Revisiting the First ISLSCP Field Experiment to evaluate water stress in JULESv5.0

The First ISLSCP Field Experiment (FIFE), Kansas, US, 1987-1989, made important contributions to the understanding of energy and CO2 exchanges between the land-surface and the atmosphere, which heavily influenced the development of numerical land-surface modelling. Thirty years on, we demonstrate how the wealth of data collected at FIFE and its subsequent in-depth analysis in the literature continues to be a valuable resource for the current generation of land-surface models. To illustrate, we use the FIFE dataset to evaluate the representation of water stress on tallgrass prairie 5 vegetation in the Joint UK Land Environment Simulator (JULES) and highlight areas for future development. We show that, while JULES is able to simulate a decrease in net carbon assimilation and evapotranspiration during a dry spell, the shape of the diurnal cycle is not well captured. Evaluating the model parameters and results against this dataset provides a case study on the assumptions in calibrating ‘unstressed’ vegetation parameters and thresholds for water stress. In particular, the response to low water availability and high temperatures are calibrated separately. We also illustrate the effect of inherent uncertainties 10 in key observables, such as leaf area index, soil moisture and soil properties. Given these valuable lessons, simulations for this site will be a key addition to a compilation of simulations covering a wide range of vegetation types and climate regimes, which will be used to improve the way that water stress is represented within JULES. Copyright statement. The works published in this journal are distributed under the Creative Commons Attribution 4.0 License. This licence does not affect the Crown copyright work, which is re-usable under the Open Government Licence (OGL). The Creative Commons 15 Attribution 4.0 License and the OGL are interoperable and do not conflict with, reduce or limit each other. ©Crown copyright 2018 1 Geosci. Model Dev. Discuss., https://doi.org/10.5194/gmd-2018-210 Manuscript under review for journal Geosci. Model Dev. Discussion started: 28 September 2018 c © Author(s) 2018. CC BY 4.0 License.


Introduction
Models of the land surface and biosphere, a key component in climate predictions and projections, depend on high quality observational data sets to tune the behaviour of the modelled processes.A significant contribution in this field was produced by the First ISLCP 1 Field Experiment (FIFE), an interdisciplinary collaboration of researchers from remote sensing, atmospheric physics, meteorology and biology.It was based at and around the Konza Prairie Long Term Ecological Research (LTER) site, Kansas, during multiple campaigns, 1987Kansas, during multiple campaigns, -1989.Its principal objectives were twofold: to improve the understanding of the role of biological processes in controlling atmosphere-surface exchange of heat, water vapour and CO 2 , and to investigate whether satellite observations could be used to constrain land surface parameters relevant to the climate system (Sellers et al., 1988;Sellers and Hall, 1992).
As part of this experiment, canopy processes were related to leaf-level stomatal conductance, photosynthesis and respiration, and responses to water availability and atmospheric forcing were modelled in detail (Verma et al., 1989;Kim andVerma, 1990b, a, 1991a, b;Verma et al., 1992;Kim et al., 1992;Stewart and Verma, 1992;Norman et al., 1992;Niyogi and Raman, 1997;Cox et al., 1998;Colello et al., 1998).This work has subsequently played an important role in influencing the representation of vegetation in a generation of land-surface models.The parametrisation of water stress in the the Joint UK Land Environment Simulator (JULES) (Best et al., 2011;Clark et al., 2011), for example, originates in a canopy conductance and photosynthesis model presented in Cox et al. (1998), which was developed using FIFE observations.After tuning, the Cox et al. (1998) model gave a very good fit to the data: it explained 91.7% of the variance in net canopy photosynthesis and 89.4% of the variance in canopy conductance, as derived from FIFE flux tower observations.As part of this model, Cox et al. (1998) calculated a piecewise-linear stress factor β.This factor is zero below the wilting soil moisture and one above a critical soil moisture (Figure 1, solid line), based on the top 1.4m of soil.Crucially, Cox et al. (1998) found that the drop in carbon assimilation in the C4 vegetation as soil water content decreased at FIFE could only be reproduced if the stress factor β was applied directly to the net leaf assimilation rate.In their model, soil water stress affected stomatal conductance via the net leaf assimilation rate.
The Cox et al. (1998) stress parameterisation was adopted in early versions of JULES.It was the only implementation of soil moisture stress in JULES until version 4.6 and, to our knowledge, has been used in all published studies to date.The JULES wilting and critical soil moistures are input by users for each soil layer in each gridbox, and are defined as corresponding to absolute matric water potentials of 1.5 MPa and 0.033 MPa respectively (Best et al., 2011).A separate stress factor is calculated for each soil layer, and these are combined into an overall soil moisture stress factor by weighting by the root mass distribution.
Other options have been more recently implemented into JULES.These include a 'bucket' approach, in which the stress factor β is calculated from the average soil moisture to a specified depth, and the introduction of a new variable p 0 which reduces the soil moisture at which a vegetation type first starts to experience water stress (Figure 1

