Regional Interactions of Climate and Ecosystems


IGBP/GAIM REPORT SERIES

REPORT #3

By: Ann Henderson-Sellers and Catherine Ciret


Click here for the PDF (Adobe Acrobat) version of this report


NOTE- THE FIGURES IN THIS REPORT HAVE BEEN REMOVED FOR EASE OF DOWNLOADING. CLICK ON EACH FIGURE'S LINK TO SEE THE FIGURES.

OVERVIEW

The Land surface and Global Change

Global change involves biogeochemical as well as physical processes. This is particularly true of near-surface continental processes where radiant energy is converted into latent and sensible heat, in addition to living material through photosynthesis.

Human welfare depends critically upon climatic conditions near the continental surface because this is where crops are grown and fresh water is collected. It is now recognized that representation of continental surface processes affects both short-period meteorological prediction (e.g. Beljaars et al., 1993) and longer-term climatic projections (e.g. Xue and Shukla, 1993). Moreover, poor or inadequate representation of both physical and biogeochemical processes at the continental surface can have significant impacts on future projections of carbon uptake by the biosphere and soil moisture available for crop production.

The project on Regional Interactions of Climate and Ecosystems (RICE) is a modeling activity of GAIM. The goals of RICE are to:

- ascertain the regional effects of vegetation and soils on climates simulated by global models;

- establish the sensitivity of vegetation and ecological schemes to regional climates derived from global models; and

- facilitate the integration of new vegetation/ecological schemes into global models.

RICE research has synergistic overlap with GCTE and BAHC within IGBP, as well as, with the WCRP/GEWEX's Project for Intercomparison of Land surface Parameterization Schemes (PILPS).

The activities of the RICE project have been focused in three complementary research areas:

(i) ascertaining the effects of climate and soils on climate simulated by global models

(ii) establishing the sensitivity of terrestrial vegetation models to the climates simulated by global climate models (GCMs) and

(iii) facilitating the integration of new Core Project-derived models into coupled global models. These are outlined below and described in detail in the body of this report.

A joint RICE/PILPS workshop took place in 1994, at the Climatic Impacts Centre, Macquarie University, Sydney, Australia [Henderson-Sellers, 1996b]. The major goal of the workshop was to increase the understanding of the parameterization of soil moisture in climate and vegetation models. The major objectives of the workshop were to:

- quantify the differences in soil moisture predictions among the land surface parameterization schemes;

- determine whether these differences are important for atmospheric and/or ecological models (vegetation models, hydrological models, atmospheric boundary-layer models and GCMs); and

- understand whether differences occur because of:

- theory,

- numerical implementation,

- coding

- choice of parameters.

These objectives were to be achieved in the time frame of the workshop as a whole including the 9 month preparation, the workshop itself and subsequent and continuing evaluation of the results. Carefully designed numerical experiments were carried out before the workshop and during the two week period of the workshop additional experiments were conducted as well as the analysis of results. Intensive scientific discussions continued throughout.

Fifteen scientists from around the world, representing 14 land surface schemes, participated in the workshop. The workshop was organized and coordinated by Ann Henderson-Sellers.

Another focus of the RICE project was to assess the uncertainties in the prediction of vegetation using climate model output. Although GCMs and vegetation models have been widely used in the assessment of climatic impacts on ecological systems, it is commonly admitted that there remain large uncertainties in the GCM predictions. Specifically, climate predictions are often inaccurate at the regional level. Hence, the use of climate models in the estimation of the impact of climatic change on global and regional ecosystems remains highly problematic.

In order to evaluate the reliability of climate models with respect to ecosystem modeling, two global equilibrium vegetation models, BIOME1 [Prentice et al., 1992] and a version of the Holdridge scheme [Holdridge, 1967] were used in conjunction with several climate model experiments from the Model Evaluation Consortium for Climate Evaluation (MECCA) project [Henderson-Sellers et al., 1995]. The aim was to identify which simulated ecosystems were sensitive to the biases of the climate simulations.

The GCMs and global vegetation models are linked in a one way mode (i.e. no feedback from the vegetation model to the GCM is allowed). The approach consists of comparing the vegetation distributions predicted by the two vegetation models using simulated climates against the vegetation distributions predicted using observed climate. The differences in vegetation predictions are overall due to the overestimation of the soil moisture index and precipitation, to the overestimation of growing degree days and to the underestimation of the annual minimum temperatures. Certain biomes appear to be particularly sensitive to the biases in the simulated climates (e.g. grassland, xerophytic woods). Overall, the discrepancies in vegetation predictions are predominantly due to biases in the simulation of the hydrology (i.e. soil moisture index and total annual precipitation) and these results indicate that the uncertainties in the simulation of the soil moisture availability should be carefully evaluated before a high degree of confidence can be vested in the prediction of vegetation, and moreover in the prediction of vegetation change.

The results indicate that the overall performance of coarse resolution climate models with respect to vegetation prediction is rather poor. The discrepancies between vegetation distributions computed from observed and simulated climatologies represents more than 50% of land area. The comparison of vegetation distributions shows that there are some common tendencies amongst the GCMs used in this study to induce the over-prediction or under-prediction of certain biomes. For example: the biomes belonging to dry climate regions are under predicted, and the woodlands (i.e. xerophytic woods/shrub) and temperate/cold forests are over predicted.

In this part of the RICE project, the sensitivity of vegetation models to changes in the spatial horizontal resolution of GCM was also investigated [Ciret and Henderson-Sellers, 1997a]. The climate integrations come from a set of experiments undertaken by Williamson et al. [Williamson et al., 1995] in which the GCM CCM2 is run with increasing spatial resolutions. The global scale vegetation prediction is found to be improved when using higher resolution climate simulations. These results confirm those from Claussen and Esch [Claussen and Esch, 1994] who found that the GCM ECHAM-T42 generated better climate simulations with respects to ecosystem prediction than ECHAM-T21. However the best results are not necessarily obtained with the highest resolution [Williamson et al., 1995]: for instance a more "realistic'' vegetation distribution was obtained with the BIOME model using the T42 climate integration instead of the T63 climate integration. It must also be noted that certain biases in the climate simulation persist and in some cases are enhanced at higher resolution. Therefore the vegetation prediction remains, in some regions, particularly poor. Hence the improvement in the biome prediction appears to be uneven. The biomes which benefit clearly from the increased spatial resolution of CCM2 are cool forests, seasonal tropical forest, tundra and hot desert.

In general, this part of RICE research has shown that the prediction of biomes using simulated climatologies is not yet fully satisfactory. It is nevertheless possible to increase our level of confidence in the prediction of vegetation by carefully evaluating the performance of the vegetation models driven by simulated climatologies and by identifying the causes of the biases.

The third focus of the RICE project has been to evaluate the effects that land-surface processes, particularly soil moisture simulation, have on the simulated climate and to assess, in turn, the impacts of the simulated climate on natural ecosystems. The interactions between vegetation and atmosphere are a key issue in climate system modeling, however the vegetation characteristics represented currently in the soil-vegetation-atmosphere transfer schemes (SVAT) are often prescribed and are not allowed to fully respond to the climate forcing. Hence it was decided that a necessary first step was to improve the representations of the short-term dynamics of the vegetation functions in land-surface schemes (i.e. seasonal and interannual variations of the vegetation functions). The RICE project proposed first to focus on selected regions and/or vegetation types for which the simulation of the hydrological cycle and of the plant-water relations are particularly critical, such as tropical seasonally dry regions.

The parameter chosen for the analysis was the Leaf Area Index (LAI) because of its critical role in controlling evapotranspiration and rainfall interception [Parton et al., 1996], and the region selected was a site in West Africa where the marked seasonality of LAI is induced by the pronounced seasonality of soil water content. An approach was developed to simulate the seasonal variations of LAI. The aim was to employ a model simple enough to be implemented without delay in a GCM, and which could be driven by simulated climate variables (including soil moisture variables), generated by the GCM LMD. The approach consisted of using a plants primary productivity and phenology model developed by Le Roux [1997] according to the conditions found in Lamto Scientific Site, Western Africa. This plant primary productivity phenology model was modified and simplified to be implemented in the GCM LMD [Ciret et al., in review]. This model simulates the temporal variations of green and dead aboveground biomass and leaf area inex. In addition, it simulates the occurrence of savanna fire. The predictions were compared to harvest measurements of grasses from two different regions of savanna; the region of Lamto, west Africa, and the region of Victoria River district, Northern Australia. Results show that this plant production and phenology model generates reasonably realistic plant seasonal variations in these two regions, despite the existing biases in the simulated climate variables. Moreover, both the frequency and timing of fire occurrence are realistically simulated.

One of the expected consequences of climate change from increasing greenhouse gases over the next century will be changes in distribution of biomes (and rates of carbon pool cycling). These changes will, in turn, modify the climate changes. Present model simulations of climate change from greenhouse warming assume prescribed distributions of biomes and non-interactive scenarios for changing atmospheric carbon dioxide. Two-way coupling between climate change on the one hand and the effects of biome distribution and carbon fluxes, on the other hand, must eventually be addressed. A prerequisite to carrying out such two-way coupling is first the validation of the individual subcomponents. There is presently developing a considerable body of information regarding the effects of vegetation changes on regional climates. Thus, an important activity of this project has been to synthesize this present body of knowledge, identify gaps and put it into a global framework. Likewise, as a second activity, we explored how to use GCM supplied regional climate excursions to drive simulations of terrestrial ecosystems or changes in biome distributions. As such, this research effort spans the contemporary period and the futures time frame established for GAIM research.

In summary, the purpose of the RICE activity has been to understand better the relationships between climate and terrestrial ecosystems by focusing upon changes (natural and human-imposed) to terrestrial ecosystems and regional climate.

 

RICE Research Activities:

I. Ascertaining the effects of vegetation and soils on climates simulated by global models.

Questions to be addressed:

(i) Is the simulated climate affected by the prescription of the continental surface?

(ii) Do feedbacks increase or dampen divergence between climate and vegetation and soils?

(iii) How long must a continental surface change persist for a climatic change to become detectable?

