nonlinear function of soil water saturation (Figure S3). The plant halts photosynthesis if the minimum leaf water potential becomes lower than simulated soil water potential and gravitational potential resulting from tree height. CLM(ED): The CLM(ED) model is a hybrid of the CLM4.0 model (Oleson et al., 2010) and the Ecosystem Demography model (Moorcroft et al., 2001) subject to the modifications described in Fisher et al. (2010) and Bonan et al. (2012), which include the development of a carbon storage model that predicts starvation as an empirical function of low carbohydrate reserves. The model is based on the concept of ‘average’ individuals, that have similar height and are in similar post-disturbance states. Thus, the community-scale mortality rate, which is a function of the internal carbon status of the average individual, must be parameterized to avoid the entire community being subject to mortality at a single mortality threshold (Fisher et al., 2010). In contrast to the individual and plot-scale approaches described above, the CLM(ED) model has prognostic predictions of leaf area index, plant size structure, density and canopy cover, as well as soil moisture and surface energy balance. 10 Detailed model descriptions: FINNSIM: A model of xylem transport and cavitation (Hölttä et al., 2009), combined with a model a phloem transport (Hölttä et al., 2006), was used in FINNSIM simulations. Temperature driven respiration rate, sugar to starch dynamics, and a sub-model for embolism refilling were added to the already existing models. Whole tree level transpiration and photosynthesis rates, soil water potential, and temperature (soil water potential was taken from the pre-dawn water potential measurements, and the other values were taken from MUSICA output) were used as environmental drivers for the model. The model was used to predict tree carbohydrate concentrations (sugar+starch), water balance, and PLC. The model structural parameters (e.g. root, xylem and phloem hydraulic conductivities, and PLC curves) were estimated based on measurement results and literature values. Embolism refilling-submodel: Embolism refilling was made to occur at a pre-determined rate (the rate was varied in the simulations) when xylem water potential was larger than a threshold value (a threshold value of -1.0 MPa was used in the simulations). The water which was taken into the refilling conduits was taken from the xylem, so refilling decreased the water potential of the xylem. The amount of sugar required for refilling was estimated from the total amount of sugar required to raise the osmotic potential high enough for water to flow into the refilling conduits from the surrounding xylem. It was further assumed, that a certain amount of these sugars could be retrieved after the refilling process, so that only a part of the sugars used to create the osmotic pull for refilling were irreversibly lost to metabolism of refilling (sugar pumps etc.). The refilling rate (and the corresponding rate of sugar consumption in refilling) in the figures refers to the fraction of conduits that could be refilled in one second provided the conditions for refilling were otherwise met (i.e. water potential above – 1 MPa and sugars available). For “fast”, “intermediate” and “slow” refilling, the fraction is 10-5 s-1, 10-6 s-1, 10-7 s-1, respectively. Sugar to starch dynamics and potassium cycling: Sugar to starch conversion was modeled so that sugar was turned into starch at a rate dependent on sugar concentration, and starch to sugar conversion was made to be dependent on starch concentration. Xylem water potential has to be balanced by a phloem osmotic potential of at least equal to xylem water potential in magnitude order for the turgor pressure to remain positive. If the high osmotic potentials were maintained with sucrose alone, the viscosity of the solution will become very high. For example, osmotic potentials of –5MPa, – 7MPa, and -7.5 MPa will increase the viscosity of a sucrose solution in a highly non-linear fashion to approximately 40-fold, 600-fold and 3000-fold in relation to pure water, respectively (see equation in Hölttä et al., , which is valid up to 7.5 MPa osmotic concentration). Therefore the circulation of potassium, which contributes to the osmotic potential without inducing a major increase in viscosity (Thompson & Zwieniecki, 2005), within the phloem and xylem along with sucrose was modeled for juniper. Potassium was only used in the juniper case, as the pine phloem osmotic concentration never reached values that phloem sap (sucrose solution) viscosity would increase severely. Potassium was loaded to the phloem at a given rate when phloem osmotic concentration increased above a given threshold value (3 MPa). It was unloaded from the phloem when osmotic concentration decreased below 3 MPa. Loading/unloading of potassium was made to cost one mole of sugar per 100 moles of potassium. 