, dashed line).
There is currently a community-wide effort to improve the response of JULES to drought conditions.This effort requires a large amount of data to evaluate against, covering a wide variety of climate and vegetation conditions.This will ensure that any improvements have global applicability, rather than being effective only for a small subset of sites.This is important for 1 International Satellite Land Surface Climatology Project generating worldwide predictions of responses to any changing future drought conditions.The FIFE data is well documented, easily available and was fundamental to the development of the original water stress parametrisation.Hence we revisit its use in the next stage of development of water stress in JULES.We do this as follows.We first create a simulation that closely reproduces the Cox et al. (1998) study.Secondly, we update this configuration to make use of more recent model developments, with parameter values based on the generic C4 grass tile from the global analysis of Harper et al. (2016).We then use FIFE observations to tune some of these generic C4 grass parameters to more accurately represent tallgrass prairie.The model setup for each of these simulations is described in Section 2. In Section 3, we compare the results from the model simulations to net canopy carbon assimilation, derived from CO 2 flux measurements, and latent heat energy flux measurements at the FIFE site.
We conclude in Section 4 with a summary of the lessons that can be learnt for improving water stress in JULES from FIFE and how this dataset can be useful to the JULES community into the future.Throughout, we refer to the appendices, which give more information about the use of the observations and the alternative datasets considered, in order to assist future modelling work at this site.A important component of this study is the provision of a complete JULES setup that can be downloaded and used to run FIFE data through the JULES model, to allow easy inclusion of this site into a comprehensive evaluation framework for JULES.
2 Experimental set-up We will use three different configurations of JULES: -Simulation 1: repro-cox-1998 .A simplified JULES run which reproduces the original Cox et al. (1998) study as closely as possible.This requires the simple 'big leaf' canopy scheme, prescribes the Leaf Area Index (LAI) and soil moisture from observations, and calculates the soil moisture stress from the average soil moisture in the top 1.4m of soil.
-Simulation 2: global-C4-grass .This run uses parameter settings from Harper et al. (2016), which has a generic representation of C4 grass.It uses many of the 'state-of-art' features of JULES, such as the layered canopy scheme with sunflecks, and calculates soil moisture stress using a weighted sum of the stress factor in each soil layer.LAI and soil moisture are prescribed.
-Simulation 3: tune-leaf .As above, but we investigate whether the generic C4 grass leaf parameters can be tuned to site measurements, to give a more accurate representation of the prairie vegetation.
These configurations are described below and summarised in Table 1.All the FIFE datasets used in this study are given in Table A1.
This tallgrass prairie site is roughly central within the 15km × 15km FIFE study area.It had been lightly grazed by domestic livestock, but was ungrazed in 1986 and 1987 and was burned on 16th April 1987 (Kim andVerma, 1990a, 1991b).At the flowering stage in 1987, more than 80% of the vegetation was composed of C4 grasses (Kim and Verma, 1990a).
For their analysis, Cox et al. (1998) selected daylight hours that were both after 10 am local time, to exclude dew evaporation, and from days with no rainfall during that day or the preceding day.This minimised the effect of evaporation of rainfall from the canopy and soil surface and let them focus on modelling transpiration and net canopy assimilation.We will also restrict our analysis to these same time periods.The model was spun up by repeating the entire run ten times, and the output from the eleventh run was analysed.
For driving data, we use a site-averaged product of the FIFE Portable Automatic Meteorological Station (AMS) data at 30 minute resolution (Betts and Ball, 1998).We prescribe both LAI and soil moisture from observations (Stewart and Verma, 1992) rather than calculating these variables internally using the JULES phenology or soil hydrology schemes.We use a 'bucket approach' to calculate the soil moisture stress factor from the average soil moisture in the top 1.4m (this option has been available from JULES 4.6 onwards), again to mimic the Cox et al. (1998) analysis.The wilting soil moisture θ wilt was set to 0.205 m 3 m −3 and the critical soil moisture θ crit was set to 0.387 m 3 m −3 , taken directly from Cox et al. (1998).The resulting stress factor is plotted in Figure 2, and clearly shows the dry period during late July and early August.
JULES and the Cox et al. (1998) optimal configuration both use the Collatz et al. (1992) C4 photosynthesis scheme.They also both use the same stomatal conductance parametrisation: Jacobs (1994), which is in turn a simplified version of the Leuning (1995) scheme.We select the 'big leaf' option from the available canopy schemes in JULES, again to mimic Cox et al. (1998).
In this way, we are able to closely reproduce the Cox et al. (1998)   full energy balance is modelled.There are also differences in the calculation of evaporation from soil and canopy, which are not the focus of this study.The calculation of aerodynamic resistance also differs.For example, in this run, canopy height is prescribed using the data from Verma et al. (1992) for this site in 1987 (see Section A5 for more information), whereas it was not modelled explicitly as part of the Cox et al. (1998) analysis.
Many of the key FIFE datasets used in this run have large uncertainties.LAI measurements have an error of approximately 75% due to the inherent variability of prairie vegetation.LAI measurements are also affected by leaf curling or folding as the leaves pass through the detector.There are therefore significant differences between datasets (for a more detailed description, see Section A2).For example, at the beginning of August, LAI measurements vary from 2.5 (Stewart and Verma, 1992) to 0.7 (the FIFE_VEG_BIOP_135 dataset).Soil moisture was also comprehensively measured across the FIFE area by multiple groups (see Section A3).While these observations are qualitatively consistent, one of the datasets shows a bias in the lower soil levels at site 4439 in 1987 compared to the other datasets.Within-site variability in soil moisture is also large.Soil properties were similarly well studied: there are four different datasets which can be used to calculate the wilting and critical soil moistures, plus the values from two additional published studies (described in Section A4).However, measurements differ from each other by more than 0.15 m 3 m −3 in some cases.There also appear to be differences between layers, with the top 10 cm having consistently lower wilting and critical thresholds than soil at a depth of about 30 cm, for example.It is therefore vital that we consider the implications of the spread in observed LAI, soil moisture and soil properties at this well studied site when drawing our conclusions.

Simulation 2: global-C4-grass
In our second simulation, we use a recent JULES configuration, presented in Harper et al. (2016).This study introduced a trait-based approach to calculating leaf physiology in JULES, and tuned plant parameters to observations in the TRY database (Kattge et al., 2011).This resulted in improved site-scale and global simulations of plant productivity.This configuration takes advantage of many of the modern features of JULES.This includes a layered canopy scheme that treats the direct and diffuse components of the incident radiation separately (as in Sellers (1985)) and includes sunflecks (Dai et al., 2004;Mercado et al., 2007Mercado et al., , 2009)).It also calculates the overall soil moisture stress factor β from the sum of the stress factor in each layer, weighted by the root mass distribution.Since we are focussing specifically on the parameterisation of water stress, we continue to prescribe LAI and soil moisture, rather than calculate these parameters dynamically with the JULES phenology and soil hydrology schemes.
The driving data was taken from the site-averaged Betts and Ball (1998) product.The diffuse radiation fraction was calculated from shortwave radiation using the method in Weiss and Norman (1985) (see Section A1 for more information).A spherical leaf angle distribution was used, as in Harper et al. (2016).LAI was prescribed using the Stewart and Verma (1992) observations and the vegetation was set to generic C4 grass.
The Stewart and Verma (1992) soil moisture observations were partitioned into the four JULES soil layers (thicknesses 0.1m, 0.25m, 0.75m and 2.0m) using an offline version of the soil hydrology scheme in JULES, assuming the same root distribution as natural C4 grass in Harper et al. (2016).This is described in more detail in Section A3.1.The wilting and critical volumetric soil moistures and the soil albedo were set to the same values as the repro-cox-1998 run.As Figure 2 shows, the resulting soil moisture stress factor is almost identical to the simulation repro-cox-1998 .Canopy height was also prescribed using the same observations as the repro-cox-1998 configuration, and the run was initialised from the spun up repro-cox-1998 run.

Simulation 3: tune-leaf
For the third configuration, tune-leaf , we calibrate the JULES parameters to measurements of the tallgrass prairie vegetation at this particular site.At the flowering stage in 1987, the vegetation at FIFE site 4439 was dominated by three C4 grass species: 27.1% Andropogon gerardii (Big bluestem), 22.2% Sorghastrum nutans (Indiangrass) and 16.6 % Panicum virgatum (Switchgrass) (Kim and Verma, 1990a).Since individual LAI observations for each species (as used in e.g.Kim and Verma (1991b)) were not available, we continue to model this site with a single plant tile.We tune the leaf parameters of this tile to be approximately representative of the dominant species at this site, A. gerardii.

