How carbon (C) is allocated to different plant tissues (leaves, stem, and
roots) determines how long C remains in plant biomass and thus remains a
central challenge for understanding the global C cycle. We used a diverse set
of observations (AmeriFlux eddy covariance tower observations, biomass
estimates from tree-ring data, and leaf area index (LAI) measurements) to
compare C fluxes, pools, and LAI data with those predicted by a land surface
model (LSM), the Community Land Model (CLM4.5). We ran CLM4.5 for nine
temperate (including evergreen and deciduous) forests in North America
between 1980 and 2013 using four different C allocation schemes:
dynamic C allocation scheme (named “D-CLM4.5”) with one dynamic
allometric parameter, which allocates C to the stem and leaves to vary in
time as a function of annual net primary production (NPP); an alternative
dynamic C allocation scheme (named “D-Litton”), where, similar to (i), C
allocation is a dynamic function of annual NPP, but unlike (i) includes two
dynamic allometric parameters involving allocation to leaves, stem, and coarse
roots; a fixed C allocation scheme with two variants, one representative of
observations in evergreen (named “F-Evergreen”) and the other of
observations in deciduous forests (named “F-Deciduous”).
D-CLM4.5 generally overestimated gross primary
production (GPP) and ecosystem respiration, and underestimated net ecosystem
exchange (NEE). In D-CLM4.5, initial aboveground biomass in 1980 was largely
overestimated (between 10 527 and 12 897 g C m
Our results highlight the utility of using measurements of aboveground biomass to evaluate and constrain the C allocation scheme in LSMs, and suggest that stem turnover is overestimated by CLM4.5 for these AmeriFlux sites. Understanding the controls of turnover will be critical to improving long-term C processes in LSMs.
Over the last half century, on average a little more than a quarter of global
CO
The amount of carbon stored in biomass is dependent on how photosynthetically fixed carbon is allocated between C pools and how long these pools persist (Bloom et al., 2016; Koven et al., 2015; De Kauwe et al., 2014). How long-lived different plant pools are (leaf, stem, and root) influences whether ecosystems are projected to act as C sources or sinks (Delbart et al., 2010; Friend et al., 2014). Once C is taken up by the plant, the carbon is allocated either to short-lived leaf or fine-root tissues, or to longer-lived woody tissues. Here we use turnover time of C in a plant pool as the total carbon pool divided by the total flux into or out of the pool (Sierra et al., 2017). Plants that allocate a greater proportion of C to tissues with long turnover times (e.g., stem) have a higher standing biomass than the plants that allocate a greater proportion of C to tissues with short turnover times (e.g., leaf). Ecological theory suggests that variation in C allocation to different plant pools is governed by functional trade-offs (Tilman, 1988), with plants investing in either aboveground or belowground tissues depending on which strategy would maximize growth and reproduction. If the functional trade-off hypothesis is relevant on forest or regional scales, land surface models (LSMs) for forests should represent it using dynamic C allocation schemes, which are responsive to above- (e.g., light) and belowground (e.g., water or nutrients) factors that limit growth.
Allocation of C between pools in terrestrial ecosystems is poorly represented in LSMs (Delbart et al., 2010; Malhi et al., 2011; Negron-Juarez et al., 2015). Some LSMs use fixed ratios for each plant functional type (PFT), while other models use allocation fractions that are altered by environmental conditions (Wolf et al., 2011; De Kauwe et al 2014). Though many LSMs use the same fractional allocation for both evergreen and deciduous forests, global syntheses show differences in inferred C allocation patterns, for example, the percentage of NPP allocated to leaves that is greater in deciduous than in evergreen forests (Luyssaert et al., 2007). In part this is because it is difficult to measure allocation to different pools on ecosystem or landscape scales, and instead we infer what partitioning was required to result in different biomass pools. While eddy covariance observations can be used to parameterize and benchmark LSMs either at single sites or, using geospatial scaling methods, across regions or the globe (Baldocchi et al., 2001; Friend et al., 2007; Randerson et al., 2009; Zaehle and Friend, 2010; Mahecha et al., 2010; Bonan et al., 2011), these data inform fluxes in and out but do not provide information on allocation between pools (Richardson et al., 2010).
