The Ent Terrestrial Biosphere Model (Ent TBM) is a mixed-canopy dynamic global vegetation model developed specifically for coupling with land surface hydrology and general circulation models (GCMs). This study describes the leaf phenology submodel implemented in the Ent TBM version 1.0.1.0.0 coupled to the carbon allocation scheme of the Ecosystem Demography (ED) model. The phenology submodel adopts a combination of responses to temperature (growing degree days and frost hardening), soil moisture (linearity of stress with relative saturation) and radiation (light length). Growth of leaves, sapwood, fine roots, stem wood and coarse roots is updated on a daily basis. We evaluate the performance in reproducing observed leaf seasonal growth as well as water and carbon fluxes for four plant functional types at five Fluxnet sites, with both observed and prognostic hydrology, and observed and prognostic seasonal leaf area index. The phenology submodel is able to capture the timing and magnitude of leaf-out and senescence for temperate broadleaf deciduous forest (Harvard Forest and Morgan–Monroe State Forest, US), C3 annual grassland (Vaira Ranch, US) and California oak savanna (Tonzi Ranch, US). For evergreen needleleaf forest (Hyytiäla, Finland), the phenology submodel captures the effect of frost hardening of photosynthetic capacity on seasonal fluxes and leaf area. We address the importance of customizing parameter sets of vegetation soil moisture stress response to the particular land surface hydrology scheme. We identify model deficiencies that reveal important dynamics and parameter needs.
Phenological timing remains a major weakness of land surface dynamic global
vegetation models (DGVMs) that are coupled to general circulation models
(GCMs) and a primary cause of uncertainty in predicting the trajectory of
global atmospheric CO
Terrestrial biosphere models (TBMs) or dynamic global vegetation models have been developed and coupled to general circulation models (e.g., Foley et al., 1996; Cox, 2001; Sitch et al., 2003; Bonan and Levis, 2006; Dunne et al., 2013) to account for biophysical and biogeochemical processes and sometimes biogeography, allowing for the prediction of transient terrestrial ecosystem interactions with climate (Cramer et al., 2001; Friedlingstein et al., 2006). Thus, the active role of vegetation phenology can be incorporated into climate modeling. TBMs have been parameterized and evaluated on the basis of local, regional, or global scale studies. It has become common to evaluate the models at the individual field scale at sites with eddy flux measurements and detailed ground data (e.g., Delire and Foley, 1999; Arora and Boer, 2005; Krinner et al., 2005; Kucharik et al., 2006; Friend et al., 2007; Stöckli et al., 2008; Bonan et al., 2011). Still, parameterizations for vegetation processes (such as phenology and carbon allocation) implemented in TBMs are often limited to local-scale derivations due to the lack of high-quality global-scale observations of vegetation structure and function together with meteorological conditions and mechanistic understanding free of local effects.
Prognostic phenology models have been developed to predict the phenological response of vegetation to climate based on empirical evidence, as a mechanistic, process-based treatment is still not fully realizable with current understanding (Sala et al., 2012). The more commonly used climatic rule-based approach accounts for cues by temperature, soil moisture and day length cues to phenology, to predict leaf-on and leaf-off, with these controls often represented as cumulative functions of one or several climate variables that reach an empirically defined threshold (White et al., 1997). Another approach is based on plant carbon status (Bonan et al., 2003) and predicts leaf-out and senescence on the basis of potential positive carbon assimilation, which is in turn affected by temperature, moisture and sometimes nutrient conditions.
All of the above approaches require empirical parameterization of the responses to climate, and a model scheme that is independent of plant functional type (PFT) or geographical variation is still a research goal. Jolly et al. (2005) have proposed a very simple and promising bioclimatic growing season index (GSI) for phenology based on linear relations to minimum temperature, photoperiod and vapor pressure deficit (VPD, as a proxy for soil moisture), which seems to perform well compared to satellite observations at diverse sites. However, it performs less well for arid systems for which VPD may not be a good indicator of available deep soil moisture, and it is not able to capture any seasonal moisture or light sensitivity that has been observed in tropical evergreen forests (Stöckli et al., 2011). Forkel et al. (2014) adopted the concept of GSI but used the soil water availability instead of VPD for water limiting functions. Phenology depends not only on atmospheric water demand but also on water supply from soil moisture as Migliavacca et al. (2011) have shown that GSI performed better when using a soil moisture limiting function instead of the VPD limiting function. Recently, Caldararu et al. (2014) introduced a promising optimality approach based on the hypothesis that phenology is a strategy for optimal leaf area index, rather than explicit carbon exchange, driven by canopy-level demand for – and constrained by availability of – light and water, limited by leaf aging. They fitted the model to satellite observations of LAI (leaf area index) and demonstrated its capability to reproduce phenological patterns for different vegetation types over the globe within 8–16 days of observations. Top-down optimality approaches such as this may indeed be the smart way for global-scale models to capture the strategic behaviors inherent in phenology, in lieu of mechanistic understanding at the leaf or molecular level; the next step remains to couple them with explicit carbon exchange and allocation.
In this study, we perform a site-based model evaluation study for the
Ent Ent is not an acronym but the name of a sentient species of tree
in J. R. R. Tolkien's fantasy novels: Enumeration
is in order for different levels of dynamics and different physics versions
available for each of these. In order, the digits denote (1) primary
biophysics (leaf, soil biogeochemistry) and base release version (1: leaf
biophysics as described in Schmidt et al., 2014; soil biogeochemistry
described in this paper); (2) canopy radiative transfer (0: two-stream as
described in Schmidt et al., 2014; 1: ACTS model, Ni-Meister et al., 2010;
Yang et al., 2010); (3) leaf phenology (0: prescribed from satellite data; 1:
prognostic, this paper); (4) carbon allocation/growth (0: allocation with
prognostic phenology, without structural growth, this paper; 1: allocation
with structural growth); and (5) ecosystem dynamics (0: none; 1: Ecosystem
Demography scheme).
