This meeting was a continuation of the SCOPE tree-grass modelling group which bought together data holders with data from a wide variety of tree-grass (savanna) systems, and ecosystem modellers with models capable of handling both tree and grass components and the interactions between them and their environment. It was the first of two workshops to be held at NCEAS, Santa Barbara, USA. The goal of this workshop was to identify what kind of product(s) the group wished to produce, the format of the data, the nature of the data analysis and modelling experiments and how this will be carried out.

NCEAS has provided a web site for general, public access, information about the project and a password protected area onto which the group can place and access the common data sets. Once the group has finished their analyses, the data sets and results will also become public access. NCEAS normally maintains these sites for two years, after this the data could be stored on the Oak Ridge DAAC site.

This exercise should help to focus on what kind of science we need to improve our understanding of tree-grass interactions. We need to:

• identify what kind of data is available
• test theories to explain the data
• identify what regulates tree-grass interactions
• investigate the ability of the models to capture this information and use the models to investigate the mechanisms involved
• identify useful synthesis products

Steve Archer, Texas A & M University, USA

David Breshears, Los Alamos National Laboratory, USA

Mike Cougenhour, NREL, Colorado State University, USA

Mike Dodd, Ruakura Research Centre, New Zealand

David Hall, King’s College London, UK

Niall Hanan, University of California, Santa Barbara, USA

Jo House, King’s College London, UK

Richard Joffre, CEFE-CNRS, France

John Ludwig, CSIRO Wildlife and Ecology, Australia

Ruben Montes, Universidad Simon Bolivar, Venezuela

Charles Moyo, Matopos Research Station, Zimbabwe

William Parton, NREL, Colorado State University, USA

Jose San Jose, CIET, IVIC, Venezuela

Joe Scanlan, Robert Wicks Research Centre, Australia

Robert Scholes, Forestek, CSIR, South Africa

Jonathan Scurlock, Oak Rdige National Laboratory, USA

Bruce Thurrold, Ruakura Research Centre, New Zealand

Apologies:

Jaques Gignoux, ENS, France

Xavier Le Roux, INRA, France

Jean-Clude Menaut, Orstom, France

During the first day of the meeting we had presentations from all of the participants who have tree-grass savanna data sets. The presentations ranged from 15-30 minutes and described the type of data collected at the sites and theoretical tree-grass competition issues that could be addressed using the data sets. A detailed description of the presentations for each data set will be attached to this report later, however, Table 1 summarizes information about the sites where tree-grass competition data sets are available.

Table 1. The models available for the synthesis exercise

*MUlti-stratum Spatially Explicit model

The models included in this exercise represent a gradation of complexity and scope. GRASP is a pasture growth model driven by transpiration at a daily time step where the trees are regarded only in terms of their competition with grasses for resources and allocation is weighted according to demand. CENTURY-SAVANNA is a biochemically based model operating at a monthly time step (daily or annual also possible) where available nutrients (N) are the primary control on production and trees take what they need first according to tree basal area. SAVANNA models woody population dynamics and is driven by carbon assimilation at the plant scale with trees and grasses competing for water and nutrients based on demand, and for light, at a weekly time step. It is quasi spatial with different patches of vegetation tracked across heterogenic landscape patches. Each was designed for a different purpose and is likely to be appropriate for different purposes in this exercise. The MUSE model was unfortunately not represented in this meeting as none of the French participants was able to attend, but it will hopefully be included in future exercises, workshops and meetings. It is not currently intended to include more models in this exercise unless they can add a significant new perspective or dimension. Agroforestry models were discussed such as ICRAF’s tree grass model.

GRASP (Scanlan)

GRASP is a pasture production model where tree production, population dynamics (except Moore version) and spatial distribution are not simulated, but the tree does compete with grasses for resources. Growth and competition are driven by the water needs of the plants to meet their respective transpiration demands which are based on basal area x pan evaporation for trees and green cover x pan evaporation for grasses. Soil water availability is divided into layers, and trees and grasses are supplied with the same proportion of their demand according to water availability and plant demand in each layer which is based on root profiles. Nitrogen is taken up in the transpiration stream at a constant concentration (although a switch allows tree uptake to be limited). Moisture stress, frost or lack of nitrogen reduce grass transpiration demand, however, there are no environmental effects on trees. Measurements of the effects of trees on microclimate at a community level are being progressively implemented. Grazing and fire effects the grass layer, but not the trees (except Moore version).

