Accounting for herbaceous communities in process‐based models will advance our understanding of “grassy” ecosystems

Grassland and other herbaceous communities cover significant portions of Earth's terrestrial surface and provide many critical services, such as carbon sequestration, wildlife habitat, and food production. Forecasts of global change impacts on these services will require predictive tools, such as process‐based dynamic vegetation models. Yet, model representation of herbaceous communities and ecosystems lags substantially behind that of tree communities and forests. The limited representation of herbaceous communities within models arises from two important knowledge gaps: first, our empirical understanding of the principles governing herbaceous vegetation dynamics is either incomplete or does not provide mechanistic information necessary to drive herbaceous community processes with models; second, current model structure and parameterization of grass and other herbaceous plant functional types limits the ability of models to predict outcomes of competition and growth for herbaceous vegetation. In this review, we provide direction for addressing these gaps by: (1) presenting a brief history of how vegetation dynamics have been developed and incorporated into earth system models, (2) reporting on a model simulation activity to evaluate current model capability to represent herbaceous vegetation dynamics and ecosystem function, and (3) detailing several ecological properties and phenomena that should be a focus for both empiricists and modelers to improve representation of herbaceous vegetation in models. Together, empiricists and modelers can improve representation of herbaceous ecosystem processes within models. In so doing, we will greatly enhance our ability to forecast future states of the earth system, which is of high importance given the rapid rate of environmental change on our planet.


| INTRODUC TI ON
Herbaceous and grass-dominated ecosystems are critical components of Earth's land surface (Faber-Langendoen et al., 2016), occupying a large proportion of Earth's ice-free terrestrial surface (Dixon et al., 2014;White et al., 2000) and providing many important ecosystem services such as carbon sequestration, livestock grazing, and wildlife habitat (Bai & Cotrufo, 2022;Gibson, 2009).
Examples of these "grassy" ecosystems are grasslands, shrublands, tundra, open tree savannas, and woodlands.Human impacts, such as land use change, suppression or ignition of fire, and climate change, are changing herbaceous ecosystems rapidly (e.g., Bowd et al., 2018;Dieleman, 2016;Harrison, 2020;Komatsu et al., 2019;Myers-Smith & Hik, 2018).Such changes not only alter local and regional ecosystem functioning provided by these ecosystems (Smith et al., 2009), but they may also feed back to the climate system to impact ecosystem processes and services beyond their geographical boundaries (Cook et al., 2009;Zhou et al., 2019).To capture this broad range of ecosystem changes and their impacts on human and climate systems, it is necessary to develop and improve models that provide detailed representations of herbaceous ecosystem processes.With these models, it is also possible to evaluate mechanisms underlying ecological patterns at scales that are difficult to measure empirically (e.g., Luo & Reynolds, 1999;Walker et al., 2015), and to make predictions for novel environmental conditions without the need for spatial extrapolation (Evans et al., 2013).This is particularly important as global change and human demand for natural resources are pushing many herbaceous ecosystems beyond historical boundaries (Alexander et al., 2015;Easterling et al., 2000;IPBES, 2019;IPCC, 2021;Wang et al., 2022).
Herbaceous plant communities can be significantly transformed by climate change, including shifts in dominance from one herbaceous species to another (Collins et al., 2020) or replacement by other plant functional types (e.g., shrubs; Collins et al., 2021;D'Odorico et al., 2012).Herbaceous plant community dynamics have been shown empirically to depend on the type, magnitude, and duration of global change drivers (Komatsu et al., 2019).In some cases, punctuated variability in environmental conditions (e.g., extreme climatic events, pulse events sensu Lake, 2000) may drive temporary shifts in plant community composition (Figure 1a); alternatively, monotonic shifts in the environment (e.g., increased average rainfall, press events) are more likely to result in altered coexistence dynamics (Van Dyke et al., 2022) and directional change in plant community composition (Figure 1b; Smith et al., 2015).These community changes may have additive or multiplicative consequences for ecosystem function (Smith et al., 2009;Suding et al., 2008).Experimental and observational studies have shown effects of herbaceous community change on soil biogeochemical cycling (Hobbie, 1996;Kardol et al., 2010;Sundqvist, 2011;Williams & Vries, 2020;Xu et al., 2015), net primary productivity (Shi et al., 2018;Wilcox et al., 2016), carbon sequestration (Petrie et al., 2015), abundance and diversity of other trophic levels (Hartley & Jones, 2003;Scharfy et al., 2010;Stevens et al., 2018), and changes in surface albedo and evapotranspiration (Chapin et al., 2005;Scott & Sudmeyer, 1993;Tang et al., 2020).
Given that shifts in plant community composition and structure can have a range of effects on ecosystem function (Figure 1c; Langley & Hungate, 2014;Polley et al., 2014;Shaw et al., 2022), improved forecasting capability is needed to represent ecosystem functioning more effectively as plant community composition changes in response to global environmental change.Additionally, we posit that better representation of herbaceous community dynamics will improve our ability to use first principles to model distributions of herbaceous vegetation, which is currently limited (Griffith et al., 2015;Still et al., 2019).
Here, we address the above challenges and provide a way forward for both model development and empirical research for improving herbaceous representation in land surface and earth system models (LSMs and ESMs, respectively).More specifically, we: (1) Provide a brief history of how herbaceous vegetation dynamics have been represented in process-based models, which offers the necessary background for the discussion of current knowledge gaps and future improvements.(2) Report on a model simulation exercise where we compared modeled plant community change and aboveground net primary productivity (ANPP) to observations from a long-term water addition experiment in a US tallgrass prairie.This was done to assess current model capabilities to simulate herbaceous community dynamics and resulting effects on ecosystem function.(3) Identify herbaceous properties and ecological processes important for accurately simulating herbaceous vegetation dynamics.In so doing, we hope to provide general approaches for modelers and empiricists to work together toward a better understanding and more predictive model representation of herbaceous plant community and ecosystem dynamics.

| REPRE S ENTATI ON OF HERBACEOUS PL ANT COMMUNIT Y DYNAMIC S IN MODEL S: A B RIEF HIS TORY
There has been a rich history of using process-based models to simulate and predict vegetation dynamics and their effects on ecosystem functioning.Dynamic global vegetation models (hereafter DGVMs) are gridded, process-based ecosystem models that represent the biophysics, biogeochemistry, and biogeography of global vegetation and soil, with different models having substantial variation in capabilities, completeness, and approaches.
Because of their computational expense, and to be able to map vegetation with satellite observations, these models simplify vegetation diversity into a few major types to capture the dominant contributors to surface-atmosphere exchange and the carbon storage in biomass and soil carbon.Vegetation types are typically categorized into a limited selection of "plant functional types" (PFTs), on the order of 5-20 classes, distinguished variously by growth form (grass, shrub, tree), allometric relations, photosynthetic pathway (C 3 , C 4 ), phenology (deciduous/evergreen, rain green, annual/perennial), leaf type (broadleaf, needleleaf), and climate range (e.g., arctic, temperate, tropical; reviews by Argles et al., 2022;Fisher et al., 2018;Prentice et al., 2007).Each PFT's physiology and structure is charac- functions, such as carbon and water cycling (e.g., Sitch et al., 2008).