II. Establishing the sensitivity of terrestrial vegetation models to the character (including initialization and extreme events as well as mean conditions) of climates simulated by global models.

Questions to be addressed:

(i) How sensitive are existing and developing vegetation models to poor climate simulations?

(ii) How sensitive are existing and developing vegetation models to extremes in climate simulations?

(iii) How good does initialization have to be for existing and developing vegetation models? (cf. PILPS)

III. Facilitating integration of new Core Project-derived models into coupled global models.

Questions to be addressed:

(i) Are the prescription/calculation of physical components e.g. soil moisture and soil and canopy air temperatures compatible between the vegetation and host models?

(ii) What variables need to be exchanged for coupling and how frequently?

(iii) What sensitivity does the host model exhibit to variations likely to be prompted by the new component?

I. SOIL MOISTURE SIMULATION

Introduction

Soil moisture is a key component in the land surface schemes as it is closely related to evaporation and thus to the apportioning of sensible and latent heat fluxes. Accurate prediction for soil moisture is crucial for the simulation of the hydrological cycle and of soil and vegetation biochemistry and thereby plays a significant role in atmospheric models, hydrological models and ecological models. The major objective of the soil moisture workshop was to increase our understanding of the parameterization of soil moisture in different schemes and to provide a quantitative assessment of soil moisture simulation in current land surface schemes.

It became clear early on in the exercise that soil moisture is of critical importance to the physics and the biogeochemistry of the continental surface. It has become increasingly clear that different communities, particularly meteorologists, hydrologists and ecologists have different perceptions of the relative importance, the time constants and even the 'laws' governing the processes occurring at the land surface. These different perceptions were being encoded into numerical schemes, all of which captured some of the attributes of the soil and vegetation, and the water, energy and momentum exchanges but none of which were exactly similar. A common element, and one of critical importance, is soil moisture. This recognition prompted the decision to hold the 'Soil Moisture Simulation Workshop' in November 1994. The workshop was co-sponsored by IGBP/GAIM/RICE and WCRP/GEWEX/PILPS. It also drew funding from IGBP/BAHC and the Climatic Impacts Centre, Macquarie University, the host institution.

In this report are described the preparatory work (lasting about 9 months) which underpinned the workshop, the participating schemes, the very intense two week workshop itself, and the outcomes and recommendations made by the participants.

Land surface schemes used in atmospheric models and ecosystem models have as their principal commonalty the need to simulate soil moisture. While atmospheric models require accurate descriptions of the state and fluxes of water at the surface to assure the realistic partitioning of incoming energy into sensible and latent heat fluxes, terrestrial ecosystem models require this same information to predict the cycling of carbon and nutrients through various organic and inorganic phases. Both groups, insofar as model validation is concerned, are interested in accurate descriptions of soil water and hydrology in general as one of the best sources, in some cases the only source, of validation data.

The treatment of these processes in a particular model is largely determined by the spatial and temporal resolution of the model employed. Atmospheric models have short time steps (less than 0.5 hours), but large spatial resolution of tens to hundreds of kilometers because of the numerical and dynamical constraints. Hydrological and ecosystem models operate under other constraints, and have time steps in the order of days to months. Because the numerics involved are relatively simple compared to atmospheric models, there is more flexibility in specifying the spatial resolution which is often constrained by the availability of spatially explicit data sets and the aims of the particular application rather than by numerical considerations.

This polarity between spatial and temporal resolution for atmospheric and hydrological models, when considering the application of each over large spatial domains, leads to a polarity in the treatment of the soil moisture processes. With very short time steps, the land surface schemes used in atmospheric models must treat the vertical movement of water and heat in the soil mechanistically. However, with a horizontal spatial resolution on the order of tens to hundreds of kilometers, explicit treatments of the horizontal movements of water at the surface are necessarily very coarse. On the other hand, with longer time steps, in the order of days to months, it is neither necessary nor possible for hydrological models to assess the details of vertical water movements in the soil column, but with a greater flexibility in the definition of the horizontal spatial resolution of the land surface, these models can and occasionally do incorporate much more sophisticated diagnoses of horizontal movements of water at and under the surface. Ecosystem modellers are concerned with some aspects of both the atmosphere and hydrological processes but tend to consider time and space scales closer to hydrology than meteorology.

While the separate research communities seem to have been fairly well served by this arrangement to date, these simple divisions are disintegrating rapidly. Hydrological models and ecosystem models are being 'plugged into' GCMs and the space and time scale incompatibilities must be recognized and overcome. The precise horizontal distribution of fluxes at the surface may be less important to the accurate assessment of long-term patterns of atmospheric dynamics than is an accurate assessment of the magnitude of those fluxes at short time steps. On the other hand, knowing the relative horizontal distribution of the quantities and fluxes of surface (including subsurface) water is critical to the accurate assessment of the major drainage features and structural and dynamic aspects of terrestrial ecosystems, while the exact magnitude of these quantities and fluxes may not be as critical. There remains one common factor: soil moisture.

In short, soil moisture is a key component in land surface schemes and is of great significance to atmospheric , hydrological and ecological models. Soil and vegetation play an important role in the hydrological cycle and influence atmospheric systems on time scales from several hours to many years. Soil water content is closely related to evaporation and thus to the partitioning of sensible and latent heat fluxes at the surface. Soil moisture, the atmospheric boundary layer and convective clouds is a coupled system leading to feedbacks in short term weather forecasting models and to longer-term climate variability and change. Apart from solar radiation and soil nutrients, the availability of soil moisture is the key to plant growth and to the net production of crops. For these reasons, the intercomparison of soil moisture simulation in land surface schemes is the focus of this joint PILPS and RICE workshop.

Land surface Parameterization Schemes

The processes parameterized in land surface schemes can be broadly divided into three categories: sub-surface thermal and hydraulic processes, bare soil transfer processes, and vegetation processes. Various treatments of surface transfer processes differ technically, but they are similar conceptually. The differences between land surface models are more clearly seen in the configuration of the soil layers and the treatment of soil hydrological processes, which in turn is closely related to the time scale problem mentioned above.

In this section, the theoretical structure of the land surface schemes tested in the soil moisture workshop are compared. The land surface schemes are described from five perspectives: scheme structure, treatment of sub-surface thermal and hydrologic processes, bare soil transfer processes and canopy transfer processes.

Structure of Land Surface Schemes

Although land surface schemes differ in many respects, differences in scheme structure, especially the configuration of soil layers, are of fundamental importance. Most land surface schemes are one-dimensional such that the starting point of a land surface scheme is the one-dimensional conservation equations for temperature and soil moisture


where T is soil temperature, C is volumetric soil heat capacity, W is volumetric soil water content, is density of liquid water and and are sink terms. The vertical heat flux G obeys a simple flux-gradient relationship

where is the thermal conductivity. The vertical soil water flux Q obeys Darcy's law

where is the hydraulic conductivity and is the matrix potential due to capillary force. A substitution of Equations (3) and (4) into (l) and (2) gives diffusion type prognostic equations for soil temperature and soil moisture

A land surface scheme is, in principle, the algorithm required to solve this system of equations for a particular soil layer configuration. The parameterizations occur in the boundary conditions and in the treatment of the sink terms and . The upper boundary conditions at the atmosphere and land surface interface include sensible heat and latent fluxes which are the most important fluxes to be determined in land surface schemes. The sink terms include the soil water runoff and drainage and transpiration by vegetation.

Therefore, differences in configuration, including soil layers, canopy layers and interaction between various components of a scheme, and differences in parameterization are the two major categories of differences in land surface schemes. The schemes tested in the workshop use a single layer canopy treated as a 'big leaf'. The only exception is the bucket model which does not incorporate the canopy explicitly. The major differences in the model configuration lie in soil layers: the number of soil layers for temperature and water content varies between one to eight and some schemes (e.g. BIOME2) do not simulate soil temperature. Figure 1 shows schematically the configuration of the models for soil hydrological processes.

Figure 1: Schematic illustration of runoff and drainage for land surface schemes represented in the workshop. According to the number of soil layers, the schemes are classified into group I (single layer models), group II (two layer models), group III three layer models) and group IV (multi-layer models).

 

Treatment of Soil Thermal and Hydrological Properties

 

Soil Temperature: In most land surface schemes, soil temperatures are determined from an one-dimensional diffusion equation. Assuming no horizontal heat transfer, soil heat conservation is governed by (1) where is the volumetric heat capacity, is density and is specific heat capacity of soil. Substitution of (1) into (3) yields (5), or if the sink term is zero,

is thermal diffusivity of the soil. The soil thermal conductivity depends on soil properties such as soil type, soil wetness and soil cover. Equation (7) is the prognostic equation for soil temperature, described as a simple one dimensional diffusion process.

 

Bare Soil Processes

For bare soil, the main processes parameterized in land surface schemes are evaporation and heat and momentum exchange between the atmosphere and the surface[Desborough et al., 1996]. These exchanges are controlled by atmospheric conditions including wind speed, temperature and moisture, and the surface conditions including temperature and soil moisture in top soil layers and surface aerodynamic roughness length.

Vegetation Processes

Most of the land surface schemes which represent explicit vegetation processes have four key elements that primarily involve the surface calculations: canopy resistance, albedo, water storage and aerodynamic transfer processes. The representation of these elements in different schemes are described below.

The HAPEX data set

High quality atmospheric forcing and validation data are both necessary for a realistic assessment of land surface schemes. Four data sets were available to the workshop: the Amazon data set, the Cabauw data set used in PILPS phase 2, the FIFE data set and the HAPEX-MOBILHY data. After checking these four data sets, it was concluded that the HAPEX data set was the most suitable for the workshop, because it included a full year atmospheric forcing, with weekly soil moisture measurements to 1.6 m with 0.1 m interval and energy flux measurements between 28 May and 3 July [Shao and Henderson-Sellers, 1996].