11 Sperry et al. 1998: As described in the text, this model predicts the steady-state transpiration vs. xylem pressure relationship for a given soil moisture and root depth profile. Rhizosphere drying associated with root water uptake is simulated using unsaturated soil conductivity relationships based on soil texture or moisture release curves. Cavitation is simulated based on vulnerability curve inputs. Richard's equations for mass balance at each node in the resistance network are solved to yield water flow and pressure. Use of the Kirchoff transform reduces the number of nodes required to discretize the network to just those compartments with different conductivity functions. The model was used to predict multi-year time courses of soil-canopy hydraulic conductance (K), flux per sapwood area (E), and safety factors from critical flux rates (E/Ecrit). Relative loss of hydraulic conductance (PLC = 1-K/Ksat, Ksat = maximum seasonal tree conductance) was also predicted. Six trees per species were analyzed, three from a drought plot (plot 10) and three from the ambient control plot (plot 12). Vulnerability curve inputs for piñon were the same used in Plaut et al. (2012), and separate curves were inputted for roots vs. shoots. For juniper, stem curves generated by JC Domec (described in the Supporting Notes S1) from trees on site were used in preference to the Willson et al. (2008) data used previously (Plaut et al., 2012) that were from trees at a distant site. In principle, the model can divide the rooting zone into multiple layers, and use measured soil water potentials as input. In practice, however, the data indicated that both species were drawing water from a missing soil layer that was wetter than any of the three measured layers. On many occasions, the predawn xylem pressure was less negative than the wettest measured soil layer, and there was significant sapflow. Significant flow means soil water uptake, and the predawn xylem pressure means the uptake was from a wetter (deeper) soil layer than what was measured with soil psychrometers. Rather than using the demonstrably incomplete soil moisture profile and estimates of the root distribution as inputs, the model was calibrated to predict the predawn xylem pressure (^PD, superscript ^ denotes a model prediction) from the measured sapflow (E, daily maximum) and midday xylem pressure (MD): ^PD = E/^K + MD. The ^K term is the soil-to-canopy hydraulic conductance at the MD pressure calculated by the model for an arbitrarily small soil-to-canopy ∆P of 0.2 MPa. The calibration was run for the subset of days (between April 1 and October 31) on which the PD was measured (7-9 days each year). The mean square error (MSE) between the ^PD and the measured PD was minimized by varying the initial K (Ksat) at the beginning of the time series. Days where PD < MD were excluded from the calibration. In piñon during 2007, the monsoon season corresponded with an obvious systematic deviation where ^PD became much more positive than the measured PD, indicating that ^K was under-predicted post-monsoon. This suggested recovery of hydraulic conductance in the trees. In these cases, the model was calibrated separately for pre-monsoon and post-monsoon data. The model was also calibrated separately for each year (2007 through 2010) years (April-October data only) to account for any changes in tree K occurring in the November to March period that was not modeled. Once the model was calibrated, it was used to generate the full time sequence of ^K's and ^PD's (i.e., not just on the days where PD was measured). The ^PD time sequence was re-inputted as a substitute soil water potential for the entire rooting zone (in lieu of accurate profiles for soil 12 water potential and rooting depth). From the sequence of soil water potential and interpolated MD pressures, the model calculated ^E and ^Ecrit values. The model fit was re-evaluated by comparing ^E to measured E. Fit was generally very good, but occasionally with distinct outliers. These obvious outliers (<< 0.1% of the data) were eliminated from subsequent analysis. In one tree and season (a droughted piñon, 2007 premonsoon), the PD calibration did not work well as evidenced by a poor fit between E and ^E. For this one tree and period, the model was recalibrated to fit measured E data (by varying ksat). The PLC (1-K/Ksat) was calculated relative to the mean maximum K (i.e., Ksat) over the analysis period for the surviving trees (piñon: 12-1, 12-2, 12-4, mean = 2.38 mole s-1m-2MPa-1; juniper: 10-7, 10-8, 12-6, 12-9, 12-10, mean = 0.94 mole s-1m-2MPa-1). TREES: The Terrestrial Regional Ecosystem Exchange Simulator (TREES) (Mackay et al., 2003; Samanta et al., 2007; Loranty et al., 2010; Mackay et al., 2012) that operates as a physiology model at the scale of individual trees or as an ecosystem model for whole stands. At the plant scale the model couples photosynthesis, stomatal conductance, and transpiration in a steady state solution for sun and shade canopy at 30-minute time steps, and forced with micrometeorological data (air temperature, wind speed, radiation, vapor pressure deficit, soil temperature). This coupled canopy model and the plant water balance model (Sperry et al., 1998) were combined into a single, integrated model to explicitly simulate soil-plant hydraulics and hydraulic failure, and to provide both demand and supply limits on stomatal control of carbon uptake and water loss (Roberts 2012), as well as carbon utilization and allocation. At the whole plant canopy scale stomatal conductance (GS) was calculated by combining Darcy’s Law and Fick’s law of diffusion as GS = KL()/D (S – L) (1) where KL() and L are whole-plant hydraulic conductivity and leaf water potential, respectively, D is vapor pressure deficit in the canopy, and S is soil water potential integrated over the rooting depth of the plant. The canopy and plant water balance model components are solved iteratively until they converge on a transpiration rate, with simultaneous solution of photosynthesis and stomatal conductance. For this study TREES discretized each modeled tree into three root modules, each having an absorbing and conducting element, and one canopy module having a conducting element and a lateral element with sun and shade sub-elements for gas exchange. The rhizosphere around each absorbing root element was discretized into five subelements for transporting water between the bulk soil and the absorbing root (see Sperry et al., 1998 for details). The root zone soil