In this paper, we describe the Ent TBM's phenology and allometry scheme coupled to the ED carbon allocation scheme and evaluate their performance at Fluxnet sites (Baldocchi et al., 2001), focusing on seasonal and interannual variations of LAI and carbon and water fluxes. We compare site simulations using both observed soil moisture and that modeled by a land surface hydrology model coupled to the Ent TBM. The phenology schemes synthesize several observational data sets, combining both climate responses and a carbon balance approach, described in detail below. Here we evaluate the performance for temperate broadleaf deciduous forest, C3 annual grassland, evergreen needleleaf forest and tree/grass savanna (mixed drought deciduous broadleaf and C3 annual grassland). Through these evaluations, we are interested in quantifying the accuracy of the current model at the site level, and we identify ecosystem processes needing further improvement, with regard to both plant growth dynamics and the representation of soil moisture.
Schematic diagram of the Ent model.
The Ent TBM can be run with observed soil moisture and temperature and canopy temperature inferred from eddy flux measurements of sensible heat fluxes or, given precipitation and air temperature, it can obtain modeled soil moisture, temperature and canopy temperature if run coupled to a land surface hydrology model. For the coupled mode, we use the land model of the National Aeronautics and Space Administration (NASA) Goddard Institute for Space Studies (GISS) general circulation model (GCM) (Schmidt et al., 2006). The NASA GISS GCM land hydrology consists of six soil layers down to 3.5 m depth based on Rosenweig and Abramopoulos (1997), with updates described in Schmidt et al. (2006, 2014). The LSM computes the fluxes of heat and water vapor to the atmosphere, and the energy balance of the soil and vegetation canopy. Surface runoff is calculated based on saturation and infiltration capacity of the upper soil layer. The underground runoff is computed according to a formulation of Abramopoulos et al. (1988), which takes into account the average slope and the density of underground sinks in the cell. When running the Ent TBM coupled to the GISS LSM, soil physics parameters are taken from the land-surface-mapped data sets of the GISS LSM.
The Ent TBM is a stand-alone model developed specifically for coupling the
fluxes of water, energy, carbon and other trace gases between LSMs and GCMs.
It is structured like the ED model (Moorcroft et al., 2001) for simulating
competition in mixed canopies and disturbance dynamics by representing
vertical canopy structure through ensemble cohorts of identical individuals
and horizontal heterogeneity via subgrid patch communities. The
specifications of canopy geometry and allometry of biomass pools are
consistent with individual ellipsoidal crown geometry that is integrated with
the coupled phenology/growth model. This paper presents simulations of
seasonal variation in leaf area and mass and in fluxes of CO
Figure 1 shows a conceptual diagram of the Ent model, and how it is coupled with a GCM (or offline meteorological forcings) and an LSM. Ent's biophysics modules operate at the physical time step of the GCM or LSM. The photosynthetic uptake of carbon utilizes the well-known photosynthesis model of Farquhar et al. (1980) and Farquhar and von Caemmerer (1982) coupled with the stomatal conductance model of Ball and Berry (Ball et al., 1987), while Ent uses its own cubic solution for these coupled equations. Canopy radiative transfer is optionally modeled as in Friend and Kiang (2006) for homogeneous canopies or as in Ni-Meister et al. (2010) and Yang et al. (2010) for clumped canopies. In this paper, in lieu of detailed site allometric and canopy structure data, we utilize the homogeneous canopy radiative transfer scheme. Carbon uptake is accumulated over a day so that carbon allocation to growth, phenological behavior and mortality are updated once per day. An individual plant has distinct biomass pools, including a “labile” or carbohydrate reserve pool into which photosynthetic uptake and re-translocated carbon are accumulated: “active” pools consisting of foliage, fine roots, a reproductive pool and, for woody plants, live sapwood; and “dead” pools consisting of dead stem wood and coarse roots. Autotrophic respiration is the sum of maintenance respiration as a function of biomass and temperature, “activity growth respiration” as a function of gross assimilation and tissue growth respiration as a function of amount of new growth.
Ent takes its meteorological drivers and hydrological balance at the grid-cell or catchment-zone scale of the LSM and subgrid heterogeneity is represented as dynamic patches of vegetation communities, comprised of cohorts of plants that are ensembles of identical individuals (patch and community dynamics are not part of this study). The biomass pools and geometry of an individual woody plant are illustrated in Fig. 2. Canopy conductances from each patch are summed to the grid-cell or catchment-zone level to couple with the atmosphere. Also, root density vertical profile distributions in Ent are used to calculate a depth-weighted average of soil moisture stress. These profiles are a modification of those in Rosenzweig and Abramopoulos (1997), with details given in Appendix A.