The HAPEX data were obtained from HAPEX-MOBILHY at Caumont (SAMER No.3, 43¡41' North, 0¡6' West, mean altitude is 113 m). Detailed information on the SAMER network and the site can be found in Goutorbe [1991]. Most of the forcing data were taken from Caumont, particularly during the Intensive Observation Period (May, June and July, 1986). If data at Caumont were missing, measurements from neighboring meteorological stations were used. Therefore, the forcing data may be not fully consistent with the validation measurements for short and intermittent time periods. However, this small inconsistency in the data should not have any significant effect on the intercomparison and validation of land surface schemes, in particular if time step values are not of concern. Figure 2 shows the atmospheric forcing data.

Figure 2: HAPEX atmospheric forcing data, including downward short-wave solar radiation (SolDn), downward long-wave radiation (LWDn), precipitation (Pr), air temperature (Tair), wind speed (Uair), pressure (Pres) and specific humidity of air (Qair), in 30 minutes time interval.

Design of the Numerical Experiments

The numerical experiments were designed to achieve the goals of the workshop, including quantifying the capability of land surface schemes in modeling soil moisture content, quantifying the differences among land surface schemes and understanding why the differences occur. The design of the experiments took into account the interests in climate, hydrological and ecological modeling among the workshop participants, and the integrity and the feasibility of the experiments.

As listed in Table 1, five experiments were proposed and carried out with the pre-workshop version of schemes during the preparation period before the workshop. In these experiments, Experiment 1 is the control experiment in which the atmospheric forcing data and various land surface parameters obtained from the HAPEX experiments were used to drive the land surface schemes. As discussed in the previous section, the HAPEX data also included homogenized validation data, including measurements of soil water content and surface energy fluxes for an intensive observation period. Experiment 1 not only permits a study of the performances of the land surface schemes against observations, but also provides the reference result set for the other experiments.

Despite the differences in structure of the models and the differences in the details of parameterization, all land-surface schemes are comprised of three basic components: bare soil transfer processes, canopy transfer processes and soil thermal and hydrologic processes. The pre-workshop experiments were aimed at increasing the understanding of the treatment of evaporation, transpiration and soil water content. In addition to the control experiment, four additional experiments were designed to test the treatment of evaporation, transpiration and soil water content separately or partially linked in the light of Experiment 1.

Table 1:  Soil Moisture Workshop Experiments

Workshop Experiments

Additional tests for model sensitivity to atmospheric forcing, specification of parameters and tests for the behavior of certain model aspects were conducted during the workshop. These additional experiments are designed to address the questions identified in studying the results of the five pre-workshop experiments. Some of the pre-workshop experiments were repeated after improvement and clarification in the design. The numerical experiments conducted during the workshop are also listed in Table 1.

Assessment of Soil Moisture Simulation

The control experiments allowed an assessment of soil moisture simulation in various land surface schemes by intercomparison and comparing the numerical results with HAPEX observations. Taking into account the configuration of most land surface schemes tested in the workshop, we considered soil moisture in three soil layers.

- Soil moisture in the total soil layer (1.6m for HAPEX);

- Soil moisture in the top 0.5 m soil layer which contains 90% of roots and is most important for vegetation;

- Soil moisture in the top 0.1 m soil layer, which is closely related to evaporation.

The analysis of soil moisture in the two top layers is more difficult, because of the high frequency fluctuations and the noise contained in the data. Most of the results presented here are based on Experiment 13 which is the final control experiment for the workshop.

Workshop Results

Summary

The first major effort of the workshop was to quantify the differences in the soil moisture predicted from the 14 participating land surface schemes. Intercomparison among land surface schemes and with HAPEX soil moisture measurements revealed a large difference of about 200 mm initially among the schemes (from control experiment 1, see Fig. 2) and this difference in soil moisture was accompanied by differences in evaporation, transpiration, sensible heat fluxes and in runoff and drainage. The disagreement in simulated soil moisture and other simulated variables was reduced in Experiment 13 compared to Experiment 1 after improvement in schemes and careful adjustment in land surface parameters, especially in the parameters characterizing the soil hydraulic properties (wilting point, saturation soil water potential and saturation soil hydraulic conductivity), and in the parameters characterizing the surface properties (leaf area index, fraction of vegetation cover, vegetation height, aerodynamic roughness length and zero-displacement height). The scatter in soil moisture for the total soil layer (1.6 m) is about 70 mm for the bare soil period and 100 mm with a maximum of 143 mm in the growing season in Experiment 13, compared about 200 mm for throughout the year with a maximum of 240 mm for Experiment 1 (Fig. 3). Thus, the range in the 'improved' control experiments remains large, even after more careful and consistent choice of parameters.

Figure 3: Soil moisture (mm) in the total soil column (1.6 m) simulated for various schemes for Experiment 1 (top) Experiment 13 (bottom).

The other major effort of the workshop was to study to what extent these differences are due to the theoretical divergence among the models, including structure and parameterization. The results obtained during the workshop indicate that the difference in model structure is a primary reason for disagreement in soil moisture predictions. For example, BGC initially had a single soil layer. Incorporation of three soil layers into the model made huge differences to the resulting simulations. Most land surface schemes tested in the workshop conserve water. However, the partitioning of sensible and latent heat fluxes (in the surface energy budget equation) and the partitioning of evaporation and runoff-drainage are profoundly different among the schemes as shown in Figure 4 and Figure 5. It was observed that despite the significant improvement in the agreement of soil moisture prediction in Experiment 13 compared to Experiment 1, there was no significant improvement in the scatter of evaporation and runoff drainage as illustrated in Figure 6.

Figure 4: Running integration of surface energy budget residual (W/m2) for different schemes from Experiment 13

Figure 5: Comparison of partitioning of annual mean sensible and latent heat fluxes (W m-2)

Figure 6: Same as Figure 3 but for running integration of total runoff plus drainage.

It was found that surface energy fluxes are well balanced for most of the land surface schemes with the maximum imbalance being smaller than 0.05 W m-2. However, the partitioning of surface energy fluxes, which is closely coupled to the partitioning of evaporation and runoff and drainage, is profoundly different. Although on average, the difference in net radiation is about 8 W m-2 that of latent heat fluxes is approximately 17 W m-2 and that of sensible heat fluxes is of the same magnitude. In terms of energy transfer, these differences correspond to 250 MJm-2 for net radiation transfer and 536 MJm-2 for latent and sensible heat fluxes.

Other important differences in land surface schemes are embedded in detailed parameterizations of individual processes, choice of parameters and numerical implementations. To study the individual processes that control these differences, further experiments were designed to try to evaluate: (a) evapotranspiration, (b) bare soil processes and (c) drainage.

Transpiration: It was clearly demonstrated during the workshop that the disagreement among schemes in soil moisture prediction, evaporation and sensible heat fluxes are largest during the growing period (see Fig. 7a and 7b), caused possibly by profound differences in the treatment of transpiration [Parton et al., 1996]. During the workshop, much attention was paid to trying to understand the variation between models in simulated evapotranspiration and the changes in soil moisture during the growing season. We were able to document the theory of the transpiration formulations, and to compare simulations with observations.

Figure 7a: Comparison of running mean (over 30 days) latent heat flux (W/m2) for various schemes and comparison against HAPEX data from Experiment 1.

The formulations of transpiration are extremely variable between the models represented at the workshop. Some of this variation is due to fundamental structural differences between models, such as different time-steps or different soil layer structures, and some of the variation is due to differences in the conceptualization of the transpiration process by the scheme designers. However, stomatal resistance is commonly used to parameterize complicated mechanisms controlling transpiration involving the interaction of abiotic with biotic factors.

After comparing the control experiments with observed HAPEX data and analysis of additional experiments, it can be concluded that there appear to be two categories of transpiration components in the schemes represented in the workshop: those falling into the low evapotranspiration group and those in the high evapotranspiration group. It is suggested that these differences may be related to the structure of the schemes (although the differences caused by the choice of parameters and the consistency of this choice with the physics represented by the model are yet to be quantified): all three of the vegetation models and BATS, CLASS, LAPS and VIC fall together in the low evaporation group. However, because of the incomplete nature of the observation data sets, it remains uncertain which of these groups is more likely to be closer to the truth.

Start of growing season water content vs. the total soil water depletion over the growing season shows two distinct groupings and one rather diffuse group of models: a) a group that has both initial soil water and growing season depletion resembling the observations; b) a group with initial water content close to observations but with water depletion significantly greater than observed; and c) another widely scattered group with generally high values of both initial water and water depletion. These groupings should serve as the basis for further analysis, by which greater understanding of model differences should be gained.

Figure 7b: Same as Figure 7a, but for sensible heat flux.

Bare Soil Evaporation: Bare soil evaporation is possibly the most well studied process in land surface schemes and remained an important subject of the workshop. Experiments 2al and 2a2 were designed to try to determine the differences in bare soil evaporation formulations. In Experiment 2a2, the role of soil moisture and aerodynamic drag coefficients are eliminated by specifying them. Even after specifying these two parameters the scaling evaporation still differs and the bare soil evaporation differs by a factor of 2 among the models for Experiment 2a2. This difference in bare soil evaporation is a result of the different parameterizations used in the models. The main conclusions from these experiments are: (a) the difference in the scaling evaporation between the models is expected to be due to the differences in the philosophy used; and (b) the differences in the bare soil evaporation is due to the differences in the scaling evaporation and due to the differences in the and parameterizations used.

Runoff and Drainage: The third major component in soil moisture simulation is runoff and drainage [Mahfouf et al., 1996]. Its treatment is crucial to soil moisture simulation, but is possibly the least adequately considered in land surface schemes. During this workshop, we studied the different theoretical treatment of runoff and drainage and conducted analysis of the control experiments. Since total runoff and drainage is related to soil moisture and evaporation, a 'pulse precipitation' experiment (Experiment 2cl) was designed to study the difference in total runoff and drainage induced by the difference in model structures and formulations.

It was shown that a combination of the model structure and the different formulations of runoff and drainage led to large differences in the cumulated animal total runoff. The difference in annual total runoff occurred nearly entirely in the first 120 days when soil moisture is near its field capacity. Limited observational evidence indicated that there may be some kind of quick subsurface and drainage flow (such as macro-pore and pipe flow) at the studied HAPEX site. Therefore, any models that consider macro-pore and/or pipe flow in their runoff and drainage formulations should perform well for the HAPEX site if evaporation was correctly simulated. Also, Experiment 2cl indicates that the percentage differences in total runoff and drainage are reduced among at least the seven models as compared with the percentage differences among the same seven models in Experiments 13 and/or 15.