water balance was maintained by the model and updated, in separate layers defined by discrete root depth, using rhizosphere flux rates determined as part of the plant water balance model solution. The model moves water at the soil-root interface either from soil to root or from root to soil as a function of the pressure gradients. Once the plant hydraulic solution converges the photosynthetic assimilation is accumulated and for daily updating of NSC. Plant mortality due to hydraulic failure can be predicted using TREES because of cavitation. Plant mortality due to carbon starvation is not explicitly modeled. However, changes in NSC are 13 simulated as the difference in carbon uptake and utilization. A reduction in carbon uptake occurs when stomatal closure reduces photosynthetic assimilation of carbon. Using hydraulic conductance as a proxy for carbon transport reduces carbon utilization. Consequently, as a simulated tree approaches a condition that suggests that it would be susceptible to mortality due to stomatal closure and reduced water for carbon transport, both carbon uptake and utilization decline, which means the rate of change of NSC can be negligible. While this would not directly predict mortality due to carbon starvation a combination of plant hydraulic conductivity, hydraulic safety, cavitation, changes in NSC, carbon uptake, and carbon use collectively can be used to diagnose the health status of a simulated tree. Changes in NSC for the whole plant were calculated at daily time steps as dCNS/dt = CA - CG – CM (2) where CNS is NSC, CA is photosynthetically assimilated carbon for period t (i.e. 1 day), CG is growth and growth respiration allocated in time t, and CM is maintenance respiration over period t. Carbon is allocated first to CM and then to CG. CM was calculated using separate temperatebased respiration rates for leaf, stem, and roots as CM = (Rroot Croot rTroot + Rstem Cstem rTstem e0.67*log(10Cstem)/10 + Rleaf Cleaf rTleaf) fM~K (3) where R terms refer to root, stem, and leaf intrinsic respiration rates (fraction), C terms are carbon pools, T terms are temperatures, r is a respiration coefficient, and fM~K is a function that reduces the transport of NSC to sites for maintenance respiration as a function of hydraulic conductivity and saturated hydraulic conductivity KLsat as FM~K = KL()/KLsat (4) When root temperature is at least 5 deg. C then CG is calculated as a parameterized fraction (G) of CA as CG = GCAfG~K (5) where fG~K is function that reduces the transport of NSC to sites for growth as a function of hydraulic conductivity and saturated hydraulic conductivity KLsat as fG~K = [KL()/KLsat]2 (6) TREES was parameterized and run on individual trees (three drought piñon, three ambient piñon, three drought juniper, and three ambient juniper) using individual tree data to the extent possible. The model was tuned to each tree using species-specific allometric equations and the basal area of each respective tree, and sap flux data for each respective tree. TREES carbon pools were initialized for each individual tree using allometric equations for the root, stem, and leaf structural carbon pools and measured NSC (NG McDowell et al., unpublished data). TREES was parameterized for hydraulics by species using vulnerability to cavitation curves (Plaut et al., 2012), and by individual tree using sap flux data to obtain midday transpiration at saturated 14 hydraulic conductivity. Measured pre-dawn and mid-day water potentials at saturated hydraulic conductivity were also used. Site-specific soil texture data was used to parameterize the soil hydraulic properties. The photosynthesis routines were parameterized using species and treatment specific data collected in the study. All canopy calculations were expressed on a per unit leaf area basis, and so leaf area index by individual tree was obtained from allometry and taking the calculated total leaf area divided by projected crown area (Loranty et al., 2010; Mackay et al., 2010). We assumed that each tree operated independently of its neighbors, and so there were no interactions between root uptake rate among trees. The trajectory of carbon and water pools and fluxes for each tree was therefore independently calculated, and determined as a function of each respective tree’s carbon pools, hydraulic properties, and effect on its local soil water conditions. TREES was driven using gap-filled half-hourly micrometeorological data from the site, where gap-filling followed standard procedures (e.g., Falge et al., 2001). To simulate a reduction of water input in the drought plots we reduced precipitation input by 50 percent starting on June 1, 2007. All 12 trees were simulated starting from January 1, 2007. For the trees that died in August 2008 the simulations ran out to the end of August 2008. For all other trees the simulations were run to about June 1, 2011. MUSICA: The multilayer, multi-leaf process-based biosphere-atmosphere gas exchange model MuSICA used here has been primarily developed to simulate the exchanges of mass (water, CO2) and energy in the soil-vegetation-atmosphere continuum and is particularly well designed for studies on conifer trees because it deals with needle clumping of various needles cohorts (Ogée et al., 2003). MuSICA assumes the terrain to be relatively flat and the vegetation horizontally homogeneous. Several species can share a common soil and the mixed canopy is partitioned into several vegetation layers (typically 10-15) where several leaf types (sunlit/shaded, wet/dry) for each cohort and species are distinguished. Stand structure is therefore explicitly accounted for and competition for light and water between species can be explored. The model typically