Leaf properties prior to the application of water stress in the model
As discussed above, JULES uses the Collatz et al. (1992) C4 photosynthesis scheme to calculate the unstressed net leaf photosynthetic uptake and the Jacobs (1994)  is given in Table A2, and they are defined in full in Clark et al. (2011) and Best et al. (2011).For the purposes of this calibration work, the JULES parameterisations have been reproduced with the Leaf Simulator package (Williams et al., in prep).Knapp (1985) compared leaf-level measurements of A. gerardii and P. virgatum in burned and unburned ungrazed plots on the Konza Prairie Research Natural Area in 1983, and the response of these two species to different water stress conditions.
Their plots were located at 39 • 05' N, 96 • 35' W, which is within what subsequently became the FIFE study area.The burning occurred in April 1983, to prior to initiation of growth of the warm-season grasses.They found significant differences between vegetation in the burned plot and unburned plots during the May to September period.The particular FIFE site we are modelling in our simulations, site 4439, was also burned prior to the start of the experiment (15th April 1987, Kim and Verma (1990a)), and was ungrazed throughout the FIFE period.Therefore, we use the observations from the burned plot in Knapp (1985) during May-June 1983, when water was 'not limiting', to constrain our unstressed leaf photosynthesis parameters in the tune-leaf configuration.First, we set specific leaf area and the ratio of leaf nitrogen to leaf dry mass for A. gerardii and P. virgatum to Knapp (1985) observations taken between 25th May and 10th June 1983.Once these parameters are fixed, we then fit the other parameters in the model light response curve by comparison with the light curve presented in Knapp (1985), which was compiled from observations taken May-June 1983 at 35±2 • C (Figure 3).Knapp (1985) also investigated the temperature dependence of net leaf photosynthesis by artificially altering the temperature of leaves of A. gerardii and P. virgatum.Their observations showed that the peaks in both species occurred at approximately the same temperatures, but that the peak was significantly broader in A. gerardii than P. virgatum.In JULES, the temperature dependence of net leaf assimilation for C4 plants is introduced through a temperature-dependent parameterisation of the maximum rate of carboxylation of Rubisco V cmax .This enters the calculation of both the gross rate of photosynthesis and the dark leaf respiration R d (since model R d is proportional to model V cmax ).Therefore, we can use the relation between net leaf assimilation and temperature presented in Knapp (1985) to calibrate the JULES parameters governing the temperature dependence of V cmax in the model.The result is illustrated in Figure 4, alongside the parametrisations used in the repro-cox-1998 and global-C4-grass runs.The lines calibrated to the Knapp (1985) Knapp (1985) observations.
For the tune-leaf configuration, we use JULES parameters fit to the A. gerardii data from Knapp (1985), since A. gerardii is the dominant species at this site.However, to investigate the uncertainty introduced by the variation between species, we repeat the runs using parameters fitted to the approximate midpoint of A. gerardii and P. virgatum light response curves and V cmax temperature relations.We would expect that the best parameter set to lie between these two parameterisations.However, note that Knapp (1985) does not have data for Sorghastrum nutans, the second-most dominant plant species at FIFE site 4439, so we were not able to take this species into account in this part of the calibration.
It should also be noted that Knapp (1985) reported a drop in the ratio of leaf nitrogen to leaf dry mass over the course of the 1982 season of more than 50% in the burned plots.This could be a contributing factor to the drop in leaf assimilation they observed over the course of 1983.We were not able to incorporate a time-varying ratio of leaf nitrogen to leaf dry mass into our simulations, which could lead to an overestimation of leaf assimilation in the senescence period.
There were also gas exchange measurements on individual leaves of A. gerardii, S. nutans and P. virgatum taken as part of the FIFE intensive field campaigns in 1987 (Polley et al., 1992).These observations were taken on upper canopy leaves perpendicular to the direct beam of the Sun, with varying absorbed PAR and internal CO 2 concentrations (FIFE_PHO_LEAF_46).This includes observations taken before, during and after the dry spell.Therefore, if we are to use these observations to calibrate the unstressed model parameters, we have to process them in such as way as to minimise the influence of water stress.
To achieve this, we identified individual net leaf assimilation (A l ) versus leaf internal CO 2 concentration (c i ) curves from the FIFE_PHO_LEAF_46 dataset for A. gerardii and P. virgatum.We normalised each A l -c i curve using the mean at high c i .
We then selected A l -c i curves with mean incident radiation greater than 1200 µmol PAR m −2 s −1 .This procedure minimises the dependence on water stress or individual leaf nitrogen levels, since these factors approximately cancel out in the relations used internally in JULES when they are manipulated in this way.We can then use these normalised curves to calibrate the model A l -c i response at low c i .For A. gerardii and, to a lesser extent, P. virgatum, this leads to a decrease in the initial slope of the A l -c i curve (Figure 5).
We also attempted to use the A l -c i curves identified in the FIFE_PHO_LEAF_46 dataset to calibrate the parameters in the JULES c i -c a relationship.Each individual curve was taken at a constant humidity.JULES uses the Jacobs (1994) parameterisation where Γ is the photorespiration compensation point (Γ = 0 for C4), dq is specific humidity deficit at the leaf surface.f 0 and dq crit are plant-dependent parameters: f 0 is a scaling factor on c i and dq crit governs the strength of humidity dependence of c i .This parameterisation predicts that plotting c i against c a at constant humidity would give a straight line, with gradient f 0 1 − dq dqcrit .However, plotting c i against c a implied that the slope of the c i -c a relationship changes as c a increases (Figure 6).Therefore, we are unable to calibrate the JULES c i -c a relationship to this data.Instead, we keep dq crit at the same value as the global-C4-grass configuration.Both Knapp (1985) and Polley et al. (1992) found that leaf stomatal conductance g s is proportional to the net leaf assimilation at this site.We therefore set f 0 using the gradient of the relationship between net leaf assimilation and g s fitted in (Polley et al., 1992) to their leaf observations, assuming dq = 0.02 (mean of the ambient FIFE_PHO_LEAF_46 measurements).We will investigate the effect that other values of dq crit would have on our results in Section 3.
As discussed above, JULES dark leaf respiration R d is calculated from model V cmax , scaled by a constant.For the tune-leaf simulation, we tune this constant such that the model dark leaf respiration at 30 • C matches the dark leaf respiration from Polley et al. (1992) at 30 • C (Figure 7).This is roughly double the dark leaf respiration at 30 • C in the repro-cox-1998 and global-C4-grass configurations.The Polley et al. (1992)  range.This is particularly true for the higher temperature values, where the observations in Polley et al. (1992) show an increase with temperature, whereas the tune-leaf JULES configuration shows a decrease.Polley et al. (1992) found no significant difference between A. gerardii, S. nutans and P. virgatum for a variety of leaf properties: net leaf assimilation under ambient conditions, maximum assimilation under high light and CO 2 saturation and relationship between assimilation and stomatal conductance under ambient conditions.This implies that the uncertainty we have introduced by not considering S. nutans data throughout most of this calibration is relatively minor.Polley et al. (1992) also found that there was 'no apparent relationship' between leaf temperature and net leaf carbon assimilation in their measurements of A. gerardii, S. nutans and P. virgatum, taken at ambient temperatures between 24.1 • C and 47.8 • C .They speculate that the difference between their results and the temperature relations found by Knapp (1985) is due to seasonal acclimatisation.On the one hand, this supports the change from using the rapidly varying V cmax with temperature in this regime in both the repro-cox-1998 and global-C4-grass simulations to using the relatively more stable tune-leaf parameterisation.
On the other hand, it implies that an even more stable parameterisation would be desirable.We will revisit this issue in Section 3.