The Ent TBM is designed to support a flexible number of PFTs. A parameter set for 17 PFTs has been developed, as listed in Table 1; however, we note that only a subset of these PFTs is evaluated here according to data availability, and the others must be approximated from the available similar types and theoretical/empirical relations from the literature. Following the rationale first advocated by Defries et al. (1995) and adopted by all vegetation models since then to varying degrees, Ent's PFTs distinguish photosynthetic pathway (C3 and C4), phenological type (evergreen, cold deciduous and drought deciduous), leaf type (broadleaf and needleleaf), growth form (tree, shrub and herbaceous) and cultivation (herb crops). In addition, to better capture community dynamics in mixed canopies, if parameter sets are provided, Ent has the capability to distinguish early and late successional species through differences in leaf life span, following the approach of the ED model (Moorcroft et al., 2001), which is based on leaf physiological relations found in Reich et al. (2007).
Ent's individual plant biomass pools and geometry. Herbaceous plants exclude woody tissue.
Plant functional types in Ent.
To capture total net carbon fluxes, the Ent TBM incorporates the code implementation of CASA' (Carnegie–Ames–Stanford Approach) from the Community Land Model 3.0 (CLM 3.0; Randerson et al., 2009; Doney et al., 2006; code kindly supplied by Jasmin John), which is based on Potter et al. (1993). For the Ent TBM, the CASA' temperature and soil moisture responses of respiration were replaced with functions derived from new fits to field data of Del Grosso et al. (2005). Details are provided in Appendix B.
As mentioned earlier, the Ent TBM can be run in several different modes of coupling: (1) a stand-alone mode when the meteorological (e.g., radiation, precipitation, air temperature, air pressure, humidity and wind) and land conditions (e.g., soil moisture, soil temperature and canopy temperature) are provided (“Ent-standalone”); (2) a mode coupled with an LSM for prognostic soil moisture and temperature given meteorological forcings (“Ent-LSM”); and (3) a fully coupled mode with an atmospheric GCM. Ent-standalone and Ent-LSM modes can be used for site-specific simulations or regional/global simulations using observed meteorological and soil moisture data.
The Ent TBM can also be run with different levels of vegetation dynamics turned on or off. In a biophysics-only mode, canopy structure and leaf area are prescribed to simulate only fluxes of water vapor, carbon dioxide and other trace gases. In an “active biomass” phenology-only mode, canopy stem structure is prescribed and static, while seasonal leaf and fine root dynamics are prognostic, and carbon that would have been allocated to stem and coarse root growth instead is allocated to litter. In a phenology-woody growth mode, in addition to leaf phenology, stem and coarse root growth are also enabled. In an ecosystem-dynamics mode, mortality and disturbance ensure that plants cannot grow indefinitely and are subject to succession and cover change (ecosystem dynamics are not covered in this paper).
The plant growth submodel integrates phenological timing and allocation of carbon to growth and litter fluxes (background litterfall and seasonal), and respiration fluxes are tied to tissue growth. The phenology scheme determines the phenological status of plants based on various environmental and climate rules studies, which determine budburst, frost hardening and senescence according to the phenological types of plants such as drought deciduousness and cold deciduousness. The carbon fixed over the course of each day from photosynthesis is accumulated and placed into a labile carbohydrate reserve pool. Carbon from the labile pool is then allocated once per day into different plant pools of foliage, sapwood, heartwood, fine root and coarse root as well as a reproductive pool according to empirical allometric relationships and leaf phenological status. In addition, tissue lost to background litter fluxes is replenished, and respiration fluxes are produced from growth of any tissue. A portion of litterfall is re-translocated back to the labile pool.
In the Ent TBM, the carbon allocation scheme takes a traditional approach of “static allocation”, based on fixed allometric relationships between different pools, adopted from approaches of the ED models (Moorcroft et al., 2001; Medvigy et al., 2009). Appendix C provides the descriptions of the ED allocation scheme, which treats “active” and “dead” biomass pools as bulk sinks, with modifications for Ent. We identified some deficiencies of the ED allocation scheme and suggest future work for improvement in Sect. 5. Also note that Appendix D provides the biophysical, phenological and allocation parameter values used in this study.
Full prognostic growth entails growth of woody structure and the size of woody plants, which would require in addition full mortality and establishment dynamics so that there is no unlimited growth; these population and community dynamics will be presented in future papers. This study focuses on the “active biomass” performance of Ent given seasonal phenology, keeping woody structure static, allocating the amount that would have gone to growth to litterfall instead.
The phenology scheme in the Ent TBM provides a synthesis and combines the climatic rule-based approach and carbon balance for deciduous plants to determine the timings and rates of leaf-out and leaf senescence by integrating several different modeling studies (Bonan et al., 2003; Botta et al., 2000; Foley et al., 1996; White et al., 1997). We present a diversity of PFTs, adding those with known behaviors that depart from common representations of cold, drought or light responses. While globally applicable parameterizations of climate rule-based phenology may still be elusive, where available in the literature, we draw from wide surveys that attempt to extrapolate to the global scale.