produces output at a 30-min time step and can be run over multiple years or decades as long as the vegetation structure is given. So far, it has been tested mostly on forest ecosystems. However, the model is general enough to be applied to other forest types and also crops or bare soils. The version 2.0.x of MuSICA used in this study (Domec et al., 2012) has been upgraded compared to the versions 1.x.x used in previous publications (e.g. Ogée et al., 2003; Ogée et al., 2009; Wingate et al., 2010). In this new version, all routines are now organized in independent modules according to the Fortran 90 standards. The radiative transfer scheme has been modified and is now based on the radiosity method to support multiple species in a given vegetation layer (Sinoquet et al., 2001) and can be applied to both broad-leaf and needle-leaf species. In particular, the so-called force-restore scheme used previously to describe the water and energy transfer in soils and litter has been replaced by a multilayer coupled heat and water transport scheme that explicitly accounts for root water uptake for each species. The model also now accounts for water storage in the plants with a single water storage capacity for each species that scales with leaf area (Williams et al., 2001) and for plant loss of hydraulic conductivity (cavitation function). Leaf-to-air energy, water and CO2 exchange are described in a similar fashion as in the original version and consists of a photosynthesis model (Farquhar et al., 1980), a stomatal conductance model (Leuning, 1995), a leaf boundary-layer model (Nikolov et 15 al., 1995) and a leaf energy budget equation. The only defined parameters for the leaf photosynthesis model (Farquhar et al., 1980) are the maximum rate of carboxylation (Vmax) And electron transport rate (Jmax), the night respiration rate (Rd) and the quantum yield. All These parameters are allowed to vary with leaf temperature and leaf age. Therefore, these parameters are prescribed at a given temperature (25°C) and for young and old leaves or needles. Day respiration is computed using the night respiration rate parameterization and a light inhibition factor. Soil water stress is primarily affecting stomatal aperture, which in turns reduces photosynthesis. The response curve is described by a sigmoid curve of xylem pressure (xylem) below a certain threshold where stomatal conductance is reduced by 50% (gs_50) at a constant rate (gs_shape) such as: gs= 1/(1 + (xylem/gs_50)^gs_shape) (7) This response is similar to the response of xylem hydraulic conductivity to xylem pressure (vulnerability to cavitation curve). Rain interception, leaf wetness duration and evaporation are computed for each species and vegetation layer. The MuSICA model allows the computation of scalar vertical profiles (e.g., air temperature and CO2) and the different component fluxes of the carbon, water, and energy budget. Notably, it gives separate estimates of not only tree water use, gross primary productivity, plant respiration, soil respiration (autotrophic plus heterotrophic) and net ecosystem exchange (NEE), but also soil moisture profile and root water uptake for each modeled soil layer. So far, MuSICA does not incorporate a full carbon cycle model and so respiratory terms are simply scaled using living biomass, basal respiration rates and Q10 values. Although MuSICA does not compute changes in whole plant hydraulic conductance as soil dries, for this study we generated some percent loss of root and branch hydraulic conductivity (PLC) from the modeled water potentials and from the vulnerability curves. Stem respiration rates are assumed to depend on air temperature and are scaled between crown and non-crown areas assuming all the branch biomass is inside the crown. Leaf respiration is calculated on a leaf area basis. Soil and litter respiration rates are a function of soil temperature and soil moisture. Tree water loss is controlled by energy inputs, evaporative demand, and photosynthetic need for CO2. As in TREES, water extraction is determined by the soil hydraulic conductivity, tree hydraulic conductance, water storage capacity, soil texture, and xylem vulnerability to cavitation. Water uptake can continue as long as hydraulic continuity is maintained from the soil through the xylem. Hydraulic failure can occur in the model as a result of xylem cavitation if xylem pressure falls below the structurally defined limits, or if the hydraulic conductance at the root–soil interface falls to zero due to high rates of plant water extraction or desiccation. Values of total xylem conductance were deduced from the ratio between transpiration rates and soil to leaf xylem pressure drops. The soil hydraulic conductivity depended on the volumetric soil moisture content according to the model of either Van Genutchen (1980). For each plot, the rooting zone was partitioned into seven soil layers of 10 cm each. Maximum rooting depth and root distribution were determined by fitting modeled root xylem pressure to measured predawn xylem water potentials when soil was close to saturation (early 2007). For each plot, most MuSICA parameters such as sitespecific soil physical parameters (profiles of soil porosity, soil matric potential and soil hydraulic conductivity at saturation), rooting profiles, leaf area index, hydraulic and photosynthetic 16 parameters are described in Table S1. Branch vulnerability to cavitation curves were determined using the air injection technique (Sperry & Saliendra, 1994) on five samples collected in December 2011 at the site from the control plot. Root vulnerability to cavitation curves were taken from (Plaut et al., 2012). Living tissue respiration was also parameterized using basal