However, even after differences were identified and to some extent and removed, very large discrepancies remain: (a) among the models, and (b) between all the models and the observations. Important remaining points include:

- No scheme did well both during the main growing period of the crop and over the annual cycle. Compared with HAPEX observations, models which predict total growing season water loss well, seriously under predict evaporation during the 35 day intensive observation period (IOP) whereas models which reproduce IOP fluxes well dry the soil excessively by the end of the growing season.

- Large differences remain in the evapotranspiration between the schemes with the range in cumulative evapotranspiration for the full growing season being as large as 134 mm. In the 35 days of the IOP for which the measurements of the total latent heat flux were available, the total evapotranspiration predicted by the models varied from 81-167 mm with the measured value being 124 mm ±10%. Only about half of the schemes predicted total evapotranspiration within the error bars of the observed value.

- Large differences also remain in the bare soil evaporation with the range in cumulative evaporation for the first four months up to 135 mm.

- Treatment of drainage, runoff and lateral flows differs greatly among schemes. Observational catchments suggest that about 290 mm of water leaves the soil via subsurface processes during the experimental year. Although the consensus of all models is near 200 mm, individual model predictions of total drainage range from 40 to 300 mm.

- Considerable disparity persists in model predictions of the extremes of soil moisture and evapotranspiration during the year. At the peak of bare soil evaporation in April, model predictions range from 50 to 140 Wm-2. During the summer minimum of bare soil evaporation, model predictions range from 0 to 25 W m-2. Within these ranges it was difficult to group models, as they varied across the range.

- Different schemes achieve different annual equilibrium when forced with the same atmospheric forcing data and the same land surface parameters.

- Different schemes describe the seasonal cycle of soil moisture differently with the greatest dispersion occurring when vegetation contributes to the total evaporative flux, there is a great atmospheric demand and the available soil moisture is limited.

- Different schemes deal with incoming precipitation differently partitioning it into runoff-drainage, soil storage and evaporation differently at different times and depending (differently) upon the antecedent conditions.

- Most schemes can be tuned to observations but no single scheme predicts all the variables describing the land surface hydrology well. Indeed the consensus (single average) of all the participating schemes generally out-performs all individual schemes.

This suggests that individual schemes capture specific aspects of this complex system well but no one scheme yet captures the whole system satisfactorily and consistently.

The importance of these differences in modelled variables also impact ecological modeling. Soil decomposition, releasing carbon and nitrogen, is a function of the soil moisture and the soil temperature. The decomposition factor calculated from the individual schemes using their simulated temperature and soil moisture show a scatter from 0.2 to 1.8 between the individual schemes. This implies that the differences in soil moisture have important consequences for global carbon modeling.

From the workshop results, three major general conclusions can be made. Firstly, there exists large differences in soil moisture modeling and other modelled variables; even for simulations run with high quality atmospheric forcing data and carefully chosen parameters. Therefore, the prediction of soil moisture in climate change, weather forecast or hydrological simulations cannot be considered as reliable when the forcing data are much less accurate and the information required for specifying land surface parameters is crude. Secondly, current land surface schemes are profoundly different in structure and in their treatment of various land surface processes such as evaporation, transpiration and drainage, with the differences in scheme structure apparently being most important. Finally, land surface schemes comprise two closely coupled components responsible for: the partitioning of sensible and latent heat fluxes and the partitioning of evaporation and runoff-drainage; and among these it is the treatment of runoff and drainage which deserves much improvement and more careful consideration.

Recommendations

A number of important problems were identified during the workshop which need to be addressed in future studies. These problems may be related to the philosophy and basic structure of land surface schemes, or to the details of parameterizations and to land surface validation data.

Equilibrium States and Equilibration: Since the initial conditions required for land surface parameterization schemes are often not available, it is necessary in intercomparisons that a land surface scheme is run for many years and reaches an equilibrium with the pre-specified atmospheric forcing. In other words the atmospheric forcing (a periodic forcing over one year period) is applied repeatedly until all prognostic variables have exactly the exact same behavior as for the previous year.

One of the most important observations made during the workshop is that the land surface schemes under the same atmospheric forcing reach very different equilibrium states (e.g. Figs. 7a, 7b and 3). This result reveals some fundamental structural and parameterization differences among land surface schemes. It is readily understood that the difference in equilibrium states makes it difficult to compare the performance of land surface schemes in a particular sub-annual period (e.g. the bare soil period or the growing season), since these are inseparable entities in a complete cycle. The associated consequence is that the validation data which are often limited to a couple of weeks or months cannot be effectively applied. It is clear that detailed studies on the equilibrium states and equilibration of different land surface schemes are both important and necessary in future intercomparisons and development of land surface schemes.

Response Time: A closely related problem to that of equilibrium states is scheme response characteristics. The response characteristic can be defined as the rate at which a land surface scheme approaches equilibrium with the changing external forcing (for instance precipitation and solar radiation) and/or land surface parameters (such as leaf area index). The response characteristics are determined by a combination of parameters within the schemes and the scheme structure. It is obvious that the response time of a bucket model is very different from that of a multi-layer model. For a bucket model, the soil water throughout the soil depletes immediately as evaporation occurs (rapid response scheme), while for multi-layer schemes, water in deep soil layers is depleted through diffusion to upper layers, and then evaporated from the surface. The latter process requires a much larger time than in the bucket scheme. Therefore the differences in the scheme response characteristics is a very important aspect of land surface schemes. It is necessary to understand these response characteristics in many applications of land surface schemes.

Energy Partitioning and Water Partitioning: Energy partitioning (i.e. the partitioning of available energy between surface sensible and latent heat fluxes in the surface energy budget equation), and water partitioning (i.e. the partitioning of precipitation between evaporation and runoff-drainage in the water budget equation), are two closely coupled and fundamental aspects of land surface schemes. The various land surface schemes balance surface energy budget and water budget vary differently. Although some land surface schemes obtain a reasonable simulation of soil water, this is achieved through very different partitioning of evaporation and drainage. For some schemes soil moisture simulation is controlled by evaporation while other schemes achieve the same soil moisture by allowing more drainage (Fig. 4). Consequently, the energy partitioning among land surface schemes are also significantly different (Fig. 5). Hence, although the emphasis of land surface schemes developed for atmospheric models is the energy partitioning, and the emphasis of land surface schemes for hydrological and ecological models is water partitioning, the two partitionings should be treated as an inseparable entity in all land surface schemes.

Treatment of Runoff and Drainage: Despite the importance of runoff and drainage in soil moisture simulation, its treatment in land surface schemes appears most inadequate. It has been clearly demonstrated that both the formulation of runoff and drainage and their prediction are extremely different. It is recommended that the treatment of runoff and drainage deserves much more careful consideration in the development of land surface schemes.

Improved and Coherent Observations: There is a critical lack of adequate observational data sets for joint intercomparisons. Data should involve many years of hydrology, at least one year of atmospheric forcing data, details of soil and growing period phenology and at least one intensive observational period measuring biomass accumulation, soil moisture and atmospheric fluxes.

II. COUPLING VEGETATION AND CLIMATE MODELS

Introduction

In this part of RICE research, a pair of experiments was conducted from which the sensitivity of one global climate model to imposed "dynamic" changes in vegetational form can be assessed. The time scale chosen was intentionally as short as possible from the point of view of ecological changes (one year). Although this is much too short a time step to represent, with any verity, ecological changes, it is useful in the context of the half century climate simulations undertaken here. The experiments reported here contribute knowledge at one end of the spectrum of necessary sensitivity assessments. Here we examine the sensitivity of a global climate model to annually imposed changes in vegetational functional form [Ciret and Henderson-Sellers, 1997b]. This sensitivity testing, a necessary first step towards full coupling, employs a methodology analogous to "instantaneous" deforestation [Dickinson and Henderson-Sellers, 1988; Nobre et al., 1991] or "instantaneous" doubling of stomatal resistance [Henderson-Sellers et al., 1994; Pollard and Thompson, 1994].

Transitions between different biome or ecosystem types are a function of a wide range of processes which operate on many time and space scales. Rapid shifts (<5 years) generally, of necessity, involve decreases in biomass. Examples are severe drought, frost, hurricanes and fire [Doyle, 1981; Noble and Slatyer, 1980; Tucker et al., 1991]. Medium (10-50 years) "recovery" of forests can be observed in benign environments such as the S.E. Asia islands and if human influence is removed. Much longer (50-500 years) ecological succession can be simulated [Shugart, 1984] but at present only by presuming a fixed climate or a prescribed changing climate (i.e. known in advance and independent of vegetation changes) [Prentice et al., 1993]. In simulations of future enhanced greenhouse conditions, the climate cannot be assumed to be fixed or known a priori nor can human influence be denied [Ojima et al., 1994]. Thus it must be assumed that rapid as well as longer-term vegetation changes will occur as climate changes. There is a variety of ways of considering the changes in the continental surface characteristics associated with these vegetation disturbances:

(a) Ignore them.

(b) Simulate climate change presuming that vegetation remains fixed and when climate equilibrates generate continental ecology.

Both (a) and (b) ignore impacts of vegetation on climate.

(c) Use results of (b) and continue at equilibrium to achieve a new climate as modified by the new vegetation. This approach, which may become an iterative process, also assumes that there is a future point at which climate change will stop (presumably at doubled or tripled C02) so that vegetation can "catch-up". It ignores shorter- term climatic effects on vegetation due to droughts, frosts and fires and other extreme climate events and their feedbacks.

(d) Apply a similar vegetation type diagnostic model at intervals during a transient or equilibrating climatic change assuming that this would capture faster ecological changes (i.e. reductions in biomass) but speed-up slower (increasing biomass) changes.

(e) Employ a "dynamic" vegetation Model globally for long enough for both vegetation and climate to equilibrate (0-500 years).