CENTURY-SAVANNA (Parton)

The CENTURY-SAVANNA model represents the competition between trees and grasses by allowing for competition for soil water and nutrients. Grass plant growth can also be reduced when tree leaf area gets large enough to shade grass plants. The major mechanism for competition is the equation that controls nitrogen uptake by trees and grasses as a function of the available nitrogen content of soil and the tree basil area. The model grows trees and grasses using separate tree and grass submodels and then allocates nutrients depending on the tree basil area. The Century model also simulates the dynamics of soil organic matter and soil nutrients, and soil temperature and water using a monthly time step.

SAVANNA (Cougenhour)

SAVANNA looks at the role of landscape heterogeneity on ecosystem function and how water in the landscape effects this. It is quasi spatial in that it keeps track of patches in the landscape and the proportion of trees/herbs/shrubs, allowing for a clumping parameter. Water and nutrients are allocated according to demand in the different patches, and there is also competition for light. The model tracks woody population and individuals in different size classes. It calculates NPP according to carbon assimilation at an intermediate level of physiological detail. Population dynamics of grazers and browsers can be simulated, dietary selection and palatability can be included. Fire frequency and intensity scheduled, but could be stochastic if wish.

David Breshears also presented a four compartment water model based on Walter’s two-layer hypothesis. The model has two layers by depth and tree roots are split into deep-rooted or shallow rooted. There are also two horizontal zones with tree roots spreading into the inter-canopy as well as the below canopy zone. Although this is still simplistic it makes the predictions more realistic.

There are two core questions relating to how trees and grasses coexist in the same environment:

1. How are the resources apportioned between trees and grasses? [the NPP problem]
2. What controls the trends in time in the proportions of trees and grasses? [the coexistence problem]

These questions are related, but they operate at different time scales, the first being the static allocation of resources (water, nutrients and light) at a point in time, and the second being dynamic in relation to various control factors including disturbance (fire, grazing, flooding, etc.) as well as competition for resources. If savannas were an equilibrium system, (1) would lead to (2), but it seems more likely that savannas are in disequilibrium.

In answer to the first question, many conceptual models have been proposed to account for resource apportionment between trees and grasses. Some experiments done on individual sites give evidence for or against the above theories, but no synthesis of the situation in a large-enough number of sites has been carried out to prove of disprove the validity of these theories globally.

• Grasses get most of the water in the topsoil, while trees get all the water in the subsoil. Implicit in this assumption is that grasses have a higher water-use efficiency [the Walter hypothesis]
• Trees get resources in proportion to their canopy cover fraction [implicit in some models]
• Trees get resources in proportion to the area (volume) of soil they occupy, taking into account

• Trees get first bite of the cherry, once their demand is satisfied, grasses get the rest [CENTURY]

Different rules lead to different scenarios of how tree and grass NPP vary in relation to each other, and what the total NPP might be.

See Scholes Diagrams plotting tree NPP and grass NPP against % tree

• linear zero-sum game: The total NPP of a site is fixed, as tree NPP increases, grass NPP decreases in a linear fashion, and vice versa. This represents a straight trade off between trees and grasses and does not allow for interaction. There is little evidence for this in the literature, yet it is the assumption of many top-down NPP models.
• non-linear zero-sum game: The total NPP is still fixed, but the proportion of tree and grass NPP vary in a curvilinear fashion. The curvilinear variation is more often seen in the literature, but it seems unlikely that total NPP remains the same - however, there is little evidence in the literature to disprove this as most studies have just looked at the change in grasses with tree cover/biomass/NPP or the change in tree NPP with grass cover/biomass/NPP, and these studies almost invariably just included the above ground proportions.
• facilitation optimum: Trees and grasses facilitate each other due to changes in the microclimate, etc., and can take advantage of different resources spatially and temporally such that a combination of trees and grasses leads to a higher NPP than would be found with either trees or grasses alone.
• asymmetric optimum: Due to resource use efficiency, NPP of the system is higher with either trees or grasses. For example, as trees enter the system the NPP of the whole system might increase and be highest when there are more trees if, for example, the system is N limited (trees having a wider C:N ratio and thus being capable of more production), or if the system tends to have deep water that is beyond the use of the grasses. However, the asymmetry could also favour C₄ grasses under certain light use scenarios.