respiration rates and Q10 values determined at the site (Mike Ryan, described in Supporting Notes S2a). Soil and litter respiration rates were a function of soil temperature and soil moisture and were parameterized from soil respiration parameters determined at the control plot in 2006 and 2007 (White, 2008). Because of the lack of data for each needle class, MuSICA was parameterized with one set of stomatal conductance and photosynthetic parameters (maximum rates of carboxylation, rate of photosynthetic electron transport and mesophyll conductance) measured in September 2011, at the peak of summer soil moisture. We forced the MuSICA model with meteorological values (radiation, wind speed, temperature, humidity, precipitation) collected at the site and ran MuSICA for the control plot and the droughted plot where both tree species were mixed. The model was tuned to each tree using species-specific allometric equations and the basal area of each respective tree, and sap flux data for each respective tree. As described in the Sperry and TREES models’ section, we also ran MuSICA on individual trees (3 drought PIED, 3 ambient PIED, 3 drought JUMO, and 3 ambient JUMO) using individual tree data. Both species were modeled at the same time and so there were both competing for the same soil water. Simulations were run for 1550 days beginning January 1, 2007. All output variables (transpiration, GPP, NEE, Respiration) were expressed on a per ground area basis and then for model comparisons on a leaf area basis, and so as in TREES, leaf area index by individual tree was obtained from allometry and taking the calculated total leaf area divided by projected crown area. In order to estimate daily fluctuation in nonstructural carbohydrate (NSC), MuSICA was coupled with a single-substrate tree pool model (Ogée et al., 2009). We assumed that nonstructural carbon in the tree was represented by a single and well-mixed pool of water-soluble sugars, which is assumed to be always large enough to supply the metabolic demand during the growing season. This pool of sugars comprised leaf, wood and fine root sugars and was filled by leaf net photosynthesis (Fleaf) and used as substrate for maintenance and growth woody respiration (Rwood) and whole-tree biomass production (Pbiomass). In turn, Pbiomass was assumed to be carbon limited and defined as Pbiomass = kNSCNSC, where kNSC represented the pool turnover rate determined from mean whole tree growth rates (Dewar et al., 1998). The carbon budget of the pool was then written as: dNSC/dt = Fleaf – Rwood – kNSCNSC, or dNSC/dt + knscCNSC = Fleaf – Rwood (8) (9) Carbon pools were initialized at t=0 for each individual tree using allometric equations for the root, stem, and leaf structural carbon pools and measured NSC at the site (NG McDowell et al., unpublished data). 17 ED(X): We used the Ecosystem Demography (ED) model (Moorcroft et al., 2001) with modifications described by Fisher et al. (2010) and in this paper. The mortality in this modified version of ED results mainly from the assumption of carbon starvation, which links tree mortality to the carbon deficit experienced by trees. The carbon deficit is defined based on the ratio of current carbon storage concentration in leaf [LSCcur] to user-specified critical leaf storage carbon concentration [LSCLcrit], below which mortality occurs . Mortality due to carbon deficit (Mstarvation , fraction of trees dead per day) is simulated as follows (Fisher et al 2010), Mstarvation = max(0.0, Smort (1.0 - LSCcur /LSCcrit)). (10) A storage carbon pool (𝐵𝑠𝑡𝑜𝑟𝑒) is simulated based on the balance of GPP input and the output through respiration (R) and carbon sink (CS; or growth of new tissues). Specifically, 𝐵𝑠𝑡𝑜𝑟𝑒(𝑡 + 1) = 𝐵𝑠𝑡𝑜𝑟𝑒(𝑡) + GPP − R − Cs (11) where the sink rate Cs is dependant on the ratio of current leaf storage carbon concentration to a user-specified target leaf storage concentration (LSCtar). Specifically, 4 𝐿𝑆𝐶 Cs = Cs0 { 1.0 − exp[− ( 𝐿𝑆𝐶𝑐𝑢𝑟 ) ] } 𝑡𝑎𝑟 (12) where Cs0 ] is a user specified maximum carbon sink rate. We set Cs0 to be 10% of total plant leaf biomass in carbon. The current leaf storage carbon concentration is calculated from the common carbon pool, 𝐵𝑠𝑡𝑜𝑟𝑒, based on the equilibrium coefficients between leaf, root and sapwood. Based on the NSC measurement in the ambient plots, storage carbon concentration ratios of leaf to root and leaf to sapwood are set to be 1.0 and 0.33, respectively. The ED(X) model simulates the control and treatment plot with two tree cohorts: a juniper cohort and a piñon cohort. The carbon storage pools of cohorts are calculated among the carbon uptake through photosynthesis and carbon drawn by growth, respiration and tissue turnover. Photosynthesis in the modified ED(X) is simulated by the Farquhar photosynthesis model (Farquhar et al 1980) for each individual leaf layers for each tree cohort. A key parameter of Farquhar photosynthesis model is the Rubisco-limited maximum photosynthesis rate, Vc,max (umol CO2/m2/s). The carbon allocation is based on the allometry data from our study site. Specifically, the leaf biomass is calculated based on the diameter at breast height (DBH) as follows, Bl xb DHB yb (13) where xb and y are parameters fitted to data. The stem biomass (total of sapwood and dead wood) is calculated based on the allometry from original ED (Moorcroft et al., 2001) as follows, b Bs 0.136 DHB1.94 h0.572 0.931 (14) 18 where is the wood density and is set to 0.5 g/cm3 in this study and of the cohort with h is the mean height (meter) h xh DHB yh (15) where xh and yh are parameters fitted to data. The amount of live stem biomass (sapwood, Ba ) is calculated based on the pipe model as follows, Ba h SLA 10 (16) where is the ratio of leaf