For deciduous plants, we use parameterizations by Botta et al. (2000). With
growing degree day (GDD) and chilling requirement, they examined the
possibility of extrapolating existing local models for leaf onset date to the
global scale by retrieving leaf onset dates from the NOAA AVHRR (Advanced Very High Resolution Radiometer) satellite
normalized difference vegetation index (NDVI). They identified appropriate
leaf onset date models and estimated their parameters for each biome, which
are implemented in other ecosystem models (Medvigy et al., 2008). The
importance of a chilling requirement is confirmed by Richardson et
al. (2012), who conducted an intercomparison of phenology predictions of
11 TBMs (and three biophysics models with prescribed phenology) at five
deciduous broadleaf and five evergreen needleleaf Fluxnet sites. They found
that, for deciduous forests, the models consistently predicted an earlier
onset of the growing season and later fall senescence than observed;
meanwhile, most models underpredicted the magnitude of peak GDD sums, while
those that explicitly or implicitly included a chilling requirement did
relatively well in capturing the onset of LAI and GPP (gross primary production) for deciduous and
evergreen forests, compared to simple temperature threshold schemes. For
drought deciduous trees and grasses, we also make use of parameterizations of
White et al. (1997) who developed a regional phenology model for the US,
predicting timings of leaf onset and offset based on the satellite NDVI at
the 20 km resolution. Their prediction errors are
For evergreen vegetation, the Ent TBM includes frost hardening for boreal evergreen plants. The frost hardening (also called winter cold hardening) involves physiological changes to protect the plant from chilling injury and freezing injury, leading to a downgrading of leaf photosynthetic capacity as well as tissue turnover and respiration. Coniferous vegetation in the boreal zone has a clear annual cycle of photosynthetic activity, with photosynthesis low or zero during the winter, increasing during the spring, peaking during the summer and decreasing during the fall. While part of the cycle is due to direct responses to PAR (photosynthetically active radiation) and air temperature, the inherent photosynthetic capacity of needles also changes (Mäkelä et al., 2004). Therefore, the models that do not account for cold hardening and de-hardening will overpredict the uptake of carbon by photosynthesis for boreal systems during the late fall through early spring. This study implements a frost-hardening algorithm based on Hanninen and Kramer (2007), Mäkelä et al. (2006) and Repo et al. (1990), who developed a model of the frost hardiness of the stems of Scots pine seedlings. Below we describe the explicit model functions reflecting our choices based on the literature above.
In the Ent TBM, several “phenological factors”,
Furthermore, the Ent TBM determines “phenological status”,
Phenostatus
Parameters in phenology submodel.
During the winter, the phenological status of cold deciduous trees and
shrubs, Phenostatus
Once leaf-out starts, trees take a number of degree days
(GDD
Fall senescence (Phenostatus
The phenological status of cold deciduous (annual or perennial) herbaceous plants
is well captured with functions based on soil temperature (TS), while that of
cold deciduous woody plants with air temperature (White et al., 1997).
Similarly to Eqs. (1) and (4) for cold deciduous trees, the soil growing
degree days (SGDD) of soil temperature (TS
Grasses begin fall senescence in response to decreased soil temperature.
Leaves start dropping once soil temperature decreases down to a given
threshold, TS
Drought deciduousness depends on available soil water for the plant. In the
model, it is determined based on a 10-day running average of the physical
time step (
The phenological factor for water stress,
The leaf-on cue for drought deciduous trees is the same as that for cold deciduous trees, while for grasses the cue is sufficient soil moisture.
Boreal plants undergo winter frost hardening, which involves physiological
changes to protect the plant from chilling injury and freezing injury.
Following Repo et al. (1990), the state of frost hardiness
The state of frost hardiness is then used to adjust the maximum
photosynthetic capacity
Site descriptions.
The Ent TBM was evaluated at five Fluxnet sites, including Morgan–Monroe State Forest (MMSF), Harvard Forest, the Vaira Ranch, the Tonzi Ranch and Hyytiäla, as briefly mentioned above (Table 3). From all sites, data from the flux tower systems were available. Meteorological driver data include radiation, precipitation, air temperature, air pressure, humidity and wind used to drive the model. Soil moisture and temperature measurements were also used to drive the Ent-standalone simulations. Flux data includes net ecosystem exchange (NEE) and evapotranspiration (ET), which were used to evaluate the simulation results. Among sites, data availability, such as LAI, varied and suited different types of model simulations as described in detail in the next section.
The MMSF, located in Indiana, USA
(Schmid et al., 2000) (latitude: 39.32315
Harvard Forest (latitude: 42.5313
The Vaira Ranch (latitude: 38.4066667
Daily simulated (S) and observed (O) phenology: (top) LAI/LAImax, (middle) phenological dates (day of year) for spring leaf-out at percentage of maximum, and (bottom) phenological dates (day of year) for fall senescence in MMS and Ha1. These results show good simulated timing of initial leaf-out and final senescence but lack of the gradual rate of these, such that the maximum leaf-out occurs too soon and the period of peak growth is too long. The gradual behavior could be simulated through a rate constraint.
Hyytiäla (latitude: 61.8474150
We performed a series of numerical experiments with Ent in different model modes in order to evaluate leaf seasonal dynamics, including leaf phenology and related water and carbon fluxes. We performed simulations for each site with observed soil moisture (hereafter denoted “Ent” mode), and LSM-modeled soil moisture (“LSM” mode); and with observed LAI (without allocation of assimilated carbon to growth; “oveg”) and dynamically modeled LAI (via carbon allocation; “dveg”), giving four experiments: Ent-oveg, Ent-dveg, LSM-oveg and LSM-dveg (Table 4). In the biophysics-only mode, the observed LAI is prescribed and related active carbon allocations are calculated according to that LAI. In the “active biomass” phenology mode, the leaf phenology and active carbon allocation are dynamically simulated. For MMSF and Harvard Forest, the model was forced with 6 and 9 years' worth of drivers, respectively. In these two sites, continuous soil moisture measurements throughout the rooting depth were not available, so only Ent-LSM simulations were performed. For Vaira, Tonzi and Hyytiäla, the model was forced with a year's worth of tower-measured meteorological drivers as well as observed soil temperature and moisture.
Types of experiments.