Representation of size-structured competition in mixed communities for above-and belowground resources has only recently been introduced in a few DGVMs (see review by Fisher et al., 2018).
We refer to earlier DGVMs that lack size-structured communities as "first-generation DGVMs", to differentiate them from the newer mixed community models, often referred to as "demographic DGVMs" or "vegetation demographic models" (VDMs) that represent individuals or cohorts of plants within PFTs.Gap models-a type of VDM-are not global in scale but were used to investigate individual forest dynamics by explicit representation of different sized individual plants able to compete for light and other resources (Bugmann, 2001;Mladenoff, 2004), driving canopy growth and shifts in species composition and community structure.Gap models are computationally expensive and must be parameterized to represent species-level interactions and community dynamics at a specific site (Botkin et al., 1972;Pacala et al., 1996;Shugart et al., 1992Shugart et al., , 2018)).
Theoretical constructs have been developed to bridge the firstgeneration DGVMs and gap models, such as the ecosystem demography (ED) model (Moorcroft et al., 2001) and the perfect plasticity approximation model (Strigul et al., 2008).They group similar individuals into cohorts to simulate succession dynamics, but with different schemes for modeling crown organization and height-structured light competition.These theories have been applied singly or together within land models (e.g., Geophysical Fluid Dynamics Laboratory [GFDL] Land Model 4.0.(Martínez Cano et al., 2020;Weng et al., 2015;Zhao et al., 2018), functionally assembled terrestrial ecosystem simulator (FATES; Fisher et al., 2015;Koven et al., 2020), and robust ecosystem demography (RED, Argles et al., 2020).By incorporating these theories that allow for finer scale groupings than simple PFTs while keeping model complexity low, land models can offer a more computationally efficient approach for representing size-based competition and compositional succession that are intermediate between the first-generation DGVMs and gap models.
Model development on community dynamics has primarily been for forests, while herbaceous ecosystems have received less attention (Detto et al., 2022;Pacala et al., 1993;Shugart et al., 1992;Wilcox, Koerner, et al., 2020).However, there has been relatively recent progress in representing herbaceous ecosystems within DGVMs.First, Euskirchen et al. (2014) improved phenology of four arctic PFTs (heath tundra, tussock tundra, wet sedge tundra, and shrub tundra) using TEM-DVM (Euskirchen et al., 2009), which had substantial ramifications for projections of C fluxes in the arctic.
Second, Scheiter and Higgins (2009) describe the adaptive dynamic global vegetation model (aDGVM) which uses an individual-based framework to improve representation of tree-grass dynamics in fire adapted savannas.This model supports two general PFTs, one for trees and one for grasses.Within the tree PFT, individual trees are simulated where C allocation and phenology of each tree vary with environmental conditions.Within the grass PFT, two "superindividuals" are simulated to represent grass between and under tree canopies.Scheiter et al. (2013) added trait filtering to the model to create aDGVM2, allowing for distributions of functional traits to be assigned among individuals.Third, Boone et al. (2018) used components from CENTURY (Parton et al., 1993) and SAVANNA (Coughenour, 1992) models to create G-Range.G-Range incorporates two herbaceous vegetation types and implements a schematic for herbivory.These important forays into improving representation of herbaceous components within models can be built upon (Fitz & Trimble, 2006;Oleson et al., 2004) as we find ways to relieve constraints in computational ability and advance our understanding of the ecophysiological, morphological, and demographic mechanisms driving herbaceous plant communities.
Many models only representing two to three natural herbaceous PFTs (e.g., C 3 , C 4 , and sometimes C 3 arctic grasses, Table 1), which limits the ability of these models to predict herbaceous community shifts.For comparison, there are 11 woody PFTs in the Community Land Model 5.0 (Lawrence et al., 2019).If groups of herbaceous vegetation types respond differently to biotic and abiotic drivers, then overly coarse PFTs for herbaceous vegetation may be part of the reason for difficulties in modeling distributions of C 3 versus C 4 vegetation (Fox et al., 2018;Still et al., 2019) or forest-grassland boundaries (Poulter et al., 2011).For example, two different and globally abundant C 4 grass lineages, Andropogoneae and Chloridoideae, have substantially different climatic envelopes and fire return intervals where they exist, and the Chloridoideae distribution overlaps with many C 3 lineages (Lehmann et al., 2019).It follows that a single C 4 PFT may limit the ability of the model to accurately predict where the boundaries of C 4 plants exist (Griffith et al., 2020).How many and the types of herbaceous PFTs to add is an open question that needs further research.We approach this subject more fully in Section 4.
These model weaknesses in representation of herbaceous plant form, function, and diversity are apparent in current DGVMs' inability to reproduce globally observed herbaceous community changes that are having strong effects on ecosystem function.For example, most vegetation models do not differentiate annual grasses from typical C 3 or C 4 grass PFTs.Yet, invasive annual grasses have expanded across large portions of the United States due to their annual life cycle and impacts on fire regimes (McArthur et al., 1990;Smith et al., 2022).This expansion has caused declines of other herbaceous functional types leading to loss of forage for grazing animals, a critical function of many herbaceous systems (Bartolome 13652486, 2023, 23, Downloaded from  b Weng and Luo (2008).
In sum, we posit that current limitations in model representation of herbaceous communities may preclude our ability to forecast the full effects of both monotonic environmental shifts and increased frequency of extremes on ecosystem function in grassy ecosystems (Figure 1).

| COMPARING MODEL S IMUL ATIONS TO OBS ERVATI ON S OF HERBACEOUS COMMUNIT Y DYNAMIC S AND ECOSYS TEM FUN C TION
To demonstrate current capabilities in modeling herbaceous ecosystems, we selected four different process-based models that vary in their representation of community change from no community representation to VDMs (Table 1).These models were used to simulate a long-term irrigation experiment at a US tallgrass prairie (Konza Prairie Biological Station) following a standardized simulation protocol (Appendix S1).Specifically, we were interested in how model output under a monotonic increase in water availability matched up to experimental findings of (1) herbaceous plant community change and (2) ANPP before and after the plant community change.
When the irrigation experiment began in 1991, a single C 4 perennial rhizomatous grass species, Andropogon gerardii, comprised ca.60% of the plant canopy coverage.Andropogon gerardii largely retained dominance in both the control (mean annual precipitation: 847 mm) and water addition (mean annual precipitation + water addition: 1108 mm) plots over the first 10 years of the experiment (Collins et al., 2012).The persistence of A. gerardii during this time was likely due to inertia exerted by the clonal nature of the species, such that replacement by more mesic species only occurred after slow infiltration into the rooting and tillering zone of the grassland.While A. gerardii was the dominant species at the site, the effect of water addition on ecosystem function (ANPP) was relatively consistent, increasing ANPP by an average of 49% compared with control plots (Knapp et al., 2012).