Theories regarding the controls of competition on trends in time lead to different equilibrium theories:

See Scholes Diagrams

• "Balanced competition": Intraspecific (tree-tree or grass-grass) competition is stronger than interspecific (tree-grass competition). This will always end at a stable point of tree-grass mixture for any given climate and soil - a climatic-edaphic equilibrium. Deviations from this point are a result of disturbance.
• "Competitive exclusion": Tree-grass competition is greater than tree-tree or grass-grass competition. Any point in the tree-grass mix is unstable, and the only stable points are no trees or no grasses.
• "Multiple stable states": This assumes that there can be several stable states depending on thresholds of disturbances such as grazing.
• Disequilibrium approach: Disturbance has a strong effect on tree-grass competition. Big trees out-compete grasses (for light, water and nutrients), but grass has a negative effect on small trees as it increases the likelihood of fires which kill small trees. Herbivory has a negative effect on grasses and small trees due to consumption, but it also has an indirect positive effect on little trees by removing the grass. The mixture of trees and grasses is more dynamic, but there are thresholds. For example, beyond a certain high fire and low grazing threshold large amounts of grass lead to regular hot fires destroying all seedlings and eventually leading to no trees. There is also an inter-tree limit, often caused by heavy grazing and no fire, where tree cover increases, grasses are suppressed and there is no fire leading to dense thickets.

In many savanna sites some degree of tree clumping takes place where there can be isolated tree clumps or larger patches of trees interspersed by open grassy areas. This adds further complexity to competition and coexistence theories. Conditions are often very different between the under-canopy and inter-canopy areas. In areas with very pure, distinct clumps there may be virtually no tree-grass competition taking place, with a more dispersed pattern, there can be all kinds of relationship.

These theories to explain resource apportionment and trends in time lead to a number of more detailed questions:

• What is the maximum tree basal area (NPP/LAI/etc.) under given climate and soil conditions?
• What is the NPP of trees and grasses in mixtures (zero-sum, optimum or)?
• What are the stability domains with respect to fire and herbivory within which savannas persist?
• How is resource allocation/competition affected by the clumping of trees?
• How important are spatial issues in predicting NPP?
• How will all of the above be affected by elevated CO₂ (direct stomatal effects and indirect climate change effects)?

These kind of savanna theories and issues have been highlighted before, but there has never been a large enough synthesis of a wide variety of data in order to answer them conclusively. This will be the first large-scale savanna data synthesis, and also the first savanna model synthesis. It should be possible to answer many basic questions using statistical analyses, these statistical analyses will show if any underlying patterns emerge. The models can be used as tools to answer more complex questions. It is interesting to look at many of these questions statically and dynamically. The former is possible with statistical analyses or models, the latter is only possible with models.

This group has identified a large amount of data sites available across a variety of topo-edaphic and management regimes and at various levels of detail. All of the sites have data on either tree biomass or grass biomass at least above ground, fewer sites have both tree and grass biomass recorded above and belowground. Some sites have data on the effects of different levels of trees on grasses, or grasses on trees due to different treatments or environmental gradients. Other sites have a range of basal areas for similar climate and soils. Some sites just have measurements at one point in time, but others have measurements for a number of years with climatic and seasonal differences. Several sites have patches of different vegetation structure due to climate, soils or disturbance, and where trees are highly clumped, these can be considered as unique patches compared to the surrounding vegetation. This range of data sets available can be viewed in a hierarchical structure:

Level 1 = Structural data

1a) at one point in time

1b) over a time period

Basic data such as site information (lat., long., elevation, slope, run-on/off, disturbance history), climate (monthly ppt, t_max, t_min, S.D. for annual ppt), soil (type, texture, depth, field capacity, total N, pH [CEC]), tree (dominant species, BA/cover^*, spatial pattern, % evergreen, % N-fixers), herbaceous biomass (dominant species, %annual/perennial, % grass/non-grass, % C₃/C₄), root distribution, etc. (See Appendix A)

Questions addressable:

• Can tree BA be predicted from climate/soils?
• Equilibrium theories
• Domain theories (where tree-grass mixtures, N-fixers, grazing and fires occur in PAN-PAM plane)
• Position of end-point envelopes
• Position of disturbance thresholds
• Niche separation

Includes the above plus NPP data^* (above and belowground and litter) for trees and grasses preferably over a range of soils/climate, over time, or under different treatments (including no tree) which can be considered separate patches of the same site. Data on tree clumping, under/outside canopy data.

Questions addressable:

• Can grass NPP be predicted as a function of tree NPP for different climate/soils/treatments?
• Tree-grass interaction impacts on grass NPP under/outside canopies.
• Does total system NPP follow facilitation optimum or asymmetry?
• How does clumping effect the above?