For the Ent versus LSM simulations for annual grass phenology, it was
necessary to tailor the soil moisture stress parameters to the different
metrics of soil moisture. The phenological timings of grasses depend on the
soil moisture condition while an LSM-derived soil moisture is a
model-specific index of soil wetness, not a physical quantity that can be
directly validated with field measurements (Koster et al., 2009). The
thresholds for the root water stress factor (
For diagnostics for model performance, we examined observed monthly LAI and
monthly sums of GPP, ecosystem respiration
(
Correlation coefficients and RMSEs of LAI-based phenological dates between simulations and observations.
We evaluated the model performance for cold deciduous woody plants at two sites: MMSF in Indiana and Harvard Forest.
Average monthly fluxes in MMS for 2002–2006 and in Ha1 for 1994–2002:
Figure 3 and Table 5 show the simulated variations of the phenological factor
(ratio of LAI to the maximum LAI of the year) in comparison to observations.
First, it is clear that the gradual nature of changes in LAI during spring
and fall were not captured in the model. The phenological factor serves as an
on/off cue between environmental thresholds, while growth rate with the ED
scheme is limited only by carbon availability, for which reserve carbon is
generally not limiting in trees (Sala et al., 2012) or in grass seeds
(William Parton, personal communication, 2008). At both sites, the
interannual variations of leaf-on timings in the spring were better captured
than those of the leaf-off timings in the fall. At Harvard Forest, the dates
with the elongation factor of 0.5 in spring showed a correlation coefficient
(
Daily root water stress factor in
Correlation coefficients and RMSEs of hourly and daily fluxes between simulations and observations.
In MMSF, the predicted NEP reasonably followed the observed NEP (Schmid et
al., 2000; Dragoni et al., 2007) with correlation coefficients ranging from
0.86 to 0.94, while the peak NEP in summer was slightly underestimated
compared to the observed (Fig. 4, Table 6). However, both GPP and
In Harvard Forest, the default simulations (LSM-dveg and LSM-oveg) showed
underestimated NEP compared to the flux tower observations due to simulated
water stress (Fig. 5). As it is known that the cold deciduous plants in
Harvard Forest do not experience water stress, no root water stress (
The ET values in both LSM simulations were overestimated compared to the flux tower observations in MMSF and Harvard Forest. These discrepancies might be attributed to both model and data errors. In the model, the higher estimated GPP (although we cannot confirm this) may lead to the overestimated ET to some extent, since higher photosynthesis corresponds to higher canopy conductance and hence more transpiration. In addition, it is well known that eddy flux measurements do not close the energy balance (Wilson et al., 2002). The sum of latent, sensible and ground heat is generally smaller than the net shortwave radiation, which is often caused by measurement errors of latent heat (i.e., ET) and sensible heat (Aranibar et al., 2006), leading to imbalance in measured net radiation of as much as 20 %. The LSM-simulated peak ET is within approximately 70 % of measurements.
We evaluated the model performance for drought deciduous herbaceous and woody plants at two sites, the Vaira Ranch and Tonzi Ranch in California. As shown in Fig. 6, at both sites, the timings of C3 annual grasses' green-up and senescence are mainly controlled by soil moisture in a Mediterranean climate, in which precipitation and temperature are seasonally out of phase. Grasses are active during the winter rains but slightly cold-limited in activity and then, with spring warming, growth and activity increase, followed by rapid senescence that closely tracks soil moisture dry down in the late spring and full senescence by the beginning of the dry, hot summer. At the Tonzi Ranch, the oaks have the opposite seasonality to the C3 grasses. The oaks leaf-out at the end of winter rains around March, when grasses have reached their peak, and then the trees start gradually losing their leaves around the beginning of July due to drought stress. Their complete leaf-off appears to be cued by the November cold or fog, but this latter cue would not be considered a stress factor and is not well understood.
Monthly fluxes in the Vaira and Tonzi ranches for 2002:
At both the Vaira and Tonzi ranches, Ent-dveg and LSM-dveg reasonably captured these phenological timings (Fig. 6). The growth rate for herbaceous plants (i.e., increase in LAI during the growing season) reflected the net carbon assimilation for each day and slightly lagged observations at the beginning of the growing season in the model. Simulated soil moisture clearly decreased much more slowly in LSM-dveg during the late spring dry down compared to the observed volumetric soil moisture that was used to drive Ent-dveg.
For carbon fluxes at the Vaira Ranch, the model simulations generally
followed the observed seasonality, although the late leaf-off in LSM-dveg
leads to significant overestimation of carbon uptake, and the observed
abrupt increase in
At the Tonzi Ranch, the simulated NEP resulted in a RMSE of
The model reasonably captured the observed seasonality of ET with an
Monthly fluxes and daily states in Hyytiäla for 1998:
At Hyytiäla, the phenological behavior of interest is frost hardening,
which lowers photosynthetic capacity in the winter. In comparison to observed
LAI, assumed according to Pasi Kolari (personal communication, 2007) and
explained in Sect. 3.2, simulated LAIs (Ent-dveg and LSM-dveg) (Fig. 8) were
almost constant at 4 m
Modeled frost hardening in the spring improved the predicted seasonality of
NEP markedly in both Ent and LSM simulations (Fig. 8, Table 6).