Studies focusing simultaneously on C pools, fluxes and allocation are relatively rare (Wolf et al., 2011; Xia et al., 2015; Bloom et al., 2016; Thum et al., 2017), in part because collecting biometric data in addition to flux data is very labor-intensive. Some forest inventory data include estimates of the average biomass within the leaf, wood and root pool, and these can be used to parameterize and benchmark models (Caspersen et al., 2000; Brown, 2002; Houghton, 2005; Keith et al., 2009). The AmeriFlux network provides a rare opportunity to investigate forest allocation processes because gross primary productivity and respiration fluxes are quantified continuously. However, measurements of pool sizes in leaves, stems, etc., are less available at these sites and so have been less frequently explored.
In this study, we evaluate mechanisms by which C is stored over multiple decades in plant biomass using corresponding eddy covariance flux towers and biometric measurements of C storage in different pools. We collated biometric data (aboveground biomass and leaf area index), where available, for AmeriFlux sites and supplemented these data with novel aboveground biomass estimates from tree-ring data for AmeriFlux sites (Alexander et al., 2017). We evaluate two dynamic C allocation schemes (Oleson et al., 2013; Litton et al., 2007) and two fixed C allocation schemes (Luyssaert et al., 2007) within the Community Land Model (CLM4.5) against C fluxes, stocks, and leaf area index (LAI) data at nine temperate North American forest ecosystems.
We implemented the CLM4.5 model – a well-established and commonly used LSM
– at nine different forest sites (Sect. 2.1) and compiled observation of C
fluxes, C pools, LAI, and the C
Nine temperate forests widely distributed throughout the USA were selected
for this study, including four evergreen (Niwot Ridge, Valles Caldera mixed
conifer, Howland Forest, and Duke Forest loblolly pine) and five deciduous
forests (University of Michigan Biological Station, Missouri Ozark, Harvard
Forest, Morgan Monroe State Forest, and Duke Forest hardwoods; Table 1). All
the selected forests are AmeriFlux sites (
We compiled different data streams from diverse sources for the sites (Table 1) for benchmarking C fluxes, C pools, and LAI in the model experiments. Some of the data were only available for a subset of sites and years (Table 1).
To quantify carbon flux into and out of the different forests, eddy
covariance measurements were collated from the AmeriFlux L2 gap-filled data
product (Boden et al., 2013;
To quantify the how much carbon was stored in aboveground woody biomass and
leaf biomass in these forests, we collated already existing biomass and LAI
estimates from the AmeriFlux network; these were available for only some
sites and years (Table 1). LAI measured in situ was available from AmeriFlux
data for some sites (Table 1), and we used the annual maximum LAI for all the
available measurements in each year. We used the leaf C–LAI ratio from the
AmeriFlux sites with simultaneous measurements of LAI and leaf C during the
same year (Table 1). The C
Site general information and observations available. ID refers to site name used in the AmeriFlux network.
NA
To quantify aboveground biomass at all of the sites, we surveyed each forest and calculated aboveground biomass between 1980 and 2011 (Table 1) using a dendrochronological sampling technique (Dye et al., 2016; Alexander et al., 2017). This provided a reconstruction of year-to-year variability in diameter at breast height (dbh) of trees and biomass inferred from allometric regressions. Briefly, the dbh of trees within a 20 m diameter plot was measured; all trees above 10 cm in diameter were sampled within 13 m, and trees larger than 20 cm dbh were sampled in the remainder of the plot. In Valles Caldera, rather than subsampling within a 20 m plot, all trees were sampled from two central locations until 50 samples were collected from each location following Babst et al. (2014). At the Niwot site, a point-centered quarter method (Stearns, 1949; Cottam et al., 1953) was used to estimate stand density and to select individuals for sampling. Species, dbh, and canopy position were recorded for each tree within the plots. Increment cores were dried, mounted, and sanded using standard dendrochronological procedures (Stokes and Smiley, 1968). Increments were first visually cross-dated (Douglass, 1941) and then measured under a binocular microscope and statistically cross-dated using COFECHA software (Holmes, 1983; Grissino-Mayer, 2001). Ring widths were scaled to dbh, and allometric equations (Jenkins et al., 2004; Chojnacky et al., 2014) were applied to estimate biomass through time. When available site- or region-specific allometric equations were applied, and generalized species-level allometric equations were used where these were not available. Trees that were sampled but lacked sufficient tree-ring data were gap-filled with a generalized additive mixed model to account for their biomass on the landscape (Alexander et al., 2017). At Harvard and Howland, tree-ring reconstructed biomass was compared to biomass estimated from permanent plots established in 1969 and 1989, respectively; tree-ring biomass increment estimates fell within the 95 % confidence intervals of biomass estimated from the permanent plots (Dye et al., 2016). Both permanent plots and tree-ring reconstructed biomass are dependent on allometric equations, which contributes to uncertainty in these values.