area to sapwood area (m2/cm2). The root biomass is empirically set to be the same as leaf biomass. Respiration is divided into maintenance respiration and growth respiration. The growth respiration consumes 25% of the carbon used for growth (Williams et al., 1987) and the leaf maintenance respiration (Rm) is set to be proportional to Vc,max (Arain et al., 2002). Specifically, Rm 0.0089 Vc, max . (17) The root and sapwood respiration is set to be 80% and 5% of leaf respiration (umol CO2/g biomass), respectively (Wertin&Teskey, 2008). Hydrology in this version of ED is based on the water supply function as determined by soil water potential, xylem conductivity and minimum leaf water potential and the water demand function as determined by stomatal conductance and vapor pressure deficit. This ED version is based on a single soil layer. The soil water potential (SWP) is simulated based on an empirical equation as follows(Oleson et al., 2010), SWP SWP0 SAT (18) where SWP0 is the reference soil water potential for saturated soil. SAT is the volumetric saturation of water in soil pores. is the exponent determined by soil texture as follows 2.91 0.159Pclay where Pclay (19) is the percent of clay in the soil and is set to be 5 in this study. The water supply is calculated based on the maximum water potential gradient between leaf and soil ( Pmax , MPa) using the cohesion theory as follows, Wsupp =- Pmax R (20) 19 where R is the resistance of water transport from root to leaf. Specifically, it is calculated as the sum of resistance of root, sapwood and leaf. See Hickler et al. (2006) for details. Pmax is calculated with respect to minimum leaf water potential (LWPmin) and height of the cohort, Pmax LWPmin SWP 9.8h . 1000 (21) The root conductivity may reduce due to xylem cavitation (Sperry et al., 1998). For this study, the loss of conductivity is calculated based on the calculated soil water potential using the Weibull equation as follows (Neufeld et al., 1992), s Rr Rr 0 e( SWP /50 ) (22) where Rr 0 is the reference root resistance with no loss of conductivity. is the critical soil water potential that cause 36% loss of conductivity and is the shape parameter for conductivity loss. The water demand of each leaf layer for a cohort is calculated based on the stomata conductance and relative humidity. Specifically, 50 s Wdem 18.0 (es - ea ) rb + rs RT (23) where rb and rs are the boundary layer and stomata resistance of water (s/m). R is the gas constant (8, 314 J/K/kmole) and T is the air temperature (K). es and ea are the vapor pressure inside leaf and of the canopy air (Pa). rs is calculated based on the empirical ball-berry model (Ball et al., 1987). Specifically, rs 1 A ea (m Patm 2000) Cf Ca es (24) where Ca is the CO2 partial pressure in the canopy air and Patm is the atmospheric pressure (Pa). 2 C f is the conversion factor from s/m to s m / umol, Cf Patm 9 10 . RT (25) During the drought, filling of tissue turn-over ceases in the model when photosynthesis become zero. The only consumption of carbon storage is from maintenance respiration. Different from the initial carbon starvation model proposed by (Fisher et al., 2010), we do not allow negative carbon storage. Instead, we down-regulate the Vc,max by 50% when the carbon storage is below 20 the critical carbon storage (S Sevanto et al., unpublished). See table below for values of the key parameters. Key parameter values used in the ED(X) model Parameter Description LWPmin Minimum leaf water potential Slope of conductivity to photosynthesis rate critical soil water potential that cause 50% loss of conductivity shape parameter for conductivity loss ratio of leaf area to sapwood area (m2/cm2) Allometric parameter 1 for leaf biomass calculation Allometric parameter 2 for leaf biomass calculation Allometric parameter 1 for height calculation Allometric parameter 2 for height calculation m 50 s xb yb xh yh Value for PIED -2.1 Value for JUMO -4.1 Sources 2 2 Data -3.57 -8.45 Data 4.07 2.2 Data 0.15 0.2 Data 0.0222 0.0386 Data 1.9172 1.686 Data 0.65 0.41 Data 0.64 0.64 Data Data CLM(ED): The CLM(ED) model is a development of the Community Land Model (Oleson et al., 2010) which is coupled to a version of the Ecosystem Demography (ED) model (Moorcroft et al., 2001) subject to the modifications described by Fisher et al. (2010). CLM(ED) as used here contains three mortality sources: 1) background mortality (fixed at 1% per year), 2) mortality due to carbon starvation, and 3) mortality due to low soil moisture potentials (as a proxy for hydraulic failure). Mortality due to carbon starvation (Mstarvation , fraction of trees dead per day) is simulated as Mstarvation = max(0.0, Smort (1.0 - bstore/bleaf,max)) (26) 21 where bstore is the stored non-structural carbohydrate, and bleaf,max is the ideal leaf biomass (both in KgC individual-1) for a tree of a given DBH (see Moorcroft et al., 2001 and Fisher et al., 2010 for allometry). For the purposes of this study, we augmented the carbon storage model of Fisher et al. (2010). In that study, a fixed fraction of live biomass was replaced each day of the simulation. Any deficit between NPP and the turnover demand was matched by removal of carbon from the storage pool. For this study, we introduce the concept that plants may respond to drought (and low NPP) either by 1) utilizing stored carbon or 2) by ceasing the replacement of lost tissues. Thus, each day, we determine the carbon balance (Cbalance KgC individual day-1), as Cbalance = NPP - dturnover (27) where NPP is the balance of photosynthesis and respiration (gC individual day^_1) and dturnover is the sum of the replacement rates of leaf, fine root and sapwood tissues (aleaf, aroot and asw) dturnover = bleaf aleaf + broot aroot + bstem asw (28) To partition any negative carbon balance into loss of tissue and loss of stored carbon, we define a new parameter, falloc. This is the minimum amount of turnover demand that is met. Aturnover,min = falloc dturnover (29) The change in the storage pool is Dstore = (NPP - Aturnover,min) fstore (30) fstore represents the demand for carbon from the storage pool. If NPP - Aturnover,min is negative, then carbon is removed from the storage pool. fstore = e-1.Tf^4 fstore = 1.0 for (NPP - Aturnover,min) >0 for (NPP - Aturnover,min) < 0 (31) (32) Thus, if there is carbon available to the store, we adjust the flux into the store for Tf, which is the fraction of the target store at present. Tf = bstore / (bleaf,maxTstorage) (33) The flux to live tissue maintenance is therefore the minimum value plus however much carbon is not allocated to the store, while the flux out of live tissues is dturnover. Dleaf = Aturnover,min + (NPP - Aturnover,min) (1-fstore) - dturnover (34) Thus, the higher the allocation parameter (falloc) the higher the allocation to live tissue maintenance, and the greater the removal of carbon from the store during stress. A high target storage (Tstorage) means a higher allocation to storage, rather than tissue maintenance and growth, but allows low NPP to be tolerated longer before mortality is induced. These two parameters are plant life-history traits related to the trade off between growth and survival, and as such, are 22 expected to differ between species. Mortality due to hydraulic failure is not simulated directly, but instead the reference stress mortality rate (Smort, fraction of individuals dying per day) is imposed for each day that the effective root-fraction soil water potential reaches a threshold value t which is variable with plant type, reflecting different tolerances of water stress (xylem function, root:shoot ratio, stomatal control) between species. Thus Mhydraulic = Smort Mhydraulic = 0.0 for plant < t for plant > t (35) (36) This simplified function allows for instantaneous mortality during intense droughts, in addition to the carbon starvation driven mortality that will typically occur as a result of chronic low assimilation rates (in the simulations). The total mortality (fraction individuals per day) is therefore Mtotal = Mbackground + Mhydraulic + Mstarvat Additional references provided in Notes S5. (37) 23 Supporting Information, Notes S4: On growth efficiency as a predictor of mortality. First generation DGVM’s often calculate mortality as a function of some metric of growth efficiency, which is often defined as NPP/LAI (McDowell et al., 2011), making this potentially one of the most practical indexes of mortality because of the ease of use by existing DGVMs. Annual growth efficiency has been shown to be an accurate predictor of vegetation mortality in some northern hemisphere conifers (Waring & Pitman, 1985; Christiansen et al., 1987) and is correlated with resin production (McDowell et al., 2007). Although growth efficiency is not as directly related to theoretical mortality mechanisms as PLC and NSC, it does represent an outcome of water transport and carbon allocation, and thus should contain some signature of the net function of plants (Waring, 1987). We compared empirical measurements of growth efficiency for trees that died and survived to the model estimates. We estimated annual wood growth and leaf area for piñnon pine using diameter growth for piñon pine estimated from annual ring widths measured on extracted cores to the nearest 0.0001 mm, and tree diameter at the core extraction point (~1.4 m height). Growth was the difference in wood biomass estimated using the allometric equations given in Grier et al. (1992) for stem wood plus live branches using diameter at ‘year’ minus the diameter at ‘year−1’ (from the measured ring widths with 100% of the diameter growth assumed to be from wood). Leaf area was estimated using the allometric equation given in Grier et al. (1992) for the diameter at ‘year−1’. Growth efficiency per tree (g wood/m2 leaf area, Waring, 1983) was calculated by annual growth ⁄ leaf area at ‘year−1’. Modeled growth efficiency was always higher than observed; however, the relative ranking of dying and surviving trees’ growth efficiency was captured correctly by the models. We note that the observed growth efficiency used measured ring widths and allometry, but assumed constant leaf area, thus the observations should not be considered accurate in absolute terms and hence we consider this analysis inconclusive but encouraging. Using growth efficiency as a general predictor of mortality is appealing because it can be calculated from most DGVMs (McDowell et al., 2011). However, in our system at least, it is clear that determining the proper threshold growth efficiency below which plants die requires more work from both empirical and modeling perspectives. Given the uncertainty in absolute values of observed or modeled growth efficiency, we propose that relative values (normalized within pixel) may be the most accurate way to employ growth efficiency in current mortality simulations. Additional references provided in Notes S5. 