Around 10 years into the experiment, cover of Panicum virgatum (another C 4 perennial rhizomatous grass) increased substantially in the irrigated plots, becoming the new dominant species for the following 10 years of the experiment (Wilcox et al., 2016).This species shift was accompanied by a more than doubling of the effect of water addition on ANPP in the irrigated plots, from a 49% to 140% increase in ANPP caused by irrigation (Knapp et al., 2012;Wilcox et al., 2016).Further details about experimental implementation, measurements, and additional findings can be found in Collins et al. (2012).
To simulate the Konza irrigation experiment, we used meteorological forcing data, environmental information (e.g., soil texture), and functional traits for A. gerardii and P. virgatum (collated from author measurements [Table S1] and TRY; Kattge et al., 2020) to calibrate and operate four models (Table 1) from 1991 to 2012.
The goal was to evaluate the models' ability to simulate the community shift and accompanying changes in ANPP (see Appendix S1 for additional details about simulation protocol).For these simulations, two PFTs were parameterized based on empirical information about functional traits for A. gerardii and P. virgatum (Table S1) and models were tuned to match observations from non-irrigated plots (i.e., controls).This led to slightly different starting PFT abundances among models when irrigated simulations began (Figure 2a).
To aid interpretation of results from model simulations, we assessed model performance with regards to both the change in composition (represented by responses of two PFTs designed to represent A. gerardii and P. virgatum) and ANPP.Comparable observations from the irrigated plots were available for both metrics.

| PFT dynamics
We evaluated whether the models were able to simulate both the PFT were not sufficient to ever allow for displacement of the initial dominant species, A. gerardii.Instead, the empirical explanation for the slow turnover of plant community was likely because P. virgatum spreads relatively slowly via clonal growth when plant competitors are present (Hartnett, 1993), which leads to lagged responses of species abundances during monotonic shifts in environmental conditions (Collins et al., 2012).
SDGVM was able to capture both the direction and magnitude of PFT change in the long term but failed to do so in the short term.This is again likely due to a mismatch between the time frames of changes in plant abundance in models versus observations.SDGVM represented the P. virgatum PFT as competitively superior to the A. gerardii PFT under irrigated conditions and was even able to represent coexistence in the long term.However, in the model this change happened within the first year of irrigation while observation suggests that plant competition between these two species should take much longer.This occurrence may present difficulties when trying to represent pulses of water availability (i.e., wet years) since models would predict full plant community change in a single year.
LPJ-GUESS was able to represent all but the magnitude of long-term community change.However, this was only after a modification was made to the model.Initially, relatively wet conditions throughout the growing season and higher growth potential of P.
virgatum facilitated competitive exclusion even under non-irrigated conditions.LPJ-GUESS was then modified to allow for greater percolation rates overall and when the shallow soil layer (0-50 cm) was at field capacity.This created drier conditions overall without irrigation and provided A. gerardii an advantage due to its greater root density in shallow soil layers (Table S1).Then under irrigation, deeper soil layers became chronically wetter, which allowed for P. virgatum growth to increase and reduce the abundance of the A. gerardii PFT, eventually excluding it from the community.
Calibration of LPJ-GUESS points to the importance of properly representing water physics and rooting densities with depth when representing herbaceous systems.
Aboveground net primary productivity dynamics-Interestingly,

| DIREC TIONS FORWARD: THE INTEG R ATI ON OF EMPIRI C AL UNDER S TANDING AND MODEL REPRE S ENTATI ON FOR KE Y HERBACEOUS COMMUNIT Y PROCE SS E S
Based on our review of the model development history and the simulation exercise above, we suggest that current model representations of herbaceous communities and their dynamics, and the effects of community change on ecosystem function are still relatively limited.While several excellent texts have described many of the potential drivers of processes that structure herbaceous communities (Keddy & Laughlin, 2021;Mittelbach & McGill, 2019;Morin, 2009;Tilman, 2020), many of these drivers are either not used by models or are not in forms readily usable by process-based models.For example, we know from empirical studies that nitrogen addition F I G U R E 3 Top: Allometric descriptions of three potential shapes for herbaceous PFTs to be incorporated in vegetation models.The artist depictions of the shapes in the bottom row are inspired from illustrations of US prairie plant species by Weaver (1954).Bottom: 3-D model created from measurements from a 4.5 × 4.5 m plot in a South African savanna describing the following measurements of three different bunchgrass species: basal and canopy diameters, vegetation height, and volumetric bunch density (transparency).The yellow cylinder on the right is 1 m tall for scale.The 3D model is adapted from Wilcox (2023).
in many herbaceous ecosystems reduces the abundance of native perennial species and species that form relationships with nitrogen fixing symbionts (Suding et al., 2005).A proposed mechanism for this phenomenon is that nitrogen addition often causes taller, fast-growing (i.e."weedy") species to increase in abundance, which reduces light penetration and causes reduction or elimination of shorter statured or slower growing species (Dickson et al., 2014).To represent the tradeoff between light and nutrient competition, we need to better understand nitrogen-fixation rates for native perennial vegetation (Taylor et al., 2020) as well as specific morphological, ecophysiological, and phenological information (Table 2) about how and why these tall fast-growing species perform well under nitrogen addition (i.e., lower root:shoot ratios, higher photosynthetic rates, earlier phenology).On the modeling side, we need to further develop model structure to allow for nitrogen fixation at different rates or at least allow for nitrogen uptake rates that implicitly represent these processes without necessarily linking to root production and distribution.Additionally, finer PFT groupings for herbaceous vegetation would be necessary to represent these types of community shifts.Although challenging, we content that improving herbaceous representation of processes such as these will enhance the predictive capacity for a large component of the Earth's terrestrial surface under future global change scenarios.
In the remainder of this review, we suggest numerous ways forward for improving model treatment of grassy ecosystems.
However, it is important to note that the complexity, structure, and parameterization needed within models will vary substantially across projects.Ultimately, the improvements and level of detail required will depend on the specific objectives of particular modeling endeavors.Individual researchers or research teams can use data to parameterize models to address specific questions, but the underlying model structure is needed to enable this parameterization.
For example, in our simulation above, representation of the slow In Table 2, we provide a summary of some key empirical needs for better representation of the diverse herbaceous plant functions and competitive strategies within models.Then below, we work through several aspects of herbaceous growth and competition where additional research would enhance predictive capacity of herbaceous dynamics within models.These include structural growth of herbaceous plants, uptake and competition for water and nutrients, vegetation dispersal and spatial dynamics, and fire and grazing regimes.In each section, we outline how current models represent these processes, current gaps in model representation, current empirical understanding of these processes, and finally, directions forward for both empiricists and modelers.From empirical work, we know that many herbaceous plants have dense canopies and competition for light can drive numerous ecological processes (Dickson et al., 2014;Eskelinen et al., 2022).In addition, herbaceous plants exhibit multiple distinct growth forms, which affect their response to environmental conditions and their potential representation in models.Forb and grass species grow either as a single stem or as clones forming bunches, tufts (i.e., caespitose), or as rhizomatous spreaders forming mats of stems (Fig- Luong et al., 2021).These different growth forms may result in substantially differential effects of fire (Zomer & Ramsay, 2021), herbivory (Díaz et al., 2007), and drought (Luo et al., 2018).Further research into the generality of these effects within and across herbaceous growth forms will be important for determining how many and which PFTs to represent in model simulations.