Onset of water stress and relationship between water stress and leaf potential
In this section, we calibrate the parameter governing the onset of soil water stress in the model, p 0 .In the repro-cox-1998 and global-C4-grass simulations, p 0 is set to 1, meaning that the model vegetation starts to experience soil water stress at a volumetric soil moisture θ=θ crit =0.387 m 3 m −3 (Figure 1).This leads to a soil moisture stress factor β of 0.75-0.55during the first 10 days of June 1987, i.e. a reduction of 25-45% compared to the case where model vegetation is not limited by water availability (Figure 2).We can investigate this in more detail using leaf water potential observations as an indicator of the stress levels of the vegetation.Leaf potential is affected by both the soil water content and the atmospheric water content, as well as other factors affecting transpiration.Both Polley et al. (1992) and Knapp (1985) found a relationship between leaf water potential and net leaf assimilation in their measurements of grasses in the FIFE study area.Polley et al. (1992) measured leaves of A. gerardii and S. nutans throughout the 1988 growing season.These observations showed a drop in net leaf carbon assimilation as the leaf potential declined through the season: leaf water potentials -0.34 to -1.5 MPa were consistent with net leaf carbon assimilate rates of 16.2 to 41.5 µmol m 2 s −1 whereas lower leaf water potentials of -1.5 to -2.45MPa were consistent with lower rates of 3.9 to 15.5 µmol m 2 s −1 (at internal CO 2 concentrations of 200 µmol mol −1 and absorbed PAR of 1600 µmol absorbed quanta m 2 s −1 )).Knapp (1985) carried out weekly leaf water potential measurements of A. gerardii and P. virgatum in 1983 for late May to early October, which showed midday leaf potential dropping from -0.4MPa in late May to less than -6.6MPa (the pressure chamber limit) at the end of July.During this period, net leaf assimilation dropped from approximately 40µmol m 2 s −1 to less than 10µmol m 2 s −1 .Kim and Verma (1991b) proposed a model which considers the prairie vegetation to be completely unstressed until the leaf potential drops below -1 MPa.This was partially motivated by the Polley et al. (1992)  observations of FIFE site 4439 in 1987, i.e. the same site and time period we use in this study.Kim and Verma (1991a) proposed an alternative water stress model, also based on data in Polley et al. (1992), where both the maximum rate of carboxylation of Rubisco V cmax and the maximum rate of carboxylation allowed by electron transport J max had a dependence on leaf water potential.According to this parameterisation, a leaf water potential of -0.4 MPa introduces a factor of 0.97 into V cmax , for example, and a leaf water potential of -0.8 MPa introduces a factor of 0.91.
Midday leaf water potential for A. gerardii in the burned plot was approximately -0.4 MPa during their 'early season' measurement period.Therefore, according to both the Kim and Verma (1991b) and Kim and Verma (1991a) models, considering this period 'unstressed' is a very good approximation (i.e.β = 1, to within 3%), and agrees with their statement that "water was not limiting" the vegetation during this period.This validates our use of this data set to tune the 'unstressed' JULES parameters in the previous section.
We can now use the same arguments to determine how much water stress the vegetation should be experiencing at the beginning of June in our runs at FIFE site 4439 in 1987.Kim and Verma (1991a) present hourly leaf water potential measurements for A. gerardii leaves at this site, for a selection of days in 1987.On 5th June 1987, they measured a minimum leaf water potential of approximately -0.8 MPa at 2pm local time.According to the Kim and Verma (1991b) model, vegetation at this leaf water potential would not be water stressed, and according the Kim and Verma (1991a) model, V cmax would be reduced by approximately 9%.This contrasts sharply with the reduction in net assimilation throughout the day of 39%, due to water stress (i.e.β = 0.61), experienced in both the repro-cox-1998 and global-C4-grass simulations on this day.
For the tune-leaf configuration, we therefore reduce the early season water stress, to be more consistent with Kim and Verma (1991a) and Kim and Verma (1991b).This can be achieved by introducing a non-zero p 0 value in the stress factor β.This reduces the soil moisture threshold at which the plant becomes completely unstressed (β = 1) from θ crit to θ wilt + (θ crit − θ wilt )(1 − p 0 ), as illustrated in Figure 1.Assuming that the stress factor β is 0.9 on 5th June 1987 leads to p 0 =0.3.The effect of different values of p 0 will be shown in more detail in Section 3.
We now examine whether any previous modelling studies at this site support or conflict with this reduction in the soil moisture threshold at which the plant becomes completely unstressed.Crucially, the maximum soil moisture stress factor considered in the original Cox et al. (1998) study was 0.7, therefore a setup with a p 0 of 1-0.7=0.3 and parameters re-tuned to give a 30 % reduction in unstressed net leaf assimilation, would have given the same fit to the data.Similarly, a stress function with p 0 =0.3 fits the plot of the ratio of actual to potential evapotranspiration to available water in Verma et al. (1992) (when corrected for their different soil properties) at least as well as a stress function with p 0 =0.An increase in p 0 can also be considered a proxy for decreasing θ crit (which, as we have already noted, has a large uncertainty: see Section A4).A p 0 of 0.2, for example, can be used to mimic the impact of changing θ crit from 0.387, as used in this study and in Cox et al. (1998), to 0.348, as used in Verma et al. (1992).Kim and Verma (1991a) present hourly water potential measurement of A. gerardii leaves at FIFE site 4439 for 3 other days (in addition to 5th June 1987): 2nd July (peak growth period), 30th July (dry period), 20th August 1987 (early senescence).
These show a minimum of -1.2MPa, -2.6MPa and -1.7MPa respectively.Given the relationships between leaf water potential and net leaf assimilation described above, these leaf water potential measurements imply a drop in leaf assimilation during the dry period.In contrast, Polley et al. (1992) found 'no evident seasonal trend' in the maximum leaf assimilation rate or carboxylation efficiency, despite taking observations throughout the day before, during and after the dry spell in 1987 2 .The apparent lack of water stress in these measurements can be reconciled if the vapour pressure deficit (VPD) in the gas chamber for measurements taken during the hottest part of the day in the dry period is less than in the ambient air, which would raise the leaf water potential for leaves in the chamber compared to ambient leaves.The diurnal cycle in the leaf water potential measurements on 30th July in Kim and Verma (1991a) also shows the strong influence of the atmospheric VPD.As we will see, the effect of water stress on C4 photosynthesis in JULES is almost entirely driven by the drop in soil moisture in the root zone (parameterised by β) and the influence of VPD is negligible (since it enters via c i , as described in Eq. 1).This will limit the ability of JULES to capture diurnal variations in water stress on leaf carbon assimilation.

Canopy and optical properties
For the tune-leaf configuration, we keep the values of leaf reflectance and transmittance from global-C4-grass , as they are consistent with those measured by Walter-Shea et al. (1992) in 1988and 1989 as part of the FIFE experiment.Walter-Shea et al. (1992) found that leaf optical properties were not dependent on leaf water potential in the range -0.5 to -3.0 MPa.Leaf angle distribution measurements were taken as part of the FIFE campaign (SE-590_Leaf_Data), and tended towards erectophile (Privette, 1996).However, erectophile leaf angle distributions can not currently be set in JULES, so we continue to use a spherical angle distribution, as in the global-C4-grass run.Walter-Shea et al. (1992) noted that the leaf angle distribution of grass at FIFE site 4439 was affected by water availability: they concluded that severe water stress in 1988 probably contributed to a more vertical leaf orientation in 1988 than in 1989.The uniformity of the canopy in JULES can be parameterised by a canopy structure factor a (a = 1 indicates a completely uniform canopy, a < 1 indicates clumping).
It is difficult to get a numerical estimate of how uniform the canopy is at FIFE site 4439 because of the large uncertainties in LAI measurements, which we discuss in Section A2.However, using LAI from Stewart and Verma (1992), together with FIFE observations of the fraction of absorbed photosynthetically active radiation (LB_UNL_42) on a day with mostly diffuse radiation (7th August 1987), gives a rough estimate for a canopy structure factor of 0.8.The structure factor changes the effective LAI seen by the model radiation scheme, and so can be used to investigate the effects of the uncertainty in the LAI dataset.
Leaves of A. gerardii roll (fold) in response to water stress, which reduces their sunlit area while still allowing photosynthesis to continue (Knapp, 1985).This dynamic response of the leaves to drought conditions could be an important factor in modelling canopy photosynthesis during dry spells.However, this behaviour cannot be modelled in the current version of JULES.