The Community Land Model (CLM version 4.5) was used to simulate C fluxes, C pools and LAI at single points (Oleson et al., 2013). CLM4.5 is a component of the Community Earth System Model (CESM1.2) of the National Center for Atmospheric Research (Oleson et al., 2013).
CLM4.5 assumes that vegetated surfaces are comprised of different plant functional types. Our sites had two different PFTs: “needleleaf evergreen tree – temperate” for evergreen forests and “broadleaf deciduous tree – temperate” for deciduous forests.
Allometric parameter values for evergreen and deciduous temperate forests in the C allocation scheme in CLM described in Oleson et al. (2013; D-CLM4.5); the alternative dynamic C allocation scheme (D-Litton) based on Litton et al. (2007); and the 2 fixed C allocation schemes (F-Evergreen, and F-Deciduous) based on Luyssaert et al. (2007). Allometric parameters represented with numbers indicate constant parameters, whereas equations indicate dynamic parameters. In the equations, NPPann is the annual sum of net primary productivity (NPP) of the previous year.
CLM4.5 includes the following plant tissue types: leaf, stem (live and dead
stem), coarse root (live and dead coarse root), and fine root (Oleson et al.,
2013). The model calculates carbon allocated to new growth based on three
allometric parameters that relate allocation between tissue types (Oleson et
al., 2013):
In addition to the dynamic C allocation structure in CLM4.5 (Oleson et al., 2013), we implemented an alternative dynamic (Litton et al., 2007) and two fixed (Luyssaert et al., 2007) C allocation parameterizations with the same structure.
The alternative dynamic C allocation structure (named “D-Litton”) was based
on carbon partitioning data along an annual GPP gradient from Litton et
al. (2007), and it considered two dynamic allometric parameters. We adapted
the original equations reported in Litton et al. (2007), converted the GPP
gradient to a NPP gradient with the general assumption that
NPP
The two alternative fixed C schemes have the same structure but different
allocation parameterizations and were based on observed values reported by
Luyssaert et al. (2007), which were converted accordingly to the allometric
parameters used in CLM. One of the C allocation parameter sets was
representative of temperate evergreen forests (named “F-Evergreen”) and the
other of temperate broadleaf deciduous forests (named “F-Deciduous”).
Similarly to Litton et al. (2007), Luyssaert et al. (2007) only provided
total root allocation without considering coarse and fine root, but the
default value for parameter
CLM4.5 uses a prognostic canopy model, with feedbacks between GPP and LAI
acting through allocation to leaf C and SLA and with SLA being a critical
fixed parameter in this feedback pathway (Thornton and Zimmermann, 2007). The
model assumes a linear relationship between SLA and the canopy depth (
We compared leaf C–LAI data from available sites with the leaf C–LAI
relationship in the model. For deciduous sites, we optimized the model
parameters based on observed leaf C–LAI. To avoid using unrealistic values
for the parameters
All CLM4.5 modeling experiments were run for nine sites, including four evergreen and five deciduous forests (see Table 1). For evergreen sites, we used the default leaf C–LAI relationship in CLM4.5, whereas for deciduous forests we used the optimized leaf C–LAI relationship (Sect. 2.5).
Each experiment represents a different allocation scheme. For experiment 1 we used the original dynamic C allocation structure in CLM4.5 (D-CLM4.5; see Sect. 2.3). For experiment 2, we used the alternative dynamic C allocation structure based on Litton et al. (2007; D-Litton, see Sect. 2.4). For experiments 3 and 4, we used a fixed C allocation structure representative of evergreen (F-Evergreen) and deciduous (F-Deciduous) forests, respectively (Luyssaert et al., 2007 – see Sect. 2.4).
The standard climate forcing provided with the model is the 1901–2013 CRUNCEP dataset. While meteorological data are available at the AmeriFlux sites, these data extend only as long as the eddy covariance observations, which are less than a decade in several cases. To explore the effects of allocation on slowly changing C pools like woody biomass, we extended model runs to 30 years, which requires using CRUNCEP or some other reanalysis climate. The CRUNCEP dataset has been used to force CLM for studies of vegetation growth, evapotranspiration, and gross primary production (Mao et al., 2012, 2013; Shi et al., 2013; Chen et al., 2016), as well as for the TRENDY (trends in net land–atmosphere carbon exchange over the period 1980–2010) project (Piao et al., 2012).