A major challenge in representing herbaceous vegetation is that allometric equations between stem diameter and plant canopy or total leaf area often do not apply for forb and grass species (Johnson et al., 1988).Recently, work has been undertaken to relate basal crown area of caespitose bunchgrasses to canopy area, with relative success (X.Gao unpublished data).This is different from trees in that crown area of bunchgrasses is composed of many individual stems, which are packed into the crown in particular ways (Curasi et al., 2023).Using this allometric relationship to scale to canopy area and ultimately total leaf area also means that the model would use crown area as a proxy for stem diameter, and this introduces new challenges such as incorporating relationships of crown area with water and nutrient flow.As such, additional allometric relationships such as crown area to total stem area could be particularly useful to optimize both the canopy formation and hydraulic components of modeling grasses (Table 2).
Allometric relationships have previously been developed for herbaceous ecosystems, and these techniques could be leveraged to aid in better representation of different herbaceous growth forms.
For  Reproduction strategy is another important consideration when modeling herbaceous ecosystems.Most models represent reproduction via seed, which ignores some key processes that can drive herbaceous community dynamics.For example, rhizomatous species, such as those in our simulations above, typically reproduce close to the parent plant (Ott & Hartnett, 2015), which limits the rate of spread in a single year.Contrasted with annual reproduction strategies, clonal growth by rhizomatous species results in slower rates of displacement of one species or functional group by another.
The lack of this reproduction strategy within most vegetation models may be why our simulated community dynamics above (Figure 2) were far more rapid than observations.Also, less allocation to seed production and more to belowground carbon stores may be driving observations of herbaceous systems being able to persist during stressful conditions and to recover quickly after perturbations (Stuart-Haëntjens et al., 2018).
Belowground rooting morphology varies across species and herbaceous functional groups (Gibbens & Lenz, 2001;Weaver, 1958).Many grasses have fibrous roots occupying much of the top 30 cm of the soil profile (Jackson et al., 1996), while many forb species have branching roots that are less abundant in surface soil layers but extend deeper into the soil column (Sun et al., 1997).This has been shown to be a coexistence mechanism where grasses utilize shallow water sources during wet periods and forbs utilize deeper water sources during dry periods (Nippert & Knapp, 2007a;Ward et al., 2013).These dynamics similarly apply to interactions between woody and herbaceous plant groups (February et al., 2013;Scholes & Archer, 1997).There are also important differences in rooting morphology within functional groups.For example, rhizomatous grasses tend to have a coarse-grained foraging approach with a low density of larger roots, while caespitose grasses have been shown to have a high density of fine roots and can concentrate nutrients in the soils beneath their canopy (Derner & Briske, 1999;Weaver, 1954).This may also link to increased microbial biomass and activity under bunchgrasses (Vinton & Burke, 1995), which has implications for carbon respiration rates.
In accordance with previous calls for using morphological and structural characteristics to differentiate plant functional types within vegetation models (Wullschleger et al., 2014), we suggest starting by differentiating among caespitose/tussock-forming, matforming, and single-stem herbaceous vegetation groups.From there, these groups could be further differentiated where appropriate based on the research questions and region of inquiry.Additional splitting criteria could be by height (e.g., tall vs. short grasses-these are distinctly different plant groups in many grasslands), photosynthetic pathway (C 3 , C 4 , CAM), and reproductive strategy (e.g., annual, perennial).c Column describes the most relevant model processes that the measurements may inform.The entries in this column match up with sections below: sg = Structural growth of herbaceous plants, w = water uptake and competition, n = nutrient cycling and competition, vd = vegetation dispersal and spatial dynamics, d = fire and grazing regimes for herbaceous systems, b = benchmarking.Note that benchmarking is not a section below, but it is relevant for checking model simulations related to all the above ecological processes.
TA B L E 2 (Continued) Although we maintain that better representation of herbaceous PFTs is important for improving models, we also recognize the need for parsimony in assigning PFTs (Fisher & Koven, 2020).
After all, representing every plant species on Earth is not computationally feasible.As such, a priority for future research is assessing how functionally similar or different various groupings of species are to one another so decisions can be made as to how many PFTs are needed to parsimoniously represent herbaceous ecosystems.Interesting research is currently underway focused on using plant lineages (Anderegg et al., 2022;Griffith et al., 2020;Lehmann et al., 2019), eco-evolutionary competition tradeoff strategies (Harrison et al., 2021), and gradients of functional traits (Butler et al., 2017)

| Water uptake and competition for water
Models represent water dynamics and competition for water by simulating both biotic and physical processes.Water enters a system via rainfall and horizontal flow, and water exits the system via plant uptake and subsequent evapotranspiration as well as evaporation from the soil, overland flow, and leaching losses.Plant water stress is determined by a plant's ability to take up soil water and their rate of loss, which is controlled by water availability combined with stomatal conductance and associated parameters.
Ultimately, plant growth is facilitated or hindered by water availability and cellular turgor pressure (Boyer & Potter, 1973;Hilty et al., 2021).
Water transport to various soil layers, overland runoff, and leaching losses are simulated within models using water physics components that incorporate information about soil texture (sometimes represented implicitly via field capacity and wilting point), soil moisture from previous time steps, and rates of precipitation inputs.
Water uptake by plants is often resolved separately by soil layer and is a function of the amount of water in the soil layer and the quantity of roots in that soil layer.Total root production is a function of the amount of photosynthesis plants conduct, their maintenance costs, and their allocation strategies.Then, root proportion/ density among different soil layers is determined using a parameter within an asymptotic relationship between cumulative root fraction and soil depth (Jackson et al., 1996), which can vary across PFTs in most DGVMs and VDMs (Lawrence et al., 2019).Recent efforts have focused on enhancing representation of hydrodynamics in process-based models (Christoffersen et al., 2016;Eller et al., 2020;Liu et al., 2021).These schematics typically use water potential gradients across the soil, plant organs, and the atmosphere along with hydraulic conductivity and vegetation parameters associated with plant ability to tolerate water stress (e.g., leaf osmotic potential at 50% conductivity (P 50 ; Li et al., 2021).