Frost hardening suppressed photosynthetic capacity during the winter
(particularly in February–April) and therefore GPP and NEP. It also
suppressed transpiration and thus ET, but a relatively small difference in ET
was detected between the simulations with and without the frost-hardening
scheme as the RMSEs with observations were 7.88 and 7.89 mm s
With regard to the differences between the Ent-standalone and Ent-LSM models
(Ent-dveg vs. LSM-dveg), we found the magnitude of NEP was overestimated in
Ent-dveg due to high simulated GPP and underestimated in LSM-dveg due to low
soil moisture. During the growing season, the observed volumetric soil
moisture was above
Our experiments show that phenological timing of leaf-out and senescence can be fairly well captured within 10 days or better of observations for deciduous or annual vegetation when based on cumulative weather statistics (e.g., air and soil temperature, growing degree days and day length) derived from observations in the literature. However, the response to soil moisture is sensitive to whether deep root water access is represented to offset soil moisture stress in shallower soil. Also, the soil moisture response must be tuned to the given measure or land model, because soil water content as simulated at the spatial resolution of a land surface hydrology model does not correspond well with any field measure of soil moisture (e.g., volumetric water content, matric potential and pre-dawn water potentials). Stomatal conductance and soil respiration are sensitive to soil moisture stress and hence subject to inaccuracy dependent on the soil moisture representation. Meanwhile, we uncovered weaknesses in the representation of particular vegetation processes – autotrophic respiration and ED-based carbon allocation – that, besides differences in simulated LAI at one site, were the primary causes of differences from observed NEP.
In Vaira grassland and Tonzi savanna, the phenology parameters which are based on the plant water stress factor (a function of soil moisture), were derived from the site observations of volumetric soil water content (Eq. 8) and perform well with observed soil moisture in Ent but not with simulated soil moisture in the LSM. The GISS LSM model predicted the same seasonal trends of soil moisture but higher in magnitude and lower in variability than the observations. Koster et al. (2009) point out that simulated soil moisture is a model-specific quantity and thus can be considered as an “index” of the moisture state. The specific evaporation and runoff formulations, in addition to model-specific soil parameters such as porosity, hydraulic conductivity, wilting point and layer depth define a dynamic range of soil moisture simulated by the specific model. Therefore, the true information content of soil moisture data lies not necessarily in their absolute magnitudes but in their time variability.
Therefore, the current approach using the absolute soil moisture value for water-limited phenology parameterization could be improved by properly mapping the soil moisture values from the field sites into those in the model, or by using the surrogates for the soil moisture, such as VPD, as suggested by Jolly et al. (2005). However, Stöckli et al. (2011) note that VPD may not be a good indicator of deep soil moisture.
For the trees at MMSF and Harvard Forest, LSM-simulated water stress where the plants should be unstressed indicates that calculating the water stress factor by weighting by root depth distributions does not accurately reflect how trees actually access water. Deep roots generally supply water when shallow layers are dry, and many trees perform hydraulic lift. A future revision of the Ent water stress scheme will account for the ability of plants to preferentially access soil moisture at any depth in the root zone, such that soil moisture stress is not a simple weighted average through the root profile.
While the Fluxnet data have recently been widely used to evaluate the DGVMs and LSMs, we still find the need for more comprehensive measurements at the sites. Specifically, it was very difficult to have continuous soil moisture and temperature together along with measurements from eddy covariance towers; also, the detailed tree surveys were not always available.
For cold deciduous trees, we used the growing degree days and chilling requirements in spring phenology (Botta et al., 2000) and temperature and photoperiod in fall phenology (White et al., 1997; Jolly et al., 2005). While we have taken a widely used approach, some recent studies suggest other possible approaches. For spring phenology, the importance of the photoperiod has been pointed out in recent studies (e.g., Körner and Basler, 2010; Migliavacca et al., 2012). Körner and Basler (2010) suggested that when the chilling requirement is fulfilled, plants become receptive to photoperiod signals and such sensitivity to the photoperiod is found in late successional species in mature forests. For fall phenology, Delpierre et al. (2009) used chilling a degree-day photoperiod to model leaf coloring change for deciduous trees in France, and Yang et al. (2012) and Archetti et al. (2013) found the model suitable for New England, US, with different parameter fits. In general, despite agreement about overall climate cues for cold deciduousness, further work is needed to uncover site-independent parameterizations.
In this study, site-specific parameters were used according to the data availability. As in Appendix D, some of parameters are generic for PFTs and some are site-specific. For the model to be utilized at the global scale, further exploration is required to determine geographic variation in parameters and possible climatology-based prediction of parameters. Model parameters for biophysics or ecosystem models have been inferred with various mathematical techniques, such as a Monte Carlo simulation (Kleidon and Mooney, 2000), data assimilation with Kalman filtering (Mo et al., 2008; Stöckli et al., 2008), optimization with the Marquardt–Levenberg method (Wang et al., 2007) and optimization with the simulated annealing method (Medvigy et al., 2009; Kim et al., 2012).
In general, vegetation biophysics models can replicate observed canopy fluxes
of CO
Autotrophic respiration can range
In Ent, using site-specific parameters for leaf photosynthetic capacity,
We encountered deficiencies in the carbon allocation/growth scheme that we adopted from the ED model. Although the current carbon allocation and growth scheme results in LAI that is reasonable, with some phenological timing issues as noted, the maximum LAI is achieved thanks to a cap on LAI by allometric relations to stem structure and plant density, while the rest of the plant carbon balance is not realistic, particularly with regard to rate of LAI growth, amount of seasonal sapwood growth and conversion to heartwood, accumulation of carbon reserves and allocation to reproduction. The on/off cues of the Ent phenological factor for cold deciduous trees results in an unrealistic fast full leaf-out, which could be rectified by introduction of a physically based cell growth elongation factor (Lockhart, 1965). We also found it would be more realistic to make carbon allocation to each live pool independent. The ED scheme allocation to one live biomass total and then partition among the live pools can lead to unrealistic behaviors for sapwood patterns during spring growth and fall senescence, due to a partitioning scheme for live carbon that does not account for the different seasonal behaviors of each live pool. Finally, reproduction in ED is currently a fixed fraction of assimilated carbon, which is problematic in the plant's overall carbon balance as a large sink. Recent studies show that reproduction relies heavily on stored carbon, which often accumulates over more than a year, such that growth of other plant tissue is never carbon limited while large stores are kept in reserve. The ED scheme relies on the plant using nearly all stored carbon for deciduous plants each year. Introducing reproductive allocation based on thresholds proposed by Sala et al. (2012) would help rectify Ent's simulated plant carbon balances such that trees are not always reaching the limit of carbon starvation. Besides respiration, plant carbon allocation is currently still poorly understood. However, recent studies with carbon tracers (Epron et al., 2012a, b) are yielding new insights that could be used to improve growth schemes that continue to be a weakness in dynamic global vegetation models.