In all the experiments, we spun up the model for each site and C allocation
scheme using 1901–1920 CRUNCEP climate and assuming preindustrial
atmospheric CO
In CLM4.5 the stem turnover rate is dominated by how much woody C is lost
each year through senescence (mortality and litter). Here we define turnover
time as the total C pool divided by the rate of C input or output. We
estimated a range of plausible, site-specific stem turnover rates using
Eq. (4) below because, at individual research forest stands, rates of tree
mortality may or may not reflect averages rates across larger areas. LSMs
are typically run on scales that are coarser than those for individual forest sites and
use aggregate estimates for turnover of different C pools. CLM4.5, like many
models, is based on differential equations for the calculation of changing
biomass with time, which can be expressed as follows:
When compared to observations from the AmeriFlux sites, D-CLM4.5 usually underestimated net ecosystem exchange (NEE; Fig. 1a), and overestimated GPP (Fig. 1b) and ecosystem respiration (Fig. 1c)
Initial aboveground biomass in 1980 showed contrasting patterns in D-CLM4.5
for evergreen and deciduous forests. At evergreen sites, aboveground biomass
in 1980 was underestimated at sites with mean annual
NPP < 500 g C m
Comparisons between
Comparisons
Comparisons between
Comparisons between
D-CLM4.5 overestimated LAI relative to in situ LAI measurements (Fig. 3a). We
compared the leaf C–LAI relationship with the observed leaf C–LAI and found
important differences, especially for deciduous sites (Fig. 3b). We optimized
the parameters
Comparisons between LAI measured in situ and LAI in the model for
the different C allocation schemes (D-CLM4.5_deciduous_optimized refers to
the one with the optimized leaf C–LAI relationship for deciduous forests in
D-CLM4.5). Dashed line is
Comparisons between observed and modeled accumulated aboveground
biomass 1980–2011 for
The C
D-CLM4.5 and the alternative C allocation schemes have important differences in C allocation to each plant tissue (see Fig. S1 in the Supplement). Some of the main differences between D-CLM4.5 and the alternative C allocation schemes include increased allocation to leaf and decreased allocation to stem, especially in D-Litton at sites with low mean annual NPP (see Fig. S1 in the Supplement).
The C allocation scheme determines aboveground biomass C at
equilibrium for
The accumulated annual C fluxes (GPP, ecosystem respiration, and NEE) from
1980 to 2011 gave comparable results for the four C allocation schemes
(Supplement Fig. 2). However, the C allocation schemes resulted in
differences larger than 5000 g C m
The C allocation schemes showed differences of up to 10 % in allocation
to leaf, which produced large differences in LAI values (from
The stem turnover rate that best matched the biomass accumulation rate
estimated from the tree-ring reconstructions varied by site and was always
lower than the default rate of 2 % yr
The partitioning between leaf and stem C at these sites was best predicted by
the D-Litton scheme (Fig 8). For the range of annual NPP values at our sites
(NPP < 1500 g C m
Initial aboveground biomass showed different patterns between evergreen and
deciduous sites (Fig. 9a, b). Whereas for evergreen sites with annual
NPP < 500 g C m
From the four C allocation schemes used, two were based on fixed coefficients (Luyssaert et al., 2007), whereas the other two were dynamic and based on optimization of resources (Oleson et al., 2013; Litton et al., 2007). Of these schemes, the dynamic scheme based on D-Litton performed better than the other three. Though this scheme is imperfect, we note that on average it produces lower, and more credible, aboveground biomass estimates at the start of the simulation for these forests (Fig. 5a) and matches the biometric estimates of C partitioning between leaf and stem (Fig. 8a). The evergreen and deciduous forests appear to allocate carbon differently, and for situations where a fixed scheme is preferred our results favor the adoption of separate schemes for evergreen and deciduous forests. Below we discuss these findings in detail and make some recommendations for future development of allocation schemes.