Empirical evidence has shown that herbaceous plants utilize three primary strategies to contend with water limitation-drought tolerance, avoidance, and escape (Levitt, 1980;Ludlow, 1989;Ocheltree et al., 2020;Noy-Meir, 1973).Drought tolerance is the continued performance of plants despite water stress.Plants use multiple structural mechanisms to accomplish this, including having thick cell walls to regulate turgor pressure and avoid hydraulic failure (Marshall & Dumbroff, 1999;Ocheltree et al., 2016), regulating osmotic potential (Zhou & Yu, 2010), and stomatal control (McDowell et al., 2008).Often, plant traits such as P 50 and leaf dry matter content (LDMC) are linked with drought resistance of grasses as well as phenology (Ocheltree et al., 2020).Fortunately, these strategies and parameters are becoming better represented within hydraulics schematics in models (Christoffersen et al., 2016;Eller et al., 2020).This means that empirical information directly describing herbaceous hydrodynamics, such as turgor loss point, leaf capacitance, and P 50 can be easily incorporated into current versions of plant hydraulic schematics (Table 2).Incorporation of traits such as LDMC into models is more difficult because LDMC is only indirectly related to hydraulic characteristics and processes, being the ratio of the dry leaf mass to fully hydrated leaf mass.Yet LDMC integrates across a few important plant characteristics, such as the rigidity of cell walls, which influences turgor loss point.As such, incorporating LDMC into model hydraulic schematics could be an area for further inquiry.
Drought avoidance is when plants can access sources of water that allow for growth under drier conditions.A prime example of this drought response mechanism is when deeply rooted species persist under dry conditions while surrounding shallow fibrous-rooted grasses decline, which has been shown in a wide variety of ecosystems (Aroca et al., 2012;Báez et al., 2013;Nippert & Knapp, 2007b).
Empirical findings have shown that water uptake from various locations in the soil is an important coexistence mechanism between deeply rooted (e.g., shrubs, forbs) and shallowly rooted plants (e.g., grasses; Ward et al., 2013).Nippert and Knapp (2007a) showed evidence that shrubs use deeper soil moisture only during times of water stress, which may necessitate modification of the simple schematic of water uptake in models that simulates water uptake as a function of soil moisture and root density by soil layer.However, additional empirical understanding should be gained about the generality of this depth switching phenomenon and moisture thresholds of when this occurs.
Drought escape is when plant growth occurs during times of greater water availability despite overall drier conditions, as with desert annuals (Venable, 2007).This process helps explains a longstanding conundrum associated with the historic drought in the United States during the 1920s (i.e., dustbowl; Schubert et al., 2004) where cool season grasses increased in abundance after the drought due to early-season growth strategies (Knapp et al., 2020).This community change was likely a combination of direct impacts on cool season species since most of the water availability was during the primary growth period of these species, and competitive release by warm season species that were strongly water limited during their growth periods.Another example of drought escape is seed dormancy until environmental conditions support plant growth, which is often rapid (Chesson, 2000).
To move forward in modeling water dynamics for herbaceous systems (Hickler et al., 2006), we propose that hydraulic information be gathered (much of it already exists in the literature) for many species spanning different herbaceous growth forms, photosynthetic pathways, and reproductive strategies.The hydraulic infrastructure is rapidly being developed within many models (Chitra-Tarak et al., 2021;Powell et al., 2013), but this has primarily been for trees.
Understanding how rooting distributions (density by depth), hydraulic conductance, stomatal conductance, and other hydraulic traits (Table 2) vary both within and across plant forms would greatly aid in developing reasonable and representative groupings for herbaceous species.Additionally, to improve the representation of drought escape, we suggest that ecophysiological information, such as stomatal conductance, could be linked with soil moisture, leaf growth, and turnover at relatively high frequency, especially during the early growing season when many cool season species complete their growth cycles (Table 2).To better represent annual species flushes in arid landscapes, thresholds for germination related to rainfall (e.g., Chávez et al., 2019), soil moisture, and temperature could be incorporated into plant growth components of models.Also, empirical information describing the respiration/maintenance costs during reproduction of annual plants could help models better represent rapid senescence events of annual-dominated ecosystems.
Improving representation of water dynamics in herbaceous ecosystems should be a high priority due to increasing frequency of precipitation extremes, such as the recent extreme droughts in the southwestern United States, large portions of Europe, and Kenya, and deluges occurring in the western United States.Additionally, climate model projections suggest that the current frequency of precipitation extremes will increase by ca.30% by 2100 under medium emissions pathways (Thackeray et al., 2022) so predicting ecosystem function during and after these events will be critical for assessing future states of the earth system.

| Nutrient cycling and competition for nutrients
Incorporation of both nutrient supply and demand is key to representing nutrient constraints on ecosystem processes (Walker et al., 2015) and to allow for more flexible/dynamic ecosystem responses to nutrient availability.There are multiple ways in which different models represent nutrient dynamics.One approach is to use mass balance equations and stoichiometric information related to various soil and biotic pools to estimate nutrient pools and fluxes, such as the multiple element limitation model (Rastetter et al., 2022).The Plant Allocation and Reactive Transport Extensible Hypotheses (PARTEH; Knox et al., 2023) can model allocation of nutrients across different plant organs, building on plant nutrient demand, soil nutrient supply, storage, acquisition, and tissue stoichiometry (ideal and minimal) for carbon, nitrogen, and phosphorus.In PARTEH, these within-plant nutrient dynamics interact with a soil biogeochemical module for nutrient acquisition to model nutrient uptake competition between plants and microbes (Zaehle et al., 2014).Yet these processes are often resolved on a coarse scale which precludes heterogeneity in nutrient availability within the unit (e.g., grid cell) where competition occurs.
In CLM 5.0, there are multiple key parameters that control the uptake of nutrients, including a set of parameters controlling the energetic costs associated with mycorrhizal and non-mycorrhizal nutrient uptake (Brzostek et al., 2014), the fractional productivity of N-fixing plants, and the fine root mass per unit leaf mass (Lawrence et al., 2019).The interactive effects of nutrients (e.g., nitrogen) and other soil parameters are not represented in many models (e.g., N effects on soil pH, interactions among N, P, K, and micronutrients; Du et al., 2020).
Increasing nutrient availability generally increases primary productivity in many terrestrial ecosystems (Elser et al., 2007;Fay et al., 2015;Gruner et al., 2008), but typically reduces or eradicates certain species and groups of species (Avolio et al., 2014;Hillebrand et al., 2018;Komatsu et al., 2019;Yang et al., 2011).These effects have been demonstrated with both experimental nutrient additions (Harpole et al., 2016;Ladouceur et al., 2022;Seabloom et al., 2021) and anthropogenic N deposition (Bobbink et al., 2010;Simkin et al., 2016;Stevens et al., 2004Stevens et al., , 2015)).Multiple competitive dynamics among plant species may be at play in driving these responses to altered nutrient availability (Cleland & Harpole, 2010).Debate remains regarding the relative strength of the underlying mechanisms of competition with nutrient additions between the shifting of limiting factors to aboveground competition for light (DeMalach et al., 2017;Hautier et al., 2009, but see Eskelinen et al., 2022;Harpole et al., 2017) compared to reduced niche dimensionality belowground (Harpole & Tilman, 2007;Harpole et al., 2016).Furthermore, altered stoichiometric balance and plasticity of plants with increased nutrient availability can drive species and community responses (Tilman, 1981;Yu et al., 2011Yu et al., , 2015)).For example, Yu et al. (2015) showed that plants with less plastic nutrient contents in their tissues tended to be more dominant in natural grasslands, but species with greater plasticity became dominant in nutrient-enriched grasslands.