All published GCM simulations to date have used technique (a). IPCC 2 [Tegart et al., 1990] and Smith et al. [1992] and other impact assessors adopted technique (b). A few have tried (c) [Claussen, 1994]. It is assumed that technique (e) is optimum but such "dynamic" vegetation models, presumably based on the principles underlying succession models are still being ceveloped. This use of succession type models for global simulations of climate and vegetation changes demands recognition of the fact that not only do the climate and the biomes change but also the cause-and-effect relationships between them are subject to change.

Technique (a) is accepted by GCMs-because it is easy(?) -although it ignores vegetation feedbacks.

Technique (b) is apparently acceptable to impact modellers although it too ignores vegetation feedbacks.

Technique (e) is the goal; (c) and (d) may be steps towards this goal.

We explore technique (d) and choose a 1-year meshing (coupling) period.

The relationship between vegetation and climate is symbiotic but not exclusive: soils, fauna and human activities all impact vegetation (and climate) [Riebsame et al., 1994]. In an idealized globe where only climate and (above ground) vegetation co-exist it is possible to recognize a range of time scales: slow transitions, "speedy" opportunistic proliferations and die-back, instantaneous wind throw and fire.

It might be more realistic to select a meshing period of 5, 10 or even 50 or 100 years, because some vegetation changes are slow e.g. forest development or change in forest composition. Elongation of the coupling time step, in some senses, brings technique (d) closer to technique (c) but also removes shorter time-scale feedbacks of disturbed vegetation on climate which we can explore here. There seems to be no obvious "best choice" of time scale for coupling although Claussen [1994] has explored some options and concludes that 5 years produces satisfactory results. In the interests of reducing CPU investment in this preliminary assessment of technique (d), we chose to use the shortest possible time period for meshing the continental surface characteristics and the climate: one year.

We now address two closely linked questions:

"Can a "standard" GCM cope with sudden switches in continental characteristics?"

and "Is the climate sensitive to the changing underlying vegetation?"

It is clear that time and space scale mismatches are at least as difficult as those between the ocean and atmosphere and between ocean, ice and atmosphere where thermodynamic models can switch sea-ice on and off in 30 minute "blinks" and 250,000 km2 clouds "flicker" on the same time periods. We believe that our use of technique (d) is in line with similar developments in global climate models and especially coupling of subcomponents over the last 15 years. We have not, here, considered changes in the behavior of vegetation in response to increased atmospheric CO2 but note that this has been explored elsewhere [Henderson-Sellers et al., 1994; McMurtrie et al., 1992; Pollard and Thompson, 1994] and could, readily, be linked to the coupling developed here.

Vegetational Functional Form

Models of vegetation which could be described as dynamically interactive are being developed. Global Dynamic Vegetation Model can be contrived by linking existing land surface parameterization schemes to ecological models through the mechanism of an agreed description of vegetational functional form. In this study, we examine such a linking using 11 functional vegetation types selected from the 18 usually represented in the Biosphere-Atmosphere Transfer Scheme (BATS) [Yang and Dickinson, 1996;Dickinson et al., 1993; Dickinson et al., 1986]. These 11 functional vegetation types have been selected based on a number of criteria including (i) their combined success in reproducing plausible vegetation distributions for the present-day climate as simulated by the GCM used here and (ii) the range of functional characteristics which they represent.

Table 2(a) lists the 11 functional vegetation types used, together with the 16 parameters which BATS requires in the form of a look-up table in order to fully characterize the continental surface. (Note that in these simulations, the soil specification remains constant as originally described by [Wilson and Henderson-Sellers, 1985].) The most important characteristics of present-day land surface parameterization schemes are not yet known, although there have been attempts to rank the vegetation parameters [Henderson-Sellers, 1993b] and there is an international intercomparison under way [Henderson-Sellers, 1993b; Henderson-Sellers and Dickinson, 1992] . It is generally agreed that roughness length is of considerable importance to the physical exchanges of energy, moisture and momentum between the atmosphere and the continental surface [Henderson-Sellers, 1992]. Other characteristics of importance in characterizing the vegetation seem likely to be the leaf area index (LAI), the fractional vegetation cover and perhaps the stomatal resistance. In BATS, as in many "complex" land surface parameterization schemes, these physical properties are calculated as a function of ambient and preceding climatic conditions.

Table 2(a):Sixteen parameters associated with the eleven vegetation types employed.


Table 2(b) includes rankings for the 11 functional types used here based on the prescribed roughness length, the maximum vegetation fraction permitted and the possible seasonal range in this vegetation fractional cover. It can be seen that ranking on a single characteristic (such as vegetation height) may not produce the same ordering as ranking on fractional vegetation cover (closely related to leaf area index), root distribution (related to the ability to continue to transpire when the upper soil layer has dried) or stomatal resistance. Table 3 shows the way in which the vegetational functional types used here relate to the predictors employed in calculating their year-to-year distribution.

Table 2(b): Eleven functional vegetation types (abbreviation) and numeric code used when BATS is coupled dynamically into the global model and rankings by roughness length, maximum fractional vegetation and possible seasonal range in vegetation fraction (largest = 1 in each case).

The range of models of continental ecology

It is not clear from the literature how best to characterize the existing "ecological" models. Heal et al. [Heal et al., 1993] suggest that there are at least seven groups: leaf, crop, CNPS (biogeochemical), stand, landscape, biophysical and biome (Fig. 8(a)). On the other hand, Malanson [1993] describes only three types: transfer functions, stand models and physiological models. Pacala and Hurtt [1993] argue that stand-type models are likely to be the least successful route by which to estimate future ecological change. In particular, they identify two fundamental problems with the application of stand models such as those of Botkin et al. [Botkin et al., 1972] and Shugart and West [1977]:

1.) the confusion of fundamental and realized niches; and

2.) the practice of assuming unlimited dispersal.

There is general recognition that climax/equilibrium models cannot portray arbitrary future joint climate and ecological states, this is particularly true of human imposed land-use change and rapid climate change, both of which lead to disequilibrium. However, it is argued that equilibrium models, while not offering a picture of the future (or past) states can offer some indications of e.g. directions of change and/or possibilities of sustainable land use [Monserud et al., 1993]. McGuire et al. [McGuire et al., 1993] use a biogeochemical model, and the assumption of a fixed biome distribution to provide information about the probable trend in primary production on time scales shorter than those over which biomes respond to altered climate.

The latest description of the plans for coordinated development of a dynamic global vegetation model depends upon the assumption that vegetational functional form, if not the precise nature of the species mix and patch heterogeneity, can be described in terms of fairly simple climatic variables including temperature, precipitation and evaporative demand [Woodward, 1987]. Such assertions suggest that these vegetational functional forms could, at least in the case of assumed equilibrium between vegetation and climate, have much in common with the biome models of Holdridge [1967], Box [1981] and Prentice et al. [1992]. In these biome models, simple relationships are assumed between climate over one or more years and the resultant large-scale ecology Figure 8(b) and Table 3.

In general, there exists a range of ecological models which, while not irrelevant to the task of dynamic coupling to global atmospheric models, also do not yet fully satisfy all the likely demands.

Here, a simple biome model is used to produce the classifications into one of eleven vegetational functional forms which are then used to characterize the information demanded by a biophysical model: BATS.

Figure 8: a) Approximate space and time scales encompassed by the seven types of "ecological" models [Heal et al., 1993]. In this research we link examples of the two largest-scale model types (a biophysical model and a biome model) to a global climate model. b) Examples of the global distribution of vegetation types as a function of monthly minimum temperature (degrees Celsius) and annual precipitation (mm) (both modified from Heal et al., 1993)

Models Used and the Experiments

Models

The global model used here is a version of the NCAR Community Climate Model (CCM), CCM1-Oz, which is integrated at a spatial resolution of about 4.5¡ latitude by 7.5¡ longitude (a spectral truncation at rhomboidal wavenumber 15). A full description of CCM1 is given in Williamson et al. [1987] and circulation statistics from seasonal and perpetual January and July simulations of the standard version of CCM1 are given in Williamson and Williamson [1987].

CCM1-Oz is a modified version of CCM1 which includes the current version of the Biosphere-Atmosphere Transfer Scheme (BATS1E) and a mixed-layer, slab ocean of 50 m depth. The mixed layer ocean model includes a three-layer ice model sub-component and a standard q-flux scheme to correct for ocean advection of energy and the prescription of a fixed mixed layer depth. CCMl-Oz includes a number of modifications to the physics subroutines including cloud prediction and radiation updates used in CCM2 [Slingo, 1989]. The model simulates full seasonal and diurnal cycles and a review of a number of standard global fields shows that the general circulation of the atmosphere is well simulated.

Table 3: Percentage areas of the 11 vegetation functional types (VFT) employed here; there limits; associated parameters; % when specified for BATS; % observed; % predicted using 5-year averages of temperature and precipitation 1 x CO2 and 2 x CO2 (instantaneous) simulations.

The Biosphere-Atmosphere Transfer Scheme (BATS), [Dickinson, 1984] incorporates a single vegetation, or canopy, layer, a multiple-layer soil scheme and provision for snow cover on the land-surface. The scheme has been subjected to stability and sensitivity tests both with the NCAR Community Climate Model [Dickinson and Henderson-Sellers, 1988; Wilson et al., 1987a] and in off-line mode [Henderson-Sellers et al., 1994; Pitman et al., 1990; Wilson et al., 1987b]. The BATS scheme has evolved as a result of these experiments so that the current version (BATS1E) which is used here [Yang and Dickinson, 1996], although substantially the same as that described in Dickinson et al. [1986], does incorporate some corrections and improvements to earlier versions described in the literature [Dickinson et al., 1993]. BATS can treat a wide range of different surface types, soil characteristics and vegetation covers. At a given grid point, a seasonally dependent fraction of surface covered by vegetation is specified; the remaining fraction is assumed to be covered by bare soil. The fractional vegetation cover varies seasonally based on the assumption of a maximum value when the temperature of the total soil profile is above 298K and decreasing in a quadratic fashion to a specified minimum value when this temperature is below 273K. (It must be recognized that this temperature-only dependence of fractional vegetation cover neglects the important effects of soil moisture availability).