The C allocation scheme does not strongly influence annual GPP, ecosystem
respiration, and NEE over 34 years of accumulated effect (Fig. S2 in the
Supplement); the overestimation of GPP and ecosystem respiration in Fig. 1
was common to all allocation schemes. GPP was also overestimated in previous
versions of CLM (Bonan et al., 2011; Lawrence et al., 2011). Despite
revisions of the model structure in previous versions of CLM, and that the
GPP bias was found to be most pronounced in the tropics (Lawrence et al.,
2011), our results show that the GPP is still overestimated in temperate
forests with CLM4.5. Our results support the recommendation by Thornton and
Zimmerman (2007) that additional measurements are required to establish the
variability of SLA(
None of the allocation schemes simultaneously matched observed evergreen and
deciduous forest aboveground biomass. D-CLM4.5 underestimated the modeled
aboveground biomass for evergreen sites with mean annual
NPP < 500 g C m
Percentage of NPP allocated to the each plant pool (leaf, stem, and belowground) according to observations, the four C allocation schemes used (D-CLM4.5, D-Litton, F-Evergreen, and F-Deciduous), and C allocation schemes of other models (see Fig. S1 in the Supplement).
LSMs tend to overestimate allocation to stem in temperate forest syntheses
and therefore underestimate allocation to leaves. C allocation to leaf in
D-CLM4.5 is probably underestimated when mean annual NPP is relatively close
to or greater than 1000 g C m
Although root function is complex in reality, the controls of root dynamics
and function are highly simplified in LSMs (Warren et al., 2015). It has been
suggested that resource allocation is controlled by two separate functional
trade-offs between leaf or fine roots and their supporting woody organs (Chen
et al., 2013). If this is correct, LSMs should use an allocation scheme based
on at least two (or probably three) dynamic allometric parameters, instead of
the D-CLM4.5 which is based only on one dynamic allometric parameter
(
Initial conditions used to begin transient runs or make forecasts in LSMs are usually obtained by spin-up methods. Starting from bare ground, with prescribed physical soil characteristics and plant functional type fractions, a time series of meteorological forcing variables are cycled repeatedly until the model reaches a steady state, a point when C pool sizes and fluxes remain constant between subsequent meteorological forcing cycles. This feature is exploited by Xia et al. (2012) with their semi-analytical approach to calculating these steady-state conditions. Model simulations over timescales from days to centuries critically depend on the initial variable values obtained after spin-up, and flawed initial values may produce model output that can be severely biased or unrealistic (Yang et al., 1995; Cosgrove et al., 2003; Rodell et al., 2005; Li et al., 2009). There is an increasing awareness in Earth system modeling of the critical role of these initial values after spin-up (including the initialized size of C pools – examined in 2017) that adds an extra layer of complexity in diagnosing the impact of an incorrect representation of physical processes on the transient simulation (Kay et al., 2015; Fisher et al., 2015). Our results reinforce that concern by showing that with the same climate forcing different C allocation schemes within the same LSM can produce strongly differing initial conditions after spin-up for aboveground biomass (Fig. 9). In the Supplement “Methods and Figures”, we provide an explanation for the variability in steady-state aboveground biomass depending on the C allocation scheme used in CLM4.5. In the C allocation schemes used, changing biomass with time can be expressed as Eq. (4), which are models that behave as a linear autonomous system (Sierra et al., 2017). This implies the models, when forced with equivalent meteorology and physical soil properties, will eventually converge to a steady-state independent of the starting values of the state variables, although in the case of CLM this may take many tens of thousands of years.
The NPP
Regardless of allocation scheme, CLM4.5 overestimates aboveground NPP and underestimates aboveground biomass increments (Fig. 5c), this suggests that the stem turnover rate is overestimated in the model. The underestimation of increment can be attributed to an inaccurate representation of production in the model, an inaccurate representation of turnover time of the plant pools, or both (Friend et al., 2014; Koven et al., 2015). Aboveground NPP in the D-CLM4.5 scheme from the deciduous sites, including UMBS, Morgan Monroe, Harvard Forest, and Duke hardwoods (Megonigal et al., 1997; Curtis et al., 2002) was consistently higher than that in the observations. The D-Litton scheme, however, resulted in aboveground NPP estimations that were consistently closer to the observations (data not shown). These results suggest that, in temperate deciduous forests, the D-CLM4.5 scheme is overestimating allocation to stem and underestimating allocation to roots, as previously found in other models like IBIS (Xia et al., 2015).
It is likely that CLM4.5 overestimates stem turnover at these sites.