24 Notes S5: Additional References cited in the Supplemental Information Arain MA, Black TA, Barr AG, Jarvis PG, Massheder JM, Verseghy DL, Nesic Z. 2002. Effects of seasonal and interannual climate variability on net ecosystem productivity of boreal deciduous and conifer forests. Canadian Journal of Forest Research 32: 878-891. Ball JT, Woodrow IE, Berry JA. 1987. A model predicting stomatal conductance and its contribution to the control of photosynthesis under different environmental conditions. Progress in Photosynthesis Research 4: 221-224. Christiansen E, Waring RH, Berryman AA. 1987. Resistance of conifers to bark beetle attack: searching for general relationships. Forest Ecology Management 22: 89-106. Dewar RC, Medlyn BE, McMurtrie RE. 1998. A mechanistic analysis of light and carbon use efficiencies. Plant, Cell & Environment, 21: 573–588. Domec J-C, Warren J, Lachenbruch B, Meinzer FC. 2009. Safety factors from air seeding and cell wall implosion in young and old conifer trees. IAWA Journal 30: 100-120. Falge E, Baldocchi D, Olson R, Anthoni P, Aubinet M, Bernhofer C, Burba G, Ceulemans R, Clement R, Dolman H et al. 2001. Gap filling strategies for defensible annual sums of net ecosystem exchange. Agricultural and Forest Meteorology 107: 43-69. Farquhar GD, von Caemmerer S, Berry JA. 1980. A biochemical model of photosynthetic CO2 assimilation in leaves of C3 species. Planta 149: 78-90. Grier CC, Elliott KJ, McCullough DG. 1992. Biomass distribution and productivity of Pinus edulis-Juniperus monsperma woodlands of north-central Arizona. Forest Ecology and Management 50: 331-350. Hoch G, Popp M, Kӧrner C. 2002. Altitudinal increase of mobile carbon pools in Pinus cembra suggests sink limitation of growth at the Swiss treeline. Oikos 98: 361-374. Leuning R. 1995. A critical appraisal of a combined stomatal- photosynthesis model for C3 plants. Plant, Cell & Environment 18: 339–355. Loranty MM, Mackay DS, Ewers BE, Traver E, Kruger EL. 2010. Competition for light between individual trees lowers reference canopy stomatal conductance: results from a model. Journal of Geophysical Research - Biogeosciences 115: G04019. Mackay DS, Ewers BE, Loranty MM, Kruger EL. 2010. On the representativeness of plot size and location for scaling transpiration from trees to a stand. Journal of Geophysical Research-Biogeosciences 115: G02016. doi:10.1029/2009JG001092. 25 McDowell NG, Adams HD, Bailey JD, Kolb TE. 2007. The response of ponderosa pine growth efficiency and leaf area index to a forty-year stand density experiment. Canadian Journal of Forest Research 37: 343–355. Neufeld HS, Grantz DA, Meinzer FC, Goldstein G, Crisosto GM, Crisosto C. 1992. Genotypic variability in vulnerability of leaf xylem to cavitation in water-stressed and wellirrigated sugarcane. Plant Physiology 100: 1020-1028. Nikinmaa E, Hölttä T, Hari P, Kolari P, Mäkelä A, Sevanto S, Vesala T. 2012. Assimilate transport in phloem sets conditions for leaf gas exchange. Plant, Cell & Environment 36: 655-69. Nikolov N, Massman W, Schoettle A. 1995. Coupling biochemical and biophysical processes at the leaf level: an equilibrium photosynthesis model for leaves of C3 plants. Ecological Modelling 80: 205-235. Ogée J, Barbour MM, Wingate L, Bert D, Bosc A, Stievenard M, Lambrot C, Pierre M, Bariac T, Loustau D et al. 2009. A single-substrate model to interpret intra-annual stable isotope signals in tree-ring cellulose. Plant, Cell & Environment 32: 1071-1090. Pazur JH, Kleppe K. 1962. Hydrolysis of alpha-d-glucosides by amyloglucosidase from Aspergillus niger. Journal of Biological Chemistry 237: 1002-1006. Roberts DE. 2012. Development of a coupled ecosystem exchange plant hydraulic model to explore drought related plant mortality. Master Thesis, University at Buffalo, Buffalo, NY, USA. Ryan MG. 1990. Growth and maintenance respiration in stems of Pinus contorta and Picea engelmannii. Canadian Journal of Forest Research 20: 48-57. Ryan MG. 1995. Foliar maintenance respiration of subalpine and boreal trees and shrubs in relation to nitrogen content. Plant, Cell and Environment 18: 765-772. Samanta S, Mackay DS, Clayton M, Kruger EL, Ewers BE. 2007. Bayesian analysis for uncertainty estimation of a canopy transpiration model. Water Resources Research 43: W04424, doi:10.1029/2006WR005028. Sinoquet H, Le Roux X, Adam B, Ameglio T, Daudet F. 2001. RATP: a model for simulating the spatial distribution of radiation absorption, transpiration and photosynthesis within canopies: application to an isolated tree crown. Plant, Cell & Environment 24: 395-406. Sperry JS, Saliendra NZ. 1994. Intra- and inter-plant variation in xylem cavitation in Betula occidentalis. Plant Cell & Environment 17: 1233–1241. Thompson MV, Zwieniecki MA. 2005. The role of potassium in long distance transport in plants. In: Holbrook NM, Zwieniecki MA, eds. Vascular Transport in Plants. Boston, MA,USA: Elsevier Academic Press, 221-240. 26 Van Genuchten MT. 1980. A closed form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Science Society of America Journal 44: 892‐898. Waring RH. 1983. Estimating forest growth and efficiency in relation to canopy leaf area. Advances in Ecological Research 13:327-354. Waring RH, Pitman GB. 1985. Modifying lodgepole pine stands to change susceptibility to mountain pine beetle attack. Ecology 66: 889-897. Wertin TM, Teskey RO. 2008. Close coupling of whole-plant respiration to net photosynthesis and carbohydrates. Tree Physiology 28: 1831-1840. White S. 2008. Vegetation and environmental controls on soil respiration in a pinon-juniper woodland. Master Thesis, The University of New Mexico, Albuquerque, NM, USA. Williams K, Percival F, Merino J, Mooney HA. 1987. Estimation of tissue construction cost from heat of combustion and organic nitrogen content. Plant, Cell & Environment 10: 725-734. Willson CJ, Manos PS, Jackson RB. 2008. Hydraulic traits are influenced by phylogenetic history in the drought-resistant, invasive genus Juniperus (Cupressaceae). American Journal of Botany 95: 299-314. Wingate L, Ogée J, Burlett R, Bosc A, Devaux M, Grace J, Loustau D, Gessler A. 2010. Photosynthetic carbon isotope discrimination and its relationship to the carbon isotope signals of stem, soil and ecosystem respiration. New Phytologist 188: 576–589.