Prognostic equations are solved for the temperature of a surface soil layer (0.1 m thick) and a deep soil layer (1-2 m thick) using a modification of the force-restore method [Dickinson, 1988]. This method includes exchanges of radiant and turbulent energy between the upper soil layer and the atmosphere, heat release by water-phase changes and diffusion between the top and lower soil layers. The soil heat capacity and conductivity depend on the soil texture and moisture content.

In the presence of vegetation, the temperatures of air within the canopy and the foliage are calculated diagnostically via an energy balance equation which includes canopy-ground and canopy-atmosphere radiative and sensible heat exchanges, transpiration from stomatal pores and evaporation of intercepted moisture. The transpiration rate is calculated using a resistance formulation which includes the aerodynamic resistance to fluxes of moisture and heat from the foliage and the mechanical resistance encountered by the diffusion of moisture from inside a leaf to outside (or stomatal resistance). The stomatal resistance depends on the flux of photosynthetically active radiation, leaf temperature and vapor pressure deficit and is modified to account for the root resistance to soil water uptake by the canopy but varies only between specified limits.

Predictive equations are solved for the water content in three soil layers as distinct from the two soil layers used in the soil temperature formulations [Dickinson et al., 1993]. These equations for soil moisture include the contributions of precipitation, leaf drip from the canopy, evapotranspiration and resulting moisture uptake by the roots, surface and groundwater runoff and diffusive exchange of water between soil layers [Dickinson, 1984]. Root uptake of moisture can only occur from the upper two soil layers which correspond to the two layers used in the soil temperature formulation.

The scheme incorporates snow, frozen soil water and related phase changes. A prognostic equation including precipitation, sublimation and snow melt is solved for the snow depth for each grid point. The fractional snow cover is calculated diagnostically from the average snow depth at the grid point and the roughness length of vegetation or bare soil [Dickinson et al., 1993; Dickinson et al., 1986].

For each vegetation type, vegetation albedos are specified for the ultraviolet/visible and near-infrared regions of the solar spectrum. The albedo for bare soil depends on soil color class and decreases with soil water content. It varies from 0.05 to 0.2 in the ultraviolet/visible region and from 0.1 to 0.4 in the near-infrared region. The albedos for diffuse radiation are assumed to be the same as those for direct radiation. At a given grid point, the surface albedo is obtained by averaging over vegetated, bare soil and snow-covered areas. The BATS code also includes a calculation of net carbon gain or loss by the vegetation and soil systems due to photosynthesis, respiration and decay although the photosynthetic calculation is, at present, highly simplified.

Sensible heat, water vapor and momentum fluxes at the surface are calculated from a standard surface drag coefficient parameterization. The drag coefficient for a given grid point is obtained via an average over vegetated, bare soil and snow-covered areas. The drag coefficients are expressed as a neutral drag coefficient times an atmospheric stability correction factor. The neutral drag coefficient is a logarithmic function of the height of the bottom atmospheric model level and the specified roughness length of vegetation, bare soil, water or snow. The stability correction factor is assumed to be the same for momentum, heat and water vapor transfer and is expressed in terms of the local bulk Richardson number. In order to use this drag coefficient parameterization, the bottom atmospheric level in the host model should not exceed a few tens of metres.

When coupled to a meteorological or climate "host" model, the vegetation type, soil texture and soil color need to be specified for each grid point, along with the initial snow cover, soil moisture and ground and foliage temperatures. From the host model then, BATS requires as input the ambient meteorology such as temperature, humidity and surface radiant fluxes. From these and other internally generated quantities, BATS calculates temperatures of the surface soil, deep soil, canopy foliage and canopy air, the soil moisture in three layers, snow cover and surface fluxes of momentum, heat and moisture. The surface fluxes are then fed into the momentum, thermodynamics and water vapor equations of the host model as lower boundary conditions. The continental surface climate, upon which the vegetation prediction scheme used here is a function, is critically dependent upon these fluxes which are, in turn, dependent upon the characteristics and parameterizations in BATS.

Coupling

BATS normally uses 18 vegetation types when coupled to CCM1-Oz to attempt to represent both natural and agricultural ecologies. This global vegetation classification was originally generated from the data sets developed by Olson et al. [Olson et al., 1983], Matthews [1983], and Wilson and Henderson-Sellers [Wilson and Henderson-Sellers, 1985]. A set of eleven vegetational functional types has been derived (Table 6(a)). There are sixteen parameters associated with each of these functional types when coupling to the global climate model via BATS. The vegetation model is "coupled" to the climate model by use of annually-average biotemperatures and total annual precipitation amounts.

Experiments

The experiments conducted here are designed to answer two linked questions: (i) a numerical sensitivity study of the GCM and (ii) a climatic sensitivity study. In the first, we try to ascertain the stability (or otherwise) of the GCM when its continental lower boundary is instantaneously modified and, in the second, we assess the impact on the simulated climate. These questions are intrinsically linked. Moreover, the answers may well be a function of the coupling time step selected: here, one year.

In all, six experiments were conducted to examine the role of an interactive biosphere in simulations of the climate system. Initially, three doubled-CO2 simulations were utilized:

(i) a "standard" instantaneous doubling experiment using a specified vegetation distribution appropriate to present day observations, in which the atmospheric concentration of CO2 is raised to 660 ppmv and the GCM allowed to equilibrate;

(ii) a fast, transient experiment, in which the atmospheric concentration of CO2 is increased gradually over 35 years until the amount has doubled and then the climate model is allowed to equilibrate with this CO2 level held constant and

(iii) a doubled CO2 experiment with prescribed vegetation cover (fixed at present-day distributions) and continued from the end of case (i). In addition, a control (fixed vegetation) and interactive vegetation for 1xCO2 experiments were performed. The fast, transient, doubling uses a 2% per annum compound rate of increase justifiable [Houghton et al., 1990; Houghton et al., 1992; Houghton, 1991; Policy Makers Summary, 1990] if recent estimates of projected emissions are taken into account. The transient experiment permitted an evaluation of the final impact on climate of different rates of vegetation change (slower in the transient than in the instantaneously doubled CO2 experiment).

Model Intercomparison

The results of the CCMOz model were compared to those of five others as an indication of relative model performance and sensitivity. The differences betwen the six GCMs are numerous: there are some differences in the spatial resolutions both vertical and horizontal, the land surface hydrology is represented in various ways, the sea surface temperatures are either prescribed or simulated, and the convective schemes employed vary. In addition, the length of the runs and initial conditions differ. It must be noted that two models have a higher resolution: the BMRC model at R21 and CCM2 for which a range of resolutions are available [Williamson et al., 1995]. It was decided to spacially aggregate the BMRC results to the R15 grid as it would be too difficult to compare the results obtained by the R21 version of the BMRC with the other model results. From controlled simulations of these models (i.e., for the present day climate), monthly surface temperature, precipitation, and total cloud fraction (assuming random overlap) are averaged over the longest period readily available, that is, 20 years for CCM10z, CCM1, and CCM0 and 10 years for CCM1W, BMRC, and CCM2. The comparison of the climate control simulations with observations shows that the performances of the climate models are rather unequal.

Comparison of Vegetation Distributions

The approach employed in this study consists of comparing the vegetation distributions obtained using simulated climates and the LW/ISCCP data sets. The spatial resolutions of the different maps of vegetation have to be identical to enable this comparison. Therefore either the computed climatologies have to be interpolated to a higher resolution [Claussen and Esch, 1994] or the observed climatology needs to be spatially aggregated to a courser resolution. Both approaches have their drawbacks, but since the resolutions of the GCMs used here are coarse (7.5¡ by 4.5¡), it was deemed inappropriate to interpolate these computed climatologies. Thus the approach retained was to spatially aggregate the observed climatology to the R15 resolution of the GCMs.

The vegetation models driven by the aggregated observed climatology generate two maps of vegetation which will be considered throughout this paper as "reference" maps, as opposed to the maps computed from simulated climatologies (hereafter called "GCM" maps). The expression "reference" maps is used despite the fact that the vegetation distributions represented in these "reference" maps may differ from the observed distribution of natural ecosystems. Indeed, this expression is employed solely to express the fact that the climatology used to generate these maps of vegetation is observed and not simulated. The fact that some discrepancies exist between the distribution of the natural ecosystems and the simulated vegetation can only be related to the reliability of the vegetation models. The vegetation models used here perform relatively well (Fig. 9).

Comparing several global vegetation distributions requires that a methodology for analyzing the results be developed. The results presented in the section are displayed as maps of vegetation and frequency distributions of biomes. A statistical tool will be used for the evaluation of the agreement between maps and finally the climatic causes that lead to discrepancies in the vegetation distribution will be identified. It should also be noted that the biomes were grouped in panels for the presentation of the graphs: tropical rain forests (i.e., biomes 1, 2, and 3), forests form temperate regions (i.e., biomes 4 and 5), forests from higher latitudes (i.e., biomes 6 to 10), drought tolerant vegetation from warm regions (i.e., biomes 11, 12, and 15), drought tolerant vegetation from cold regions (i. e., biomes 13 and 16) and finally subpolar and polar biomes (i.e., biomes 14 and 17).

Figure 9: Biome distributions: (a) data from Olson et al. [1983] used to validate BIOME-1 in the work of Prentice et al., [1992]; (b) simulated by BIOME-1 and (c) simulated be Holdridge scheme. The map resolution is 0.5¡ by 0.5¡, and the climate data sets are used to drive the vegetation models are from Legates and Willmott [1990a, b] and Rossow and Shiffer [1991]. The continental areas colored in white in Plate 1a correspond to agricultural areas.

The maps of vegetation simulated by BIOME-1 are presented in Figure 10. Large discrepancies can be seen between the maps of vegetation computed from simulated climatologies and the "reference" map, particularly in certain regions (e.g., Australia, central Asia, Africa).