Although these findings show important patterns of nitrogen controls on herbaceous plant communities, the first principles driving these responses are still largely unknown.
Beyond direct competition, other biotic interactions can affect plant community responses to altered nutrient availability, including mutualisms with mycorrhizae and N-fixing bacteria (Tognetti et al., 2021).We know that associations with mycorrhizae are critical for many species' ability to take up nutrients (Van Der Heijden et al., 2008) and water (Smith & Read, 2008), and can drive coexistence in herbaceous communities (Stanescu & Maherali, 2017).Additionally, mycorrhizal colonization rates vary substantially among species (Wen et al., 2019) and functional groups (Unger et al., 2016).
Plants that associate with N-fixing bacteria may be positively impacted by P additions (Hungria & Vargas, 2000), but negatively impacted by N additions (Tognetti et al., 2021).These responses likely vary across species that range in N-fixing traits from facultative to obligate mutualists (Taylor et al., 2020).Other impacts on plant groups that associate with N-fixing bacteria include N-fixation limitation by drought (Serraj et al., 1999), although this effect may be alleviated by increased atmospheric CO 2 (Parvin et al., 2019).Fire has also been found to have positive effects on plants associated with N-fixing bacteria (Kerns & Day, 2018).
There are numerous avenues for advancement of nutrient competition and dynamics in vegetation models.One particularly important component for many herbaceous ecosystems is to increase our understanding of the drivers of N fixer abundance in the community.
Sensitivity analyses show that fertilization effects within models are highly responsive to the proportion of N fixers within the community (Fisher et al., 2019), making this a high priority for research.
Second, modeling work has shown that competition among PFTs based on root surface area creates a tradeoff between above-and belowground competition for light and nutrients (Weng et al., 2017(Weng et al., , 2019)), which should be explored further empirically.Third, associations with mycorrhizae and the variation in such associations could be better described for different PFTs-as such, information about mechanisms driving mycorrhizal association to increase or decrease, and implementation of these mechanisms within the nutrient cycle representations within models, would assist in providing better representations of eutrophication effects on ecosystems.Lastly, gathering information about aboveground-belowground tradeoffs among different herbaceous groups, and their ability to alter their allocations above-versus belowground, would allow for better parameterization of herbaceous PFTs in their acquisition of nutrients.

| Vegetation dispersal and spatial dynamics
As discussed in depth above, most vegetation models represent reproduction via seed.These models typically do this by allocating a set proportion of excess carbon (assimilated during the growing season) to seed (e.g., Moorcroft et al., 2001).These seeds enter the system and drive germination rates the following season.when resident vegetation abundance is decreased (Sato & Ise, 2012).
Spatial dynamics, such as dispersal within a metapopulation, have been shown to allow for species to exist as "sink" populations within communities despite being less competitive in the local environment (Hanski, 1982;Leibold et al., 2004;Shmida & Wilson, 1985).
Reproductive strategies of different species and functional groups are particularly important when considering how plants will respond to stressful conditions.For example, species with strategies that allow for further dispersal (e.g., wind dispersed small seeded species) and fast growth rates, may allow for faster expansion into disturbed areas (Monty et al., 2013).Additionally, germination rates of some species have particular conditions that lead to substantial increases in primary production.For example, an extreme drought occurred in a US shortgrass prairie in 2012.Prior to the drought, an annual cool season grass (Vulpia octoflora) was a relatively rare species in the community.However, the year after the drought, it became the dominant vegetation in unburned areas, and stayed dominant for many years after the drought (Dufek et al., 2018).
Investigating the relative importance within models of longdistance propagule dispersal on herbaceous community dynamics would be useful to assess whether seed dispersal across grid cells is an important driver that should be included, given its complexity to simulate (Nabel, 2015).Shorter distance dispersal representation within models would benefit from collection of information  (Wang & Loreau, 2016;Wilcox et al., 2017), yet global change drivers and disturbance can reduce this asynchrony effect (Hautier et al., 2020).Additional global change studies that bridge spatial scales will be important to answer this question.

| Fire and grazing regimes for herbaceous systems
The SPITFIRE (Spread and InTensity of FIRE; Thonicke et al., 2010) model has been incorporated into several vegetation models, including LPJ-DGVM (Thonicke et al., 2010), ORCHIDEE (Yu et al., 2015), and FATES.The model represents two major components, the probability of ignition and the rate of fire spread.The probability of ignition is based on the amount and moisture content of the vegetation, which is obtained from the vegetation model.There is also differentiation between natural and anthropomorphic sources of ignition.Human-caused ignition rates are a unimodal function of  (Thonicke et al., 2010), which typically do not follow the population size-burn frequency relationships set up for anthropogenic fire (Allen & Palmer, 2011;Hulbert, 1986).
Herbivory is not represented in most process-based vegetation models.One model that has represented grazing is the G-Range model (Boone et al., 2018).In this model, there are two herbaceous plant functional types, perennial and annual.A given total amount (determined for each grid cell) of aboveground plant biomass is removed, and is divided among PFTs such that herbaceous vegetation loss is relatively high compared with shrub and woody PFTs.Loss of plant C and N is proportional to the amount of biomass removed via grazing.Nitrogen is returned as feces and urine, and the proportion of nitrogen removed is a function of soil texture.One consideration missing from this representation of grazing effects is representation of herbaceous PFTs that differ in their susceptibility and response to grazing.
Disturbance has long been identified as a key driver of numerous ecological processes in a wide variety of landscapes (Newman, 2019;Turner, 2010), and has been found to have particularly strong effects in herbaceous ecosystems (Belsky, 1992).For example, frequent fire in tallgrass prairie has been shown to cause dominance by highly productive C 4 rhizomatous grasses (Collins & Calabrese, 2012), while grazing in the same system reduces these same grass species and negates the positive fire effect (Collins, 1998).Interactions with herbivores (i.e., apparent competition) and/or pathogens (Gruner et al., 2008;Hillebrand et al., 2007) are other important determinants of herbaceous community structure.Disturbance effects also interact with water availability.For example, in a South African savanna, herbivory resistance of a dominant bunchgrass, Bothriochloa radicans, due to high concentrations of terpenoids within leaves, provided structure necessary to support many other species despite herbivore pressure (Koerner et al., 2014).However, energetic costs associated with this herbivore resistance caused the species to experience mass mortality during drought (Wilcox, Komatsu, et al., 2020), effectively altering the future responses of the entire ecosystem to both precipitation and herbivore pressure.