Summary of tune-leaf configuration
The tune-leaf configuration improves the representation of the tallgrass prairie vegetation at this site by tuning to leaf and canopy measurements taken in the FIFE study area.The response of leaf photosynthesis to light, CO 2 and, particularly, of VPD on C4 photosynthesis during periods of low soil moisture, which is not captured by the model.We note that there remains significant uncertainty in the threshold for the onset of water stress, the calculation of internal CO 2 concentration and the uniformity of the canopy.There is also an uncertainty introduced by inter-species variation.We note that the comparison with observations has revealed some possible limitations of the model, such as the fixed leaf nitrogen content through the    growth stage, 2nd July and 11th July are in the peak growth stage, 23rd July, 30th July and 11th August are in the dry period and 17th August and 20th August are in the early senescence period (Verma et al., 1992).All of these dates comply with the selection criteria described in Cox et al. (1998) (following Stewart and Verma (1992)).Days with, or directly after, significant rainfall have been avoided, in order to reduce the effect of evaporation from the canopy surface and bare soil.The model latent heat flux is compared to latent heat flux measurements in the FIFE_SF30_ECV_33 dataset.GPP and net canopy assimilation are derived from CO 2 flux measurements in FIFE_SF30_ECV_33, using the method in Cox et al. (1998).Further net canopy assimilation estimates have also been read from Kim and Verma (1991a) (see Section A7 for more information).et al., 1992), normalised by the mean A l of each curve.Only curves with mean incident PAR greater than 1200 µmol PAR m −2 s −1 have been used.Coloured points: normalised A l calculated from observed ci and incident PAR for each data point in the curve and the mean T leaf observation for each curve, using the JULES relations.The JULES parameters are taken from the repro-cox-1998 configuration (red triangles), the global-C4-grass configuration (blue circles) and fits to A. g. data (tune-leaf default configuration, cyan diamonds) and P. v. data (yellow diamonds).Model points have been calculated using the Leaf Simulator package.

Results and discussion
fluxes reasonably well outside the dry period, although GPP is underestimated during the growth stages.For example, GPP is underestimated by approximately 30% during the middle of the day on 5th June.During the dry period, however, the global-C4-grass simulation poorly captures the early morning peak and subsequent decline in GPP indicated by the carbon flux observations.The repro-cox-1998 run captures this behaviour through its response to temperature.Recall that V cmax in the repro-cox-1998 simulation declines at leaf temperatures above 32 • C .This causes a decline in modelled carbon assimilation during the hottest parts of the day.However, as discussed in Section 2, this temperature response is not supported by observations in Knapp (1985) or Polley et al. (1992).Therefore, it appears that, while the model is successfully capturing the shape of diurnal cycle during the dry period, it is not achieving this with the correct physical process.
Similarly, net canopy assimilation in the repro-cox-1998 simulation compares well to the time series derived from the flux tower observations, although it has lower leaf respiration, particularly on 23rd July and 30th July Figure 9.As discussed in Section A7, the leaf respiration assumed when processing the flux measurements were based on observations of leaf respiration in Polley et al. (1992).In Section 2.3, we showed that the repro-cox-1998 simulation underestimates leaf respiration compared to the Polley et al. (1992) dataset, particularly at the higher temperatures experienced during middle of the day in the dry period.While the global-C4-grass configuration also simulates lower leaf respiration values than seen Polley et al. (1992), a combination of a low bias in the GPP and a peak in V cmax at higher temperatures (compared to the repro-cox-1998 simulation) reduces the impact on net canopy assimilation.
The latent heat flux is well modelled in general in both the repro-cox-1998 and global-C4-grass simulations outside the dry period.However both simulations overestimate the latent heat flux during the dry period (Figure 10).This is expected, given that the net assimilation is also overestimated and stomatal conductance is proportional to the net leaf assimilation at this site (Knapp (1985), Polley et al. (1992)).

tune-leaf simulations
The tune-leaf configuration generally overestimates both GPP (Figure 8) and net canopy assimilation (Figure 9) compared to the observations and the repro-cox-1998 and global-C4-grass simulations.On days during the dry period, the tune-leaf simulation behaves characteristically similarly to the global-C4-grass simulation in that it also does not capture the mid-morning peak and subsequent decline in GPP and assimilation.When fitting the tune-leaf configuration in Section 2, we highlighted uncertainties in some of the key parameters, and we will look at the effect of these here.
Firstly, the tune-leaf configuration is based on observations of the dominant grass species at this site, A. gerardii.In Section 2, we also fitted parameters to another grass species at this site: P. virgatum, and a 'combined' set fitted to both species.
Since A. gerardii is almost twice as abundant at this site in 1987 as P. virgatum, and in the absence of parameter fits to the other grass species at this site, we would estimate that the most representative parameters lie somewhere between these two  parameter sets.Using this combined A.g./P.v.parameter set increases GPP and net canopy assimilation on the order of roughly 10% compared to using the set fitted solely to A.g. (Figure 8, Figure 9), from which we conclude that the error introduced from using the dominant grass species is relatively minor.
A key difference between the tune-leaf configuration and the other configurations is the introduction of a non-zero p 0 .
Figure 11 shows illustrates that varying p 0 from 0 (as in the repro-cox-1998 and global-C4-grass simulations) to 5 0.4 has a strong effect on GPP, as expected.It demonstrates the importance of ensuring that the threshold for water stress is consistent with the 'unstressed' leaf observations we tuned against, since using p 0 =0 with these new parameters would have resulted GPP that is much too low the early growth period that we were using for tuning.Recall also that changing p 0 can be considered a proxy for changing the critical soil moisture and therefore this sensitivity to p 0 also demonstrates the sensitivity to the soil properties.

10
The effect of varying the canopy structure factor on GPP can be seen in Figure 12.This can also be seen as a proxy for examining the effect of reducing LAI as it changes the effective LAI seen by the model radiation scheme.Reducing the canopy structure factor from 0.8 to 0.3, has a large, negative impact on GPP, and a reduction of this size in LAI is inside the error given Less straightforward is the effect of the uncertainty the humidity response of c i in JULES, parameterised by dq crit .
Recall that we were unable to constrain this parameter from the FIFE observations, so we kept the same value as global-C4-grass .We then fitted the parameter f 0 to be consistent with the Polley et al. (1992) observations at this dq crit .Here we examine whether it is possible to find a value of dq crit that will model the drop in GPP from mid-morning during the dry period.Decreasing dq crit such that the specific humidity deficit dq approaches or exceeds it during the middle of the day in the drought period would cause a drop in modelled c i (see Eq. 1).This could push the net leaf assimilation into the c i limited region, especially as tuning to the normalised A l -c i curve in Section 2 increased the threshold at which the plant can be considered to be not limited by CO 2 .We can test this by using a more extreme value of dq crit of 0.0475 (and a corresponding value of f 0 of 0.95, to stay consistent with the Polley et al. (1992) relation).Figure 13 shows 4 days during the dry period where the GPP observations show a mid-morning peak followed by a decline.Using this extreme value of dq crit produces a sharp dip in the modelled GPP on the 30th July, which is markedly different to the flat nature of the GPP observations after the mid-morning peak.The model produces a similar localised mid-afternoon dip on the 11th August, with its minimum close to the observations.However, neither the 23rd July nor the 10th August shows any visible difference between this extreme dq crit and the standard tune-leaf value.This clearly shows that the model does not have the flexibility to reproduce the observed diurnal cycle in GPP on low humidity days during the dry period, even if it is pushed into the more extreme region of parameter space.