Currently, CLM4.5 assumes a stem mortality rate of 2 % yr
Given the high uncertainty associated with turnover relative to production, it has been suggested that research priorities should move from production to turnover (Friend et al., 2014). Our results show the need for improvements of models in carbon turnover processes, a current limitation in state-of-the-art LSMs (Thurner et al., 2017). Tree-ring widths can provide reliable estimates of biomass increment, but repeated surveys of forests are required to estimate stem turnover in nonequilibrium stands (Alexander et al., 2017; Dye et al., 2016; Klesse et al., 2016; Babst et al., 2014). However, whole-ecosystem C turnover will encompass processes other than mortality, including disturbances and land-use and land-cover change (Masek et al., 2008; Erb et al., 2016; Thurner et al., 2017). Some of the aforementioned processes are already partially incorporated in LSMs, in particular land-use and land-cover change, but the lack of a mechanistic representation of the remaining processes is therefore indirectly represented in stem turnover rates. The processes controlling turnover times influence C storage capacity, but turnover is not well constrained in models (Friend et al., 2014; Chen et al., 2015; Sierra et al 2017).
Our results highlight the importance of evaluating the C allocation scheme and the stem turnover in LSMs using measures of C stocks in addition to flux data. The four C allocation schemes translated to important long-term differences in C accumulation in aboveground biomass, but gave similar results for short-term C fluxes. We were unable to distinguish between the allocation schemes using eddy flux data alone.
Data on different carbon pools are sparse, but very useful in parameterizing allocation schemes. We found that site-specific SLA was a prerequisite to evaluating the different allocation schemes; large-scale databases might be exploited to better estimate this relationship. Fixed allocation schemes preclude dynamic changes in allocation in response to varying water and nutrient availability on seasonal to interannual timescales (De Kauwe et al., 2014), but they have the advantage of simplicity. If fixed allocation schemes are used in land surface modeling, we suggest different schemes for evergreen and deciduous forests and that databases like Litton et al. (2007) and Luyssaert et al. (2007) can be used to parameterize them.
Finally, we show that information on stem turnover rate, which varies with
forest age and successional status, is important to interpret the success or
failure of different model schemes at forest sites. Stem turnover in CLM4.5
may approximate steady-state conditions on large scales, and so is
inconsistent with forests which are not at steady state. Decreasing the stem
turnover rate from 2 % yr
Ecological theory suggests that dynamic allocation probably reflects whatever resource is most limiting, but developing allocation schemes for LSMs that respond to resource limitation is challenging. The two dynamic allocation schemes reflect forest stand development to some extent, i.e., as trees get bigger (and can grow more) they tend to invest more in stem and less in leaves. However, the two schemes both use low NPP, regardless of cause, as a proxy for resource limitation (Fig. S1 in the Supplement). Cohort representation in the model would enable ontogenetic changes in allocation but would not avoid the problem that these dynamic schemes cause sites with low average NPP to perpetually allocate more resources to leaves and roots, while sites with high average NPP perpetually allocate less resources to leaves and roots (Fig. S1 in the Supplement). As coupled C–N and functional root subroutines are developed for LSMs (Shi et al., 2016), and with better representation of vegetation dynamics (Fisher et al., 2015), we could imagine a dynamic allocation scheme for CLM4.5 based on whether aboveground (light) or belowground (water and nutrients) factors are limiting.
The code for CLM version 4.5 (CLM4.5) is available
at
The data for this paper are available upon request to the corresponding author.
The authors declare that they have no conflict of interest.
This study was supported by the DOE Regional and Global Climate Modeling DE-SC0016011, the NSF Macrosystems Award 1241851 and 1241930 and by the University of Arizona Water, Environment, and Energy Solutions (WEES) 578 and Sustainability of Semi-Arid Hydrology and Riparian Areas (SAHRA) programs. The US-NR1, US-UMB, and US-MMS AmeriFlux sites are currently supported by the US DOE, Office of Science, through the AmeriFlux Management Project (AMP) at Lawrence Berkeley National Laboratory under award numbers 7094866, 7096915, and 7068666, respectively. AmeriFlux site US-MOz is supported by the US Department of Energy, Office of Science, Office of Biological and Environmental Research Program, through Oak Ridge National Laboratory's Terrestrial Ecosystem Science (TES) Science Focus Area (SFA). ORNL is managed by UT-Battelle, LLC, for the US DOE under contract DE-AC05-00OR22725. Flurin Babst acknowledges funding from the Swiss National Science Foundation (no. P300P2_154543) and the EU-Horizon 2020 Project “BACI” (no. 640176).Edited by: Carlos Sierra Reviewed by: two anonymous referees