Assessing the Sensitivity of a Global Model to "Responsive" Vegetation

A coupled vegetation-climate model

Figure 11 compares the distributions of vegetation functional types derived for CO2 (Fig. 11(a)), at the end of the instantaneous 2xCO2 simulation (Fig. 11(b)) and at the end of the transient CO2 simulation (Fig. 11(c)). Note there is a considerable increase in the area predicted as being appropriate for agriculture (VFT cp) whereas the total area of tropical forest (VFT eb), itself a contributor to atmospheric CO2, changes very little. Comparing the response of the vegetation to instantaneous doubling to that produced during a fast, transient doubling experiment shows similar trends in vegetation distribution, but the instantaneous doubling results in a greater area of 'crop' at the expense of deciduous needle-leaf tree (VFT dn). However, the results of this interactive evaluation of terrestrial vegetation distribution and dynamics in a warming world must be treated with considerable caution; in particular because there is no CO2 fertilization effect on the predicted vegetation distribution. At the minimum, however, the vegetation functional types "prediction" scheme delineates climate zones hospitable to the identified vegetation types. The general impact of climate on vegetation is similar to earlier studies [Emanuel et al., 1985a; Emanuel et al., 1985b; Leemans, 1989].

Figure 10:Biome distributions simulated by BIOME-1. The climatologies used are computed by the GCMs (a) CCM10z, (b) CCM1, (c) CCM0, (d) CCM1W, (e) CCM2, and (f) BMRC and are (g) observed. The observed climate data sets are LW/ISCCP and are spatially aggregated to R15 resolution, that is, 4.5¡ latitude by 7.5¡ longitude (see section 2.3). See Figure 9 for color keys.

These experiments also allowed investigation of whether changes in vegetation form affect the GCM's simulated climate. The strongest signal might be expected to be in the land-surface climate parameters. Indeed, Henderson-Sellers [1993a] found that the introduction of the interactive biosphere for a present-day simulation resulted in an increase in evaporation of up to 5 W m-2 over the continental surface and increases in absorbed solar radiation and surface temperature (up to 1.5¡C). Figure 12 shows the evaporation over the continental surfaces for the 5 years of two paired sets of model runs at 1xCO2 and 2xCO2. In each case, the simulation which incorporates an interactive vegetation scheme gives rise to increased evaporation over the continents which is particularly marked in northern hemisphere summer.

Figure 11: a) Vegetation functional types' distribution predicted, off-line, form ensemble means of biotemperature and precipitation from 5-years of the 1 X CO2 control climate. b) Vegetation functional types predicted at the end of the last year (year 15) of the instantaneous doubled CO2 simulation. c) As for (b) except from the last year (year 45) of the fast, transient doubled CO2 simulation.

Table 4 summarizes some climate statistics resulting from the three 2xCO2 experiments as differences from, or percentages of, the 1 x CO2 control. The rootzone soil moisture appears to be sensitive to the inclusion of an interactive biosphere but this response is, in part, the result of changes in rootzone soil depth which is itself a function of the vegetation functional type. There is almost no difference in this variable between 1 x and 2 x CO2 when the BATS-prescribed vegetation is used. Planetary albedo decreases as in most doubled CO2 experiments, partly as a result of decreased cloud amount and partly because of the ice-albedo feedback. Here, a small component of the decrease in planetary albedo over the land is the result of the decrease in ice-cap area (Fig. 11) which can be modified to one of the other vegetational functional types if climate warms. The sea-ice area is found to be sensitive to both the inclusion of an interactive biosphere and the mode of CO2 increase, the sea-ice area remaining larger (i.e. the smallest decrease) in the case of the transiently increasing CO2.

Figure 13 compares the zonally and multi-year averaged difference between 2xCO2 (fixed or interactive vegetation) and 1xCO2 (fixed vegetation), Stevenson screen air temperature and total precipitation for January and July. Large temperature differences are seen in July in high latitude locations where sea-ice differences occur whereas tropical precipitation is sensitive to the incorporation of interactive vegetation. In common with most simulations of greenhouse climates, the temperature change signals are statistically significant (the lowest curves on the temperature graphs in Figure 13 are of one standard deviation) while the precipitation changes are generally not significant. On the other hand, the differences between fixed and interactive vegetation simulations are always smaller than the model's natural variability (Fig. 12).

Figure 12. 5-year time series of evaporative flux (W m-2) over the continents only for 1xCO2 with fixed vegetation (dash) and interactive vegetation (solid).

The evaporation from the continents is noticeably larger when the vegetation is interactive than when it is prescribed at the present-day distribution and, in the annual mean, the global (oceans as well as land) evaporation is greater when the vegetation is interactive. Preliminary analysis indicates that the vegetation changes may be prompting a feedback between the atmosphere and the oceans, via low-level convergence changes, which further increases global evaporative flux. Specifically, including the interactive biosphere in both 1 x CO2 and 2xCO2 simulations tends to enhance the Hadley circulation.

Doubling of CO2 (without any change at the land surface) results in a weaker winter branch (decreases between the equator and 35¡) but a slightly stronger summer branch (increases from the equator to 35¡) indicating enhanced circulation. The presence of an interactive biosphere produces similar changes in the meridional circulation. The winter branch of the Hadley circulation is diminished and the summer branch is enhanced. Differences at higher latitudes are much smaller and harder to identify clearly. Changes in meridional circulation induced by allowing the biosphere to respond to the climate are similar in character and of the same magnitude as changes induced by doubled CO2. In both cases, there is stronger low-level flow over summertime tropical oceans which has the potential to operate as a positive feedback by increasing evaporation.

Table 4: Differences in climatic parameters derived from the three doubled CO2 simulations (5-year ensemble means), the 1 x CO2 control integration (i) after instantaneous doubling with the interactive vegetation; (ii) after a fast, transient doubling of CO2 with the interactive vegetation and (iii) after instantaneous doubling of CO2 but reverting to the present-day prescribed vegetation. In all cases, the land area is 34.17% of the globe. Screen temperatures are differences in Kelvin and all the parameters are given as % of the 1xCO2 values. Cloud amount and rootzone soil moisture are for 00Z only, all other values are diurnal averages. Sea-ice is given for the globe and rootzone and soil water for the land entries in the lowest rows.

In summary, it is doubtful if vegetation and climate are ever in equilibrium. Certainly, these experiments, which couple a very simple (eleven class) representation of vegetation functional form to a fairly standard global climate model, show feedbacks operating in both directions: the climate alters the vegetational form and changing vegetation modifies the climate.

We conclude that an interactive vegetation modifies the climatic change due to doubling of atmospheric CO2. The most direct changes are enhanced continental evaporation which prompts intensification of the atmospheric circulation in the tropics and, in turn, enhanced oceanic evaporation. This second affirmative is a trying result for those content to generate vegetation post facto. At a minimum, it means their diagnosis is incomplete since their analysis neglects feedbacks between climate and vegetation.

There is no one, simple solution to the "best means" of coupling models of terrestrial vegetation into global climate models. Sensitivity studies of the types described here are essential first steps towards "dynamic" modelling of global change.

Figure 13. January and July differences (2x CO2 -1xCO2) in zonally averaged (land, ocean and sea-ice), screen temperature (K), and total precipitation (% of 1x CO2) with interactive and fixed continental vegetation. All differences are with a 1x CO2 simulation with fixed vegetation. For screen temperature, one standard deviation of the 1x CO2 (control) values is shown (lowest curve).

III. GCM SENSITIVITY TO STOMATAL RESISTANCE

Stomatal resistance and elevated atmospheric CO2

In elevated CO2 conditions, plants tend to exhibit increased stomatal resistance [Bazzaz and Fajer, 1992; Cure, 1985; Eamus and Jarvis, 1989] so that water lost through transpiration is decreased for the same uptake of carbon in photosynthesis. Although this response varies widely as a function of plant type, availability of sunlight, soil water and nutrients and inter-species competition, there are many examples of observations of CO2 "fertilization" enhancing plant water use efficiency.

It has been shown in numerous numerical modelling experiments that decreased evaporation from the continents can cause changes to the regional climate [Dickinson and Henderson-Sellers, 1988; Henderson-Sellers, 1993b; Nobre et al., 1991] and to the global climate [Shukla and Mintz, 1982]. Generally, the imposition of stomatal resistance on the parameterization of continental evapotranspiration causes decreases in the latent heat flux over the land and reductions in both the convective and large-scale precipitation [Blondin, 1989]. Other experiments in which the effect of stomatal resistance has been introduced also show decreased evaporative fluxes, warmer near-surface temperatures, reduced near-surface humidities and, in some cases, an increase in the diurnal cycle of the near-surface air temperature [Sellers et al., 1989]. Pollard and Thompson [1994] examined the sensitivity of the GENESIS global climate model to increased stomatal resistance as it affects the present-day climate simulation. They also find decreased evapotranspiration and near-surface air warming and, in addition, significant increases in precipitation and runoff.

It is reasonable to hypothesize that, as CO2 increases in the atmosphere, plant stomatal resistance to transpiration will also increase. This increase is most likely to cause a decrease in the transpiration through plants (Fig. 14(a)). These changes may both compound and confound the direct changes to the climate caused by the CO2 increase itself (Fig. 14(b)). A few studies have tried to evaluate the possible impact on surface hydrology of increased stomatal resistance. Idso and Brazel [1984] calculated that including decreased stomatal resistance in calculations of future stream flow changed an expectation of a 40-70% decrease into a 40-60% increase. Wigley and Jones [1985] were more cautious in their predictions: concluding that for low runoff ratios (runoff/precipitation), small changes in precipitation could cause large changes in runoff but noting that both the magnitude and the direction of these changes in runoff are strongly dependent upon the direct effect on transpiration caused by increased stomatal resistance. No studies have yet assessed the combined impacts of doubling atmospheric CO2 and plant stomatal resistance in a global climate model.

Figure 14a: Schematic representation of the relationship between evapotranspiration shown as the ratio of evapotranspiration to potential evapotranspiration as a function of canopy resistance for three different canopy types with large (short grass), intermediate (tall grass) and small (forest) aerodynamic resistances (Monteith, 1981).