Disturbance is a critical component of maintaining mixed assemblages of herbaceous and woody plant species, such as savannas (Lehmann et al., 2014;Staver & Bond, 2014).In more arid savannas, tree-grass coexistence is maintained by the ability of grasses to rapidly take up soil water, especially at the seedling stage (Belovitch et al., 2023).This effectively limits the ability of trees to proliferate and outcompete grasses for light (Holdo & Nippert, 2023).In more mesic savannas, coexistence is maintained by disturbance, where fire and herbivory keep tree canopies from closing and excluding grasses through light competition (Sankaran et al., 2004;Staver et al., 2009).
We also know that these types of disturbance may interact such that fire effects may be different in the presence of herbivory, and vice versa.For example, herbivores that eat herbaceous vegetation (i.e., grazers) reduce fuel loads through consumption, which ultimately reduces fire intensity and can cause tree encroachment if grazing intensity is too high (Ward, 2005).Conversely, the occurrence of fire often increases herbivore pressure due to short-term increases in nutrients after fire (Ojima et al., 1994) that are incorporated into plant tissues, thereby increasing forage quality.These effects may also vary depending on the plant growth form or PFT under consideration (Gao & Schwilk, 2018, 2022).
Some primary characteristics of plant species found to promote existence and spread under fire include annual life strategies (Midgley, 1996;Monty et al., 2013), dense or deep crown tissue (Wigley et al., 2020), location of reproductive buds and meristematic tissue (Clarke et al., 2013;Medwecka-Kornaś & Kornaś, 1985), and allocation of resources to rhizomes and other belowground storage organs (Clarke et al., 2013).Similar plant traits have been found to link to grazing resistance, but additional traits related to herbivore selectivity have been found to be important, including lignin content (Semmartin & Ghersa, 2006), and the presence of defensive compounds in plant tissue (Briske, 1996;Kaul & Vats, 1998;Maestre et al., 2022).
Disturbance-related traits may map neatly onto some of the plant groups outlined above (Figure 3; Hempson et al., 2015), and this may be a productive direction for future research.
To improve the representation of disturbances such as herbivory and fire into herbaceous components of models, information about herbivory and fire-related traits should be assessed within and among different plant growth forms and other splitting categories (e.g., annual vs. perennial, C 3 vs.C 4 ).When the goal is to represent susceptibility and responses to herbivory, it is likely that additional splitting of PFTs will be necessary due to variation of grazing tolerance and avoidance linked with different evolutionary histories and environmental characteristics (Adler et al., 2004).Lastly, better representation of anthropogenic fire ignition events will be important for areas where fire is used as a management tool.As one of the primary drivers of herbaceous vegetation in many ecosystems worldwide, improvements of the occurrence and effects of disturbance should be high priority for development in models if the goal is to represent ecological processes associated with herbaceous vegetation.

| CON CLUS IONS
The literature review and simulation exercises presented above highlight the current capacities and limitations of ecological knowledge for incorporating herbaceous plant communities and ecosystems into process-based models.Importantly, the need for improvement lies in both the empirical and modeling sides.Co-design and implementation of ecological studies with both empiricists and modelers is the ideal path forward for improving herbaceous dynamics within ecosystem models (Lu et al., 2011(Lu et al., , 2021)).We suggest this process should be iterative, beginning with conversations between modelers and empiricists about the goals of modeling activities and the ecological components and processes of interest (Kyker-Snowman et al., 2022).These conversations may be driven by observed patterns of ecological phenomena or other prior empirical understanding.Then, initial model simulations should be used to identify parameters or modeled processes and their ecological analogs where uncertainty exists.Empirical studies can then be designed to address these uncertainties and provide information necessary for improving model parameters and processes.Finally, using these improved models, empirical findings may be expanded to greater temporal and spatial scales, which is important for informing land management and policy decisions (Cuddington et al., 2013), as well as providing paths forward for future scientific inquiry.We contend that each iteration of this cycle will lead not only to developments in process-based models, but also to substantial improvements in our mechanistic understanding of ecological processes.Observation networks such as FLUXNET (Swenson & Lawrence, 2014) and AmeriFlux (Kucharik et al., 2006;Novick et al., 2018) have been instrumental in benchmarking and improving predictive capacity of many process-based models.We suggest that other experimental or observational networks, of which there are currently many within herbaceous ecosystem types (e.g., NutNet, DroughtNet, HerbVar, NEON data streams, BromeCast), be leveraged to improve herbaceous representation within models.The franchise nature of many of these networks could be leveraged to allow a variety of researchers to opt-in and collect standardized measurements that could then be incorporated into models (Table 2).All these efforts are important to improve our predictive capacity for herbaceous ecosystems, which will in turn reduce uncertainty of predictions of future states of the entire earth system.Ultimately, our goal should be to increase our ability to make robust predictions of future states of Earth's ecosystems, which will be vital for guiding policy and management over the coming decades.
terized by unique sets of values for large parameter sets that allow different PFTs to independently respond to environmental drivers and compete for light and soil resources in different ways from one another.DGVMs do not simulate individual plants, but there is a range of model complexity in representation of plant cover, including F I G U R E 1 Conceptualized changes in environmental variables and plant community composition under two different components of global change: (a) increased frequency of extreme events, and (b) monotonic changes in the environment.(c) Hypothesized and observed relationships between changes in plant communities and their effects on ecosystem function responses to environmental factors.Different scenarios are based on hypotheses and findings from the literature: I no change in function despite change in community-Wilcox, Koerner, et al., 2020; Wilcox, Komatsu, et al., 2020; II linear change in function with altered community-Wilcox et al. (2016), III threshold change in function-Briske et al. (2006), Smith et al. (2009); IV unimodal or oscillating change in function-Langley and Hungate (2014).Changes in plant communities and ecosystem function will likely vary widely depending on site-level characteristics, the type of existing vegetation present, and the ecosystem functions being assessed."big leaf" models where the entire canopy is represented as a single leaf, mosaicked plant cover to roughly capture coexistence of PFTs in the same grid cell, multi-layer canopies of single PFTs, and most recently communities of vertically stratified mixed PFTs to allow relative cover of PFTs or mixed community structure to shift in response to environmental drivers.When linked to ESMs, which couple land, ocean, and an atmospheric general circulation model (GCM), DGVMs can simulate feedbacks between the land surface and climate and serve as a component in the full carbon cycle.Such coupling can allow investigation of such questions as long-term vegetation-climate feedbacks under climate change, the impacts of episodic climate extremes on vegetation dynamics, as well as how vegetation change alters ecosystem direction and magnitude of plant compositional shifts observed in the short term (years 1-10) or long term (years 11-20).The criteria for whether models were able to simulate plant compositional dynamics were as follows: Direction-did the model estimate dominance by the correct species found in observations (years 1-10: A. gerardii, years 11-20 P. virgatum)?Magnitude-did the model represent the magnitude of PFT change relative to starting conditions?