In this study, we evaluated the Ent TBM focusing on the seasonal dynamics of vegetation leaf as well as carbon and water fluxes. In particular, we took a process-based approach, evaluating the Ent-standalone model with observed LAI and Ent's prognostic active growth submodel with observed soil moisture as well as coupled to the LSM model for prognostic soil moisture, allowing us to identify parameterizations that need to be improved. For herbaceous PFTs whose phenological timings depend on soil water availability, it is inevitable to find errors in phenological timing in Ent-LSM simulations due to the discrepancy in simulated soil moisture in the LSM. Also, the predicted LAI of herbaceous PFTs in Ent directly reflects the amount of assimilated carbon on the day and vice versa as herbaceous PFTs allocate assimilated carbon only to active compartments (as they have no structural tissue) and thus any errors in phenological timings propagate into errors in biophysical processes. For tree PFTs, the Ent soil moisture stress scheme should be improved to allow for deep soil moisture access to override stress that might result from weighting shallower dry soil layers too strongly.
This study evaluated the phenology and resulting seasonality of fluxes in the limited number of sites, including four different PFTs. The Ent PFTs not tested in this study include deciduous needleleaf plants, evergreen broadleaf plants, shrubs, arctic grasses and crops. Future work will involve determining the efficacy of these PFT parameterizations at the global scale and the possibility of developing some of these parameters as functions of local climate as obtained from either reanalysis data or from GCM climatology. In addition, we have identified deficiencies in the carbon allocation scheme from the ED model that can be rectified in future revision of Ent's growth submodel.
Future work will include development of phenology and allometry parameter sets that are robust at the global scale and soil moisture stress accounting for deeper soil access. In addition, due to how ED allocates biomass to all live pools (e.g., foliage, sapwood and fine roots) combined, rather than allowing for separate dynamics, alternative carbon allocation schemes that partition the dynamics of the live tissues must be developed for realistic plant carbon balances.
This work sets the foundations for coupled land carbon–GCM simulations that can utilize height-structured canopy data from remotely sensed lidar to reduce uncertainty in predictions of the land carbon balance through tighter links between seasonal growth dynamics geometrical and biomass allometry of vegetation canopies. Because the model at the global scale will involve a community of users that will continue to identify parameter sets applicable for more climatically diverse distributions of the Ent TBM's PFTs, this paper is also written to serve as a detailed reference for these users, to achieve appropriate interpretation of model results and parameter adjustment.
Depth profiles of root density are modifications of those in Rosenzweig and
Abramopoulos (1997), revised to fit the PFT categorizations in the Ent TBM.
These are modeled as cumulative normalized root density distributions
Plant functional type parameters for root density distributions.
The soil biogeochemistry submodel of Ent utilizes a slightly modified
version of the CASA' biosphere submodel originally implemented in the NCAR (National Center for Atmospheric Research)
LSM and CSM 1.4 (Bonan, 1996; Randerson et al., 1997; Fung et al., 2005;
Doney et al., 2006), which itself is a modified version of the original
NASA CASA biosphere model (Potter et al., 1993). The soil model determines
terrestrial soil carbon pools and CO
The soil biogeochemistry model consists of three litter C and N pools and nine soil C and N pools, as in CASA'. The pools are currently only simulated for the top 30 cm soil depth. This layer accounts for nearly all observable soil respiration fluxes to the atmosphere but not for full long-term carbon stocks in deeper soil. Simulating soil carbon down to 100 cm and deeper would allow comparison to existing global data sets of soil carbon and root depths (Batjes, 1996a, b; Jackson et al., 1996). Figure B1 shows these 12 pools. Ent has an optional 30–100 cm deep soil layer that is not run in the current paper.
The various pools currently have fixed C : N ratios and turnover times, listed in Table B1. The pools gain carbon and nitrogen from transfers from other pools and losses to respiration and transfers to other pools. These transfer and respiration fractions are listed in Table B2.
Soil micrometeorological conditions for the soil layers must be extrapolated from the soil layering scheme of the land surface model. For example the GISS land surface hydrology has a six-layer soil scheme with geometrically increasing layer thicknesses with depth (Rosenzweig and Abramopoulos, 1997), so soil temperature and moisture for the soil biogeochemistry layers are calculated through a weighted sum for the upper 30 cm.
Schematic diagram of the soil biogeochemistry submodel of
Ent (showing nine soil C pools only; modified from Potter et al., 1993).