Figure 14b:. Schematic and incomplete representation of the possible responses to increased stomatal resistance in plants for the three cases investigated: heavy arrows for off-line sensitivity; solid arrows for present-day climate and dashed arrows for 2x CO2 climate.

Here, the sensitivity of a global climate model to these two effects is investigated. The term "stomatal resistance" is employed to describe the parameterization of the effect of vegetation on transpiration of moisture from the soil to the atmosphere. Although a more correct term might be canopy conductance as it applies to a "big leaf" land surface scheme, stomatal resistance is retained here in line with the current terminology in land surface modeling [Henderson-Sellers, 1993a]. The climate model and land surface scheme used are described in Section 4.2. A series of off-line sensitivity tests are discussed in Section 4.3 and Sections 4.4 and 4.5 describe the results of doubling stomatal resistance in the global climate model at 1 x and 2 x CO2 levels respectively.

 

The "doubled" stomatal resistance experiment

The stomatal resistance in BATS is formulated on the basis of the parameterization proposed by Jarvis [Jarvis, 1976]

where gs is leaf stomatal conductance (m s-l), is the leaf stomatal resistance (s m-l), F is the vector flux of photosynthetically active radiation (W m-2), n is the normal vector of leaf orientation and are adjustment factors to account for the effects of temperature, leaf water potential and vapor deficit stresses.

In controlled glass-house experiments, plants have been shown to increase their stomatal resistance by a factor of between 1.5 and 2 in a doubled CO2 environment [Morison and Gifford, 1984]. However, the response differs between C3 and C4 plants and varies widely as a function of temperature, light levels, water availability and soil nutrients [Eamus and Jarvis, 1989; McMurtie and Wang, 1993]. In the simulations described here, a simple means of increasing stomatal resistance in the BATS scheme has been employed: the calculated stomatal resistance is doubled in the penultimate line of the stomatal resistance calculation. However, the final line of this subroutine is a capping calculation which forces the stomatal resistance employed in BATS to be less than or equal to the maximum prescribed stomatal resistance. Thus, in some cases, the result of the "doubling" of stomatal resistance employed here may not be an increase by a factor of two.

Doubling stomatal resistance is not expected to be a realistic representation of the future behavior of plant response to increased levels of atmospheric CO2 for a number of reasons. It can be argued that a factor of 2 is an underestimate of the stomatal closure because the intercellular resistances are larger than stomatal resistances [Rosenberg, 1981] or that its impact is a significant overestimate because CO2 fertilization will increase the net primary productivity and hence increase transpiration [Riebsame et al., 1994]. Here we accept the observations that a doubling is a reasonable representation of the response of plants to CO2 elevations [Eamus and Jarvis, 1989] around double the pre-industrial level and seek to place the simulated responses in the context of other factors controlling climatic, and particularly land surface climate, response to greenhouse gas increases.

 

Figure 14(a) illustrates typical variations of E/Ep with canopy resistance for low and high roughness canopies. Here the ratio between non-potential and potential evapotranspiration can be expressed as

where E is the non-potential evapotranspiration, Ep is the potential evapotranspiration, is the slope of the saturation vapor pressure curve, is the psychometric constant, is the canopy resistance (stomatal resistance divided by LAI) and is the aerodynamic resistance. When the roughness length is small (e.g. grassland), the aerodynamic resistance is large and even significant values of the canopy resistance do not reduce the evapotranspiration much below the potential rate. In contrast, when the roughness length is large (e.g. forest), the aerodynamic resistance is small and an increase in the canopy resistance may reduce evapotranspiration well below its potential rate. In other words, forest evapotranspiration is likely to be much more sensitive than shorter vegetation to changes in the stomatal resistance.

 

The doubled CO2 experiments

Two pairs of GCM simulations have been conducted representing control (1xCO2) and enhanced greenhouse (2xCO2) climates. The 2xCO2 simulations represent instantaneous doubling of the atmospheric CO2 content after which the global climate is allowed to achieve equilibrium. The simulations reported upon here were conducted at the end of a pair of control and 2xCO2 experiments so that the stomatal resistance alteration was a further instantaneous change after the climates were equilibrated. Results from the basic control simulation are available for 25 years while the perturbation simulations are for 10 years for 2xCO2 but standard stomatal resistance; 6 years for 1xC02 and doubled stomatal resistance; and 5 years for 2xCO2 and doubled stomatal resistance. Although it would appear to be preferable to have at least 10 years in all the simulations, CPU constraints did not permit this and the analysis of the off-line experiments indicates that BATS is adequately equilibrated to the imposed change in in much less than 5 years.

CCM1-Oz has a relatively low sensitivity to doubled CO2 conditions as can be seen in Figure 15, which compares the globally and annually averaged temperature and precipitation (which is synonymous with evaporation at equilibrium) changes from these experiments with those reported by the IPCC [Houghton et al., 1990; Houghton et al., 1992]. With the standard stomatal resistance in BATS, the effect of doubling CO2 is to increase surface air temperatures by 2.67¡C and total precipitation by 7.73%. This can be compared to the impact of doubling stomatal resistance in the present day climate which increases temperatures by 0.13¡C but decreases precipitation by 0.82%. If the differences are constructed with doubled stomatal resistance in control climate and at 2xCO2 they become 2.59¡C and 5.88% respectively. However, the most likely global response is from lxCO2 and 1x , to 2xCO2 and 2x which produces global increases of 2.72¡C and 5.01%.

Figure 15: Sensitivity of CCM1-Oz) to doubled CO2 and doubled stomatal resistance as compared to the IPCC results. The three points lying in the IPCC cluster are for the standard stomatal resistance and for doubled stomatal resistance at 2xCO2 and for 2xCO2 and 2xrs minus 1xCO2 and 1xrs. The two points for which there is a decrease in evaporation/precipitation show the differences between doubled stomatal resistance and the standard value both for fixed climatic conditions (i.e. at 1xCO2 and 2xCO2 respectively).

Off-line sensitivity of BATS to doubled stomatal resistance

Experimental design

The BATS land-surface scheme can be examined in an off-line mode with prescribed forcing, as described by Wilson et al. [1987a; 1987b] and more recently in the context of the PILPS experiment by Pitman et al. [1993a]. Here we use the PILPS forcings for tropical forest, grassland and tundra and also the observed data set from Cabauw to examine the sensitivity of this land-surface scheme to an imposed increase in stomatal resistance.

The atmospheric data used in this study were obtained from the NCAR CCM1. They represent a temperate grassland (0¡E, 52¡N); a tropical forest (60¡W, 3¡S); and tundra (95¡W, 65¡N) environments. In addition, observational forcing data from Cabauw (4¡56'E, 51¡58'N) are also used. The forcing data include downward short-wave radiation; downward long-wave radiation; precipitation; air temperature; wind speed; surface pressure and specific humidity. To perform these off-line sensitivity tests with BATS, the parameter values listed in Table 5 are used, based on Dickinson et al. [ 1993].

In all the off-line experiments the following initialization for January 1st was used: all soil moisture is initialized as 50% of full capacity (all layers, liquid or frozen are half full); the canopy is initialized as 50% full; snow mass and snow age are initialized as zero and all temperatures are initialized equal. Using the four sets of one-year forcing data for the different sites, BATS1E is run with a 30 minute time step until equilibrium is reached. In these experiments, the model was run for at least 10 years and this ensured equilibrium.

Table 5: Parameter values for grassland, tropical forest, tundra and Cabauw.

The off-line mode does not allow feedbacks to the atmospheric forcing, which is prescribed and independent of any changes in the surface fluxes. Thus the responses are restricted as shown by the heavy arrows in Figure 14(b). As BATS does not incorporate a response to altered atmospheric CO2 amounts, there is no off-line experiment comparable to a 2xCO2 simulation with the GCM. For these reasons, and because of the potential for synergy between changes in atmospheric and land surface parameters, these off-line tests are used here as indications of possible responses but not analyzed for the full range of sensitivities [Henderson-Sellers, 1992].

Off-line sensitivity to doubled stomatal resistance

The monthly average responses from the four vegetation type simulations in the off-line mode are shown in Figures 14-17. In all four cases, evapotranspiration decreases as a result of the reduction in plant transpiration; sensible heat increases but by a smaller amount; the effective surface temperature is marginally increased and runoff is increased (Table 6). It can be seen that, as anticipated, the largest response occurs in the case of tropical forest where the aerodynamic resistance is relatively small (Fig. 14(a)). Indeed, it is only in the tropical forest that the effective temperature is noticeably increased (Fig. 16). The other vegetation types show much smaller responses although these can be clearly seen where the annual cycle is small .

Table 6: Annually-averaged differences in the off-line simulations between doubled stomatal resistance experiments and the control experiments where T is the effective surface temperature (K); E is the evapotranspiration (Wm-2) and R is the total runoff (mm yr-1).

Although on a seasonal basis the impact in the off-line simulations is less than 15 W m-2 reduction in evaporative flux (Table 7), individual days can exhibit much larger effects. For example, Figure 20 shows the evaporative and sensible heat fluxes from the tropical forest, tundra and Cabauw simulations for single days (15th January, 15th July and 15th April, respectively). In all three cases, the overall effect is in line with the seasonal and annually averaged results. The differences are restricted primarily to the day-time and can be large on individual days. For example, in the tropical forest on 15th January the maximum reduction in evapotranspiration is more than 150 W m-2 with an almost commensurate increase in sensible heat flux. In the tundra and Cabauw simulations, the reductions in the evaporative flux for the doubled stomatal resistance are smaller but can exceed 50 W m-2 for short periods of time (Figs. 20(c) and 20(e)).

Figure 16:Monthly averaged responses from the off-line simulations of a tropical forest to doubling the stomatal resistance (a) effective surface temperature (K), (b) evapotranspiration (W m-2), (c) sensible heat flux (W m-2), (d) total runoff (mm per month).

Figure 17: As for Figure 16, but for grassland.

Figure 18: As for Figure 16, but for tundra.

Figure 19: As for Figure 16, but for Cabauw.

Figure 20: Single day realizations from the off-line simulations showing surface fluxes (W m