Limitations of the current water stress representation in JULES and possible extensions
As we have seen, both the global-C4-grass and tune-leaf simulations are unable to capture the diurnal cycle of GPP, net canopy assimilation and latent heat flux during the dry period at FIFE site 4439 in 1987.The repro-cox-1998 simulation is more successful, but this response is mediated by a temperature dependence in leaf carbon assimilation which is not supported by observations.Other studies have argued that the dry period diurnal cycle at this site can by captured via an explicit dependence on leaf water potential.As discussed in Section 2.3, there is an observed relationship between leaf water potential and leaf assimilation in grass species at this site, and leaf potential is lowered not just by low values of soil moisture, but also by the high values of atmospheric VPD that occur during the middle of the day in the dry period.Kim and Verma (1991a) were able to qualitatively capture the mid-morning peak and subsequent decline in net canopy photosynthesis on 30th July at this site, using a model in which both V cmax and J max had a dependence on their leaf water potential measurements.Furthermore, Kim and Verma (1991b) were able to reproduce similar behaviour in canopy conductance at this site on 30th July and 11th August 1987 using a model that included an explicit dependence on observed leaf water potential, as well as a direct dependence on VPD.This implies that one possible way to improve the performance of C4 vegetation in JULES during dry periods would be to include a parameterisation of leaf water potential within the model, used as part of the calculation of carbon fluxes during water stress conditions.Although it improved the fit to observations, the leaf potential-based model in Kim and Verma (1991a) still overestimated net canopy carbon assimilation on days during the dry period.They speculated that this could be due a decrease in apparent LAI caused by leaf rolling.As discussed in Section 2.3, leaf rolling is an observed strategy of A. gerardii, in response to drought conditions (Knapp, 1985), which cannot be captured by the current version of the model.It would be useful to investigate whether a parameterisation of leaf rolling could be added in the future.It would also be valuable to include the decline in leaf nitrogen observed by Knapp (1985).
FIFE provides an ideal case study for improving the model representation of water stress on carbon and water fluxes on a tallgrass prairie site.The extensive range of observations available means that the FIFE dataset would also be very useful for looking at other processes.These include plant and soil respiration (see the discussion in Section A7) and the modelled energy balance (see, for example, Kim and Verma (1990a) and Colello et al. (1998)).Betts and Ball (1998).Site averaged product Betts and Ball (1998).Site averaged product Betts and Ball (1998).
No diffuse radiation needed.

Diffuse radiation from shortwave radiation
Diffuse radiation from shortwave radiation using method in Weiss and Norman (1985).
using method in Weiss and Norman (1985).
Other met.data Site averaged product Betts and Ball (1998), Site averaged product Betts and Ball (1998).Site averaged product Betts and Ball (1998).apart from air pressure, which is set to a constant.
Leaf Area Index (see Table A2) described in Section 2.

Conclusions
In their closing remarks, Sellers and Hall (1992) state that "FIFE created an environment for the discussion of all aspects of the land surface component of Earth remote sensing and Earth system modeling and provided a data set which has been and continues to be used to test models and algorithms."This paper demonstrates that this is still the case, over thirty years since the collection of its first year of data.FIFE continues to be a valuable resource for the land-surface modelling community, due to the wealth of available data and the extensive analysis in the literature.Furthermore, the response of vegetation carbon and water fluxes to dry spells was an area of research that was particularly prominent in the FIFE literature, and FIFE observations were used to derive the original soil moisture stress parametrisation that was incorporated into JULES.This therefore makes FIFE an ideal test case for evaluating and improving this process.Cox et al. (1998).Extending this setup to make use of all the subsequent 10 developments that have been incorporated into JULES demonstrates that the current version of the model is still able to successfully reproduce the net canopy assimilation and latent heat energy flux reasonably well through the season.However, it highlights some important issues.JULES is not currently able to capture the diurnal cycle of net canopy photosynthesis at this C4 grass site, due to the lack of a strong dependence on the canopy vapour pressure deficit (indirectly or directly).The temperature response of V cmax can be tuned to compensate for this, but it is more desirable for the model to respond to high temperature stress and high water stress individually.These runs also showed how the default water stress parameterisation can result in large reductions in photosynthesis during periods that are not considered water-limiting at the site.Re-tuning the water stress parameterisation to address this issue must be done in conjunction with a re-tuning of the unstressed photosynthesis parameters.

JULES can closely reproduce the original runs in
The water stress parametrisation, and therefore photosynthesis and transpiration in JULES, are also sensitive to the observational uncertainty in leaf area index, soil moisture, and soil properties.These have been extensively studied at FIFE in independent investigations and yet still show a wide spread, leading to large modelling uncertainties.This is an important issue when tuning JULES based on site data and must be carefully considered when scaling up to larger scale runs.The FIFE data also indicates that it could be beneficial to extend JULES to include other drought strategies observed at the site, such as leaf rolling and senescence.
Confidence that the model is capturing key processes is necessary if the model is being run into new regimes, such as when forced with climate projections.This study provides clear examples of how improving one part of the model may initially appear to worsen the fit to observations, if there were compensating biases.It requires detailed site data to disentangle these effects.This detailed data needs to be available for a wide variety of sites, with different climates and vegetation, to avoid the risk of over-tuning to one site.It is hoped that this study can be part of the larger effort of developing and evaluating JULES.
With this aim, the publication of this manuscript will be accompanied by the release of full set of files needed to process the data downloaded from ORNL-DAAC and reproduce these JULES runs (see the 'code and data availability' section for more information).It is intended that this suite of files form a living set of configurations, which will continue to develop in the future as additional parts of the model are evaluated against the FIFE dataset, and the JULES community builds up a comprehensive body of knowledge of data and model runs at this site.
Code and data availability.JULES can be downloaded from the JULES FCM repository on the Met Office Science Repository Service at https://code.metoffice.gov.uk/trac/jules(registration required).We use JULES version 5.0 (tag 'vn5.0'), which corresponds to revision 9522.The Leaf Simulator can be downloaded from https://code.metoffice.gov.uk/trac/utils.Where data points have been read directly from published plots, this was done with the EasyNData tool (Uwer, 2007).The three JULES simulations described in this study can be reproduced using the rose suite u-bb181, available at https://code.metoffice.gov.uk/trac/roses-u/browser/b/b/1/8/1/trunk.This suite also contains instructions for downloading the driving data from ORNL-DAAC and a script to pre-process the driving data, including calculating the diffuse radiation fraction.grass LAI, non-grass LAI and total LAI are available as part of the FIFE_VEG_BIOP_135 dataset.However, the total LAI in FIFE_VEG_BIOP_135 is substantially different from the measurements in Stewart and Verma (1992), Verma et al. (1992) and Sellers et al. (1992).This was investigated in detail at the time (Kim et al., 1989).The FIFE_VEG_BIOP_135 dataset documentation estimates that there is standard error of the mean LAI in their data of around 75% due to the inherent variability of prairie vegetation and a variation of about 25% can be attributed to leaf curling or folding as the leaves passed over the detector, particularly an issue for drought-stressed leaves.Foliage Area Index measurements (i.e.includes green leaves, dead leaves, stems) are available in FIFE_LB_UNL_42 for site 4439 in 1987, and plotted in Figure A2.FIFE_LIGHTWND_43 and FIFE_LB_KSU_41 also have Foliage Area Index measurements for site 4439, but these were taken in 1988-9, not 1987.
We also experimented with the internal phenology scheme in JULES.Calculating LAI dynamically with the phenology scheme would remove the need to prescribe LAI.However, we found that this scheme did not have the flexibility to reproduce the observed seasonal cycle of LAI.(1992) for the intensive field campaigns and green LAI data from Table 4 in Verma et al. (1992).