-this would result in coexistence by both species.The three models with capacity to simulate composition shifts (i.e., SDGVM, BiomeE, LPJ-GUESS) matched observed trends in at least one of these areas, yet no models matched observed trends in all categories (Figure 2a,b).BiomeE was able to match observed trends in the short term but struggled to capture the long-term community dynamics.The model matched observations for the first 10 years of the experiment (i.e., greater abundance of A. gerardii than P. virgatum), but potentially for the wrong reasons.Whereas the model simulated dominance of A. gerardii PFT in the short term, it was not due to slow rates of displacement of A. gerardii, but because parameters of the P. virgatum only the model agnostic to community change (TECO) matched short-term ANPP simulations to observed values, yet predictably the model was unable to simulate the increased ANPP trend observed in years 11-20 (Figure 3c,d).Similarly, BiomeE under-projected ANPP in the long term, likely due to dominance of the A. gerardii PFT from years 11 to -20 in its simulations (Figure 3a).Conversely, because SDGVM and LPJ-GUESS successfully simulated the shift to F I G U R E 2 Quantitative assessment comparing observations from irrigated plots within a long-term irrigation experiment in tallgrass prairie to model simulations from a variety of process-based models.(a) Observed relative abundance of two C4 grass species (thick lines) is compared with model PFTs (symbols and lines) designed to represent these two species.(b) Authors' interpretation of the ability of each model to capture the directionality (dir) and magnitude (mag) of community shifts in the short term (1991-1999) and long term (2000-2012).Only models that simulate PFTs are included in (a) and (b).(c) Aboveground net primary productivity (ANPP) observations (grey bars) and model simulations (symbols, see panel c for legend of symbol shapes) over the course of the irrigation experiment.(d) Average simulated ANPP over the short term (1991-1999) and long term (2000-2012) of irrigated grassland.Horizontal lines represent the mean (solid) and 1 SD intervals (dashed) of the observed ANPP during each time period.Error bars on symbols represent 1 SD of the simulations through time.dominance by P. virgatum in years 11-20, they were better at capturing long-term ANPP dynamics (Figure 3c,d).Yet, both SDGVM and LPJ-GUESS over-projected ANPP during year 1-10 because the simulated PFT shift occurred within 1-2 years, which failed to capture the inertia of the current PFT.This simulation exercise teaches us two major things with respect to modeling herbaceous community changes and ecosystem responses under global change scenarios.First, the exercise highlights the importance of capturing herbaceous plant community change to accurately represent temporal trends of ecosystem function (sensuKnapp et al., 2012), especially in systems that have dynamic plant communities.Second, to capture both increased variability and monotonic global change (Figure1), we must both improve our empirical understanding of the mechanisms driving plant community shifts and incorporate this understanding into representations of herbaceous vegetation in models.It is not surprising that the models were unable to simulate all aspects of the community and ANPP patterns shown here, since we still do not fully understand the mechanisms responsible for these empirical patterns.Enhancing this understanding should be a primary goal for empiricists and modelers alike to better represent herbaceous ecological processes in models.

(
10 years) replacement of the rhizomatous C 4 perennial grass A. gerardii by the other C 4 perennial grass P. virgatum was likely limited by the seed-based reproduction scheme in vegetation models.So, although incorporating multiple C 4 perennial rhizomatous PFTs may not be feasible or advisable for researchers focused on, for example, global estimates of carbon sequestration under altered water availability, if the functionality of slow spreading clonal species is readily available in VDMs, this functionality can be leveraged for more sitelevel research questions.

4. 1 |
Structural growth and reproduction of herbaceous plantsIn vegetation models, herbaceous plants may form part of a closed understory or canopy, or open and undifferentiated canopy.For plant growth, most models use allometric relationships between stem diameter, leaf mass, and canopy area to convert leaf carbon to total leaf area and individual canopy area to the community canopy cover.These derived variables are then important drivers of various other model processes.
canopies are also important for herbaceous layers.To this end, we showcase a 3-D model visualizing empirical measurements of basal and canopy diameters, vegetative height, and volumetric density for three bunchgrass species within a 4.5 × 4.5 m plot in a South African savanna (Figure3bottom).
Success rates and energy costs of different types of reproductionmeasurements of seed and ramet characteristics Used to determine the rate of increase in PFTs.Also, this would be important in future developments of dispersal limitations for PFT spatial spread.mass/storage in the rhizome-pool of carbon and turnover rate measured using non-structural carbohydrate assays Benchmark non-structural carbohydrate storage for different plant functional groups allocation to defense compounds Not yet implemented parameters that reduce mortality or leaf and stem loss due to herbivory Need column indicates general importance of the data type for improving representation of herbaceous processes within models.* indicates lesser need, ** moderate need, *** greatest need.b Ease column indicates the difficulty of collecting the suggested measurements with * indicating difficult, time consuming, and/or expensive measurements, ** moderately difficult, time consuming, or expensive measurements, and *** indicating easy, quick, and/or inexpensive measurements.
Representation of spatial patterns (e.g., across grid cells) of seed dispersal is largely not accounted for within vegetation models.Two modeling studies have focused on this issue.First, Blanco et al. (2014) created a version of aDGVM (2D-aDGVM) that allowed for plant dispersal and found that environmental heterogeneity and dispersal of trees (dispersal of herbaceous vegetation was not represented in the model) were important for simulating tree-grass coexistence.Another study used a spatially explicit individual-based dynamic global vegetation model (SEIB-DGVM; Sato & Ise, 2012) to show how dispersal dynamics, again for trees, alter temporal vegetation trajectories, especially

TA B L E 1 Descriptions of models used in the simulation exercise. Model type Model name Herbaceous PFT (base) PFT parameter complexity a Key processes Plant competition Best use in herbaceous systems
https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950by Cornell University Library, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License , 2023, 23, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950by Cornell University Library, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Empirical needs for better incorporation of community shifts into process-based models.Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950by Cornell University Library, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License (Eskelinen et al., 2022)ion into models is how they relate to canopy cover and total leaf area since this will impact how these different groups influence and respond to light.The need for multi-layer canopies has been shown for forests(Bonan et al., 2021), yet we contend based on empirical evidence showing strong light competition among individual species(Eskelinen et al., 2022), that multi-layer 13652486TA B L E 2 13652486, 2023, 23, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950by Cornell University Library, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License to explore the best ways to represent PFTs within models.Continued work on this frontier should provide in- 13652486, 2023, 23, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950by Cornell University Library, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 13652486, 2023, 23, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950by Cornell University Library, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License (Thonicke et al., 2010)nloaded from https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950byCornellUniversityLibrary, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)onWileyOnlineLibrary for rules of use; OA articles are governed by the applicable Creative Commons Licensehuman population density; this function describes an initial increase in ignition rate with population due to accidental ignitions, and then a decline at larger population sizes due to human effects such as land fragmentation.Rate of fire spread is a function of numerous variables including wind speed, the dryness of dead vegetation, and the surface to volume ratio of vegetation(Thonicke et al., 2010).A 13652486, 2023, 23, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gcb.16950by Cornell University Library, Wiley Online Library on [12/04/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License