Surfstr – surface structural pool; Surfmet – surface metabolic pool;
Soilmet
In addition to the transfer coefficients in Table B2, three other rate
coefficients are used (following Randerson et al., 1997):
Physical inputs to the soil module from the land surface hydrology are
volumetric soil moisture, soil temperature and soil texture (percentage of
clay, sand and silt). Biological inputs consist of leaf, root and wood
litter (Fig. B1). Model outputs are soil C (and N, not used) pools and soil
CO
Values of C pool parameters: C : N ratio of all
12 C pools (used only to calculate N pools); annk
The relevant PFT-dependent litter parameters (leaf,
fine root and wood turnover times, litter C : N ratios, specific leaf
area and lignin contents) from Ent are listed in Table B2. In addition to
these parameters, a parameter representing the inverse of the residence times
of the litter pools, denoted annk
We replaced the CASA' temperature and soil moisture responses of soil
respiration with new functions derived from new fits to field data collected
by Del Grosso et al. (2005). The Ent TBM temperature response of soil
respiration is a simple piecewise linear model that increases up to
30
Values of respiration pathway coefficients: eff – microbial respiration transfer efficiencies for all 14 pathways; frac_donor – additional respiration efficiencies (both unitless).
Respiration and transfer efficiencies for the physical time step are from the CASA' code as implemented in the National Center for Atmospheric Research (NCAR) climate model (CSM) 1.4 (Doney et al., 2006).
The Ent TBM moisture response of soil respiration is similarly a piecewise
linear model that rises from 0 at zero soil moisture to 1.0 at a relative
extractable water content (REW) of 0.7, where REW is the fraction of
saturation above the hygroscopic point. Because there are no good functions
for calculating the hygroscopic point based on soil texture, we estimate the
hygroscopic point as half of the wilting point. We note that it would be
more precise to model the soil moisture response as an optimality curve
that rises from the soil hygroscopic point (minimum for microbes) rather
than wilting point (for plants) to some optimum and then declines as pore space
becomes saturated and obstructs the flux of gases. However, because of a lack of good
algorithms to calculate the soil hygroscopic point for different soil
textures, we use this version of Ent relying on the wilting point as the
point of minimum available soil moisture. We may later introduce a simple
linear decline of the soil moisture response with saturation; however, at present
we have no data on the response to saturated conditions.
The labile carbon reserves in Ent are allocated into different plant biomass
pools, including foliage, sapwood, heartwood, fine root and coarse root. In
addition, turnover of tissue due to background litter fluxes is replenished
from the carbon reserve pool. In nature, plants may allocate biomass to
different compartments in response to many different controlling factors,
such as light availability and water availability, which alter, for example,
root : shoot ratios. Among various carbon allocation modeling approaches
with different complexities, many DGVMs take a simple approach to model
carbon allocation via empirical and allometric relationships, a traditional
“static allocation” approach (Foley et al., 1996; Sitch et al., 2003) while
some models parameterize the dependency of carbon allocation on resource
availability, “dynamic allocation” approach (Friedlingstein et al., 1999;
Arora and Boer, 2005). Although carbon allocation varies with plant
characteristics such as size and age and environmental conditions, the static allocation
approach may be justified for models operating at large scale. If plant
productivity is assumed in a steady state, carbon allocation is likely to be
in a steady state. Also, spatial variability in environmental factors and
their effects on allocation can be averaged. However, the fixed allocation
approach is limited in long-term simulations as it lacks response to
environmental changes such as climate change and elevated atmospheric
CO
In the Ent TBM, the allocation submodel takes a traditional approach of static allocation, based on allometric relationships between different pools. Modified from approaches of the ED models (Moorcroft et al., 2001; Medvigy et al., 2009), the scheme allocates the labile carbon to different biomass pools according to empirical allometric relationships and leaf phenological status on a daily basis.
The biomass within each plant is partitioned between an active carbon pool
and a structural carbon pool. The active biomass pool (
Thus, the time change of the active pool can be written as
Finally,
Growth of structural tissue is handled as follows. If the stored labile
biomass is non-zero, the size of the structural pool of woody plants
increases according to the empirical allometric relationships and
consequently the size of the active pool increases. Here, the partitioning
between
See Tables D1 and D2.
Biophysics parameters for Fluxnet sites in this study.
Biogeochemical and phenological parameters for Fluxnet sites in this study.
The Ent TBM is being developed as a part of NASA GISS ModelE. Version
1.0.0.0.0, Ent biophysics, is available at
This research was supported by two grants from the NASA Earth Science, Modeling, Analysis & Prediction Program (MAP/04-116-0069), for proposals titled Ent: A Global Dynamic Terrestrial Ecosystem Model for Climate Interactions at Seasonal to Century Time Scales Through Coupled Water, Carbon, and Nitrogen Dynamics (PI: Nancy Y. Kiang), and NASA Goddard Institute for Space Studies Global Model Development (PI: James Hansen and Gavin Schmidt). Support was also provided in part by the NASA Postdoctoral Program (NPP) at NASA GISS administered by Oak Ridge Associated Universities through a contract with NASA, for a proposal titled Role of phenology in coupled vegetation-climate at seasonal to decadal timescales in Ent DGTEM (NPP fellow: Yeonjoo Kim). We also wish to express our gratitude to the many researchers who made available large amounts of data from their Fluxnet sites, particularly Danilo Dragoni, HaPe Schmid, and Craig Wayson for Morgan–Monroe State Forest; Dennis Baldocchi for the Vaira and Tonzi ranches; Steve Wofsy and co-workers for Harvard Forest; and Timo Vesala and Pasi Kolari for Hyytiälä. The websites of publicly available data are listed in Table 3. Edited by: C. Sierra