A3 Soil moisture
The soil moisture data for site 4439 presented in Figure 1 of Stewart and Verma (1992) were created from a combination gravimetric measurements and neutron probe measurements.The gravimetric measurements were taken in the top 0.1m soil daily during the FIFE intensive field campaigns and weekly between campaigns.The neutron probe measurements were taken at different depths on 15 dates, at approximately weekly intervals between the end of May and the beginning of September 2017.
These measurements were interpolated in Stewart and Verma (1992)  to get a daily soil moisture values for the 0-1.1m soil layer.Stewart and Verma (1992) also observed 'virtually no seasonal variation' in soil moisture below 1.1m.The data from Stewart and Verma (1992) for the top 1.1m of soil corresponds very closely to the 0-1.6m soil moisture values used in Cox et al. (1998) on their selected days, as illustrated in Figure A3.Stewart and Verma (1992) also presents data for an ungrazed site in the FIFE area, and state that, while the ungrazed and grazed sites received very similar season totals of precipitation, individual storms resulted in differences in soil moisture (which gives a possible motivation for using site 4439 precipitation measurements over the site-averaged data product we use here).
ORNL-DAAC contains two main datasets of soil moisture observations on levels that can be considered for site 4439 for 1987: FIFE_SM_NEUT_111, which contains measurements carried out at site 4439 and FIFE_FFONEU87_100, which is a site-averaged product for the FIFE area (Betts and Ball, 1998).These are plotted in Figure A4 for 1987.It can be seen that, at lower depths, the site 4439 measurements are considerably lower than the site-averaged product.For 1988, however, the site-averaged product is mostly within or near the edge of the spread of observations at site 4439, up to approximately 120cm.
Neither of these datasets are consistent with the Stewart and Verma (1992) site 4439 dataset when summed over the top 1.1m.
The FIFE_SM_NEUT_111 for site 8639, on the other hand, is consistent with the Stewart and Verma (1992) site 8739 dataset.
The documentation for FIFE_FFONEU87_100 also cautions that the 20cm neutron probe data is 'suspect' as the range of the probe exceeds 20cm in dry soil and says that it is 'inconsistent' with the rest of the profile in 1987.It has been linearly interpolated between observation dates.Plots of observed soil profiles for 9th July and 31st July 1987 are presented in Kim and Verma (1990a).Soil profiles for individual days are also presented in Colello et al. (1998), which are consistent with the neutron probe measurements in FIFE_SM_NEUT_111, but not the gravimetric measurements.Given these inconsistencies, we chose not to use the soil moisture observations for individual levels to directly drive our simulations.The resulting soil hydraulic and thermal parameters from these different methods are plotted in Figure A6, and shows that there are considerable differences between the different datasets.The large spread in the wilting and critical soil moistures is particularly important to note, since, as we have discussed, they both enter the soil moisture stress factor β explicitly, and therefore plant GPP and transpiration are very sensitive to variations in these parameters.The thermal and optical soil properties and the remaining hydraulic properties have a comparatively minor effect on GPP and evapotranspiration.

A5 Canopy height
In this study, we used the canopy height observations presented in Table 2 of Verma et al. (1992): 0.4-0.6m,0.6-0.75m,0.75-0.9mfor days 120-179, 180-239, 240-300 respectively for site 16 in 1987.Another available dataset for canopy height at this site is FIFE_VEG_BIOP_135, which is plotted in Figure A7, and shows considerable differences with the Verma et al. (1992) data, particularly in the 240-300 day period.As discussed in Section A2, the non-uniformity of the vegetation at this site is a significant source of error in these measurements.
A6 Canopy dark respiration Polley et al. (1992) shows leaf dark respiration as a function of leaf temperature for observations of A. gerardii, S. nutans and P. virgatum taken in the FIFE area in 1987 and fits the following relationship: When this relation was used in Cox et al. (1998), it was scaled up to the canopy level by multiplying by LAI, i.e. dark respiration was assumed to be constant on leaves through the canopy.In contrast, in the model presented in Kim and Verma (1991a), leaf respiration was calculated from where R d,25 =1.55 µmol m −2 s −1 , R=8.314 J K −1 mol −1 is the gas constant and T l is the leaf temperature in • C and leaf dark respiration was suppressed by 50% when the absorbed PAR was greater than 20 µmol quanta m −2 s −1 , to account for the light dependency of mitochondrial respiration.Air temperature near the top of the canopy was used to approximate leaf temperature.Kim and Verma (1991a) scaled this leaf respiration up to the canopy level by considering the sunlit and shaded portions of the leaf separately.
In JULES, dark respiration decreases through the canopy in the same way as V cmax and it is multiplied by the soil moisture stress parameter β.In the 'big leaf' approximation used in the repro-cox-1998 run, V cmax decreases through the canopy with light.In the layered canopy model with sunflecks used in the global-C4-grass and tune-leaf runs, the decrease of V cmax through the canopy is set by an input parameter k nl , and the leaf dark respiration is reduced by a factor of 30% above a light threshold.Table A2.JULES parameters used to represent vegetation at FIFE site 4439, which vary across runs.These parameters are all specified in the JULES_PFTPARM namelist apart from can_rad_mod (JULES_VEGETATION), co2_mmr (JULES_CO2) and beta2 (JULES_SURFACE).

Figure 2 .
Figure 2. Daily mean soil moisture stress factor β for each JULES simulation at FIFE site 4439 in 1987.
temperature has been improved.Leaf water potential observations indicate the need to delay the onset of water stress in our run, 2 Tim Arkebauer, personal communication, and timestamps from the FIFE_PHO_LEAF_46 dataset 11 Geosci.Model Dev.Discuss., https://doi.org/10.5194/gmd-2018-210Manuscript under review for journal Geosci.Model Dev. Discussion started: 28 September 2018 c Author(s) 2018.CC BY 4.0 License.compared to the repro-cox-1998 and global-C4-grass configurations.These observations also indicate an influence

Figure 3 .
Figure 3. Mean observations from Figure 1 in Knapp (1985) from the burned plot, early season (May-June 1983) for A. gerardii (cyan diagonal crosses) and P. virgatum.(yellow vertical crosses) for net CO2 assimilation rate against incident PAR, at 35±2 • C .JULES parameters are fitted to the A. gerardii observations (cyan dashed line), P. virgatum.(yellow dashed line) and a combination of both (green solid line).Also shown are the relations from the repro-cox-1998 (red dotted line) and global-C4-grass runs (blue dot-dashed line), at 35 • C .Fitted lines assume no water stress (i.e.β = 1) and ci=200 µ mol CO2 (mol air) −1 .Model lines have been created using the Leaf Simulator package, which reproduces the internal JULES calculations.

Figure 8 ,
Figure8, Figure9and Figure10show the model output for gross primary productivity (GPP), net canopy assimilation and latent heat flux for eight days during 1987.These dates sample a range of different vegetation states: 5th June is in the early

Figure 4 .
Figure 4. Vcmax against leaf temperature for A. gerardii (cyan diagonal crosses) and P. virgatum.(yellow vertical crosses), using the normalised observations from Figure 2 in Knapp (1985), scaled using the fitted light response curves of A. gerardii and P. virgatum at 35 • C shown in Figure 3. JULES parameters are fitted to these derived A. gerardii observations (cyan dashed line) and P. virgatum.observations (yellow dashed line) and a combination of both (green solid line).Also shown are the relations from the repro-cox-1998 (red dotted line) and global-C4-grass runs (blue dot-dashed line).Model lines have been created using the Leaf Simulator package.