Data at a finer temporal scale (grass growth and turnover leaf/stem/root, live/dead), more detailed spatial data (landscape and plant patches), plant allometry (leaf/stem/root), litterfall and turnover, root live/dead by depth, plant chemistry (leaf N, lignin), physiological data, population data, water balance (transpiration rates, stomatal conductance, leaf water potential, sap flow, daily ppt, pan ET, VPD, max-min), PAR, soil moisture/nutrients by depth, fire frequency/temperature/severity, grazing/removals details. (See Appendix B)

Questions addressable:

• What mechanisms are needed to explain the observed relationships?
• Factors controlling tree-grass growth
• Effects of plant spatial distribution and demography
• Allocation of water and nutrients between plant types
• Spatial and temporal separation of resource allocation and production

Level 1: to be prepared by December 1998

South Africa x 20 - Scholes

Zimbabwe x 5 - Moya

Kalahari x 5 - Scholes

West Africa x 4 - Moyo

NATT (Australia) x 32 - Ludwig

(Other Australian Burroughs/Anderson x 40, John Carter x 4 - Ludwig)

USA x 12 - Breshears

Venezuela x 26 - San Jose

Level 2 (at least three treatments of basal area): to be prepared by September 1998

E. populina - Scanlan

A. aneura - Scanlan

P. radiata - Thurrold

Eucalypt - Dodd

Acacia - Dodd

Mopane - Scholes

A. mellifera - Scholes

Pinyon-Juniper - Breshears

Pordrosa - Breshears

Acacia - Archer

Level 3: to be prepared as and when possible

Nylsvley x 2 - Scholes

Tikitere - Thurrold

Boatman - Scanlan

Sahel - Hannan

Los Alamos - Breshears

Venezuela x 2 - Montes

Te Kulti x 2 - Dodd

The heart of the way in which GRASP simulates tree-grass interactions is the competition for water to meet their respective transpiration demands.

Potential tree transpiration is calculated from tree basal area and pan evaporation. Potential transpiration increases linearly with tree basal until tree basal area reaches 25 m² ha^-1 at which potential transpiration is equal to pan evaporation. Potential grass transpiration is the product of proportional green cover and pan evaporation. Potential soil evaporation is calculated from the proportion of ground not covered by litter and pan evaporation. If tree basal area and green cover are both high, potential evapotranspiration can reach a maximum of 1.3 times pan evaporation.

The soil cannot supply these demands except when soil moisture is high. The ability of each layer of the soil to supply the transpiration demand is calculated from supply functions from which is calculated a soil water index (0 to 1). The root profiles of trees are also used to estimate the amount of water use from each soil layer. Grass root profiles are embedded in the calculation of the supply functions. The demand is multiplied by the soil water index and trees and grasses are supplied with the same proportion of their transpiration demand.

Nitrogen is taken up in the transpiration stream at a constant concentration (expressed as kg/ha/100mm transpiration). GRASP has the ability to allocate proportionally less of the nitrogen to the trees than to the grasses. This is based on the idea that grass root systems are located in the surface soil layers in which most of the organic matter and therefore where nitrogen is located. This feature has not been widely tested as data sets upon which to parameterise this behaviour are very limited.

Green cover of grasses can be reduced by moisture stress, frost or lack of nitrogen. This decreases transpiration. However, there are no environmental effects on tree transpiration demand.

Apart from the model developed by Jocelyn Moore and others for mulga, tree production and tree populations are not simulated.

Also, tree are not spatially located, rather tree effects are evenly distributed across the landscape.

Measurements of the effect of trees on microclimate (especially vapour pressure deficit) at a community level have been made. These effects are being progressively implemented in GRASP.

Grazing by livestock leads to changes in grass basal area and this influences initial growth of grass at the start of the growing season. There are no simulated effects of grazing on trees. Similarly, there are no effects of fire on trees (apart from the Moore version).

Steve Archer La Copita site @ level 3

David Breshears USA level 1 coordinator, Pinyon site @ level 2

Mike Cougenhour Level 2 & 3 model experiment

Mike Dodd Two Te Kulti sites @ level 3

David Hall Writing meeting coordinator; funding

Niall Hanan HAPEX-Sahel site @ level 3

Jo House Emerging Issues in Savanna Ecology paper

Richard Joffre Dehessa site @ level 3

John Ludwig Australia level 1 coordinator

Ruben Montes Two Venezuela sites @ level 3

Charles Moyo Africa level 1 coordinator

William Parton Level 2 modelling coordinator; level 2 modelling

Jose San Jose South America level 1 coordinator

Joe Scanlan Mulga site @ level 2 & 3; Poplar Box @ level 2; level 2 modelling

Robert Scholes Database variables description, Nylsvley site @ level 3

Jonathan Scurlock Database structure

Bruce Thurrold Level 2 data coordinator; Tikitere site @ level 2 & 3