Figure 6 .
Figure 6.Leaf internal CO2 against chamber CO2 for A.g. (cyan) and P.v.(yellow) from FIFE_PHO_LEAF_46.Each line is from the same A l -ci curve, and was taken an approximately constant humidity and temperature.

Figure 7 .
Figure 7.Comparison of leaf dark respiration against leaf temperature relations from Polley et al. (1992) (black solid line) Kim and Verma (1991a) (black dashed line), repro-cox-1998 (red dotted), global-C4-grass (blue dot-dashed), tuned to A.g. (cyan dashed), tuned to P.v.(yellow dashed) and tuned to both A.g. and P.v.(green solid line).All lines assume no light inhibition of respiration.All JULES lines are top of the canopy (TOC) values without water stress.The lines that reproduce JULES configurations have been calculated using the Leaf Simulator package.
Geosci.Model Dev.Discuss., https://doi.org/10.5194/gmd-2018-210Manuscript under review for journal Geosci.Model Dev. Discussion started: 28 September 2018 c Author(s) 2018.CC BY 4.0 License. in the LAI dataset documentation (as discussed in Section 2).Varying the canopy structure factor in the range 0.8-1.0 has a negligible effect on GPP on these days.

Table 1 .Figure 8 .Figure 9 .
Figure 8.The diurnal cycle of GPP at site 4439 in the FIFE area for 8 days in 1987: 5th June (early growth), 2nd July and 11th July (peak growth), 23rd July, 30th July and 11th August (dry period) and 17th August and 20th August (early senescence).Green band show uncertainty from fitting plant parameters to A. gerardii compared to fitting to both A. gerardii and P. virgatum.20

Figure 10 .Figure 11 .Figure 12 .
Figure 10.The diurnal cycle of latent heat flux at site 4439 in the FIFE area for 8 days in 1987: 5th June (early growth), 2nd July and 11thJuly (peak growth), 23rd July, 30th July and 11th August (dry period) and 17th August and 20th August (early senescence).Green band show uncertainty from fitting plant parameters to A. gerardii compared to fitting to both A. gerardii and P. virgatum.22

Figure 13 .
Figure 13.The diurnal cycle of GPP at site 4439 in the FIFE area for 4 days in during the dry period of 1987.Green band shows how tune-leaf simulation would vary if dqcrit, f0 were changed from the default values of dqcrit=0.075,f0=0.675 to the more extreme values dqcrit=0.0475,f0=0.95.

Figure A1 .
Figure A1.Leaf area index observations for site 4439 for 1987.Left: data from FIFE_VEG_BIOP_135.Right: literature values.Plot includes data extracted from Stewart and Verma (1992) Figure 1 and Cox et al. (1998) Figure 1, total LAI and green LAI from Sellers et al.

Figure A2 .
Figure A2.Foliage Area Index observations from FIFE_LB_UNL_42 for site 4439 in 1987.
Geosci.Model Dev.Discuss., https://doi.org/10.5194/gmd-2018-210Manuscript under review for journal Geosci.Model Dev. Discussion started: 28 September 2018 c Author(s) 2018.CC BY 4.0 License., published in Polley et al. (1992) FIFE_VEG_BIOP_135 Nelson et al. (1994) Biophysical properties of the vegetation at the FIFE study area collected for FIFE by the staff of the Evapotranspiration Laboratory at Kansas State University under the direction of E.T. Kanemasu, and by the staff of the University of Nebraska, Lincoln under the direction of B. Blad.The dedicated efforts of A. Nelson, J. Killeen, L. Ballou, T. Shah, and C. Hays in collecting and preparing these data is particularly appreciated.FIFE_LEAF_H2O_126 Blad and Walter-Shea (1994b) Total Leaf Tissue Water Potential data collected by B.L. Blad, E.A. Walter-Shea, C.J. Hays, and M.A. Mesarch of the University of Nebraska.FIFE_LB_UNL_42 Blad and Walter-Shea (1994a) Leaf Area Index and PAR Determined from the UNL Light Bar Data collected under the direction of B.L. Blad and E.A. Walter-Shea at the University of Nebraska.The dedicated efforts of C.J. Hays and M.A. Mesarch in the collection and preparation of these data is particularly appreciated.

FIFE_LIGHTWND_43FIFE_FFONEU87_100
Index Obtained from the KSU Light Wand data collected by staff of Kansas State University under the direction of T. Shaw and E.T. Kanemasu.The contribution of these data is appreciated.FIFE_LB_KSU_41 Shah and Kanemasu (1994a) Leaf Area Index and PAR Determined from KSU Light Bar Measurements collected as part of the KSU staff science effort directed by Dr. E.T. Kanemasu.FIFE_SM_NEUT_111 Kanemasu (1994a) Neutron Probe Soil Moisture data collected for FIFE by the staff and students of the Evapotranspiration Laboratory at Kansas State University under the direction of Dr. Edward Kanemasu.The dedicated effort of A. Nelson, T. Shah and G. Harbers in the collection and preparation of these data is particularly appreciated.data, published in Norman et al. (1992).

FIFE_PHO_BOX_27FIFE_SOILSURV_115
Asrar and Sellers (1994) Canopy Photosynthesis data collected by G. Asrar of NASA HQ and P.J. Sellers of NASA Goddard Space Flight Center.survey data obtained by the FIFE Information System staff from the US Dept. of Agriculture, Soil Conservation Service (USDA-SCS).Thanks are due to Dr. Elissa Levine who was instrumental in acquiring, interpreting, and preparing these data.

FIFE_SOILDERV_117
data set produced by the FIFE Information System staff using data collected by the FIFE staff science team.FIFE_SOIL_REL_112 Hope and Peck (1994) FIFE Level-3 Example Gridded Soil Moisture data provided by Drs. A. Hope and E. Peck.The assistance of Dr. James Wang, NASA, in furnishing PBMR soil moisture data was sincerely appreciated.Thanks to the FIS staff, especially Fred Huemmrich, Diana van Elburg-Obler and Jeff Newcomer, for providing information in such usable form.Thanks also to Eric Wood, Princeton University, for providing soil moisture data in digital form for the Kings Creek Basin.FIFE_SOILHYDC_107 Kanemasu (1994b) Soil Hydraulic Conductivity data collected for FIFE by the staff of the Evapotranspiration Lab at Kansas State University.FIFE_SOILREFL_114 Huemmrich (1994) Soil Reflectance Reference data obtained by the FIFE Information System staff from Stoner et al., 1980.The permission of Stoner et.al. to use these data is greatly appreciated.
relation to calculate stomatal conductance.In this section, we calibrate these parameterisations to the available in situ observations.A brief description of each of the model parameters fitted in this section 6 Geosci.Model Dev.Discuss., https://doi.org/10.5194/gmd-2018-210Manuscript under review for journal Geosci.Model Dev. Discussion started: 28 September 2018 c Author(s) 2018.CC BY 4.0 License.

Table A1 .
List of FIFE datasets from ORNL-DAAC referenced in this document.Each dataset is referred to by its folder name.Geosci.Model Dev.Discuss., https://doi.org/10.5194/gmd-2018-210Manuscript under review for journal Geosci.Model Dev. Discussion started: 28 September 2018 c Author(s) 2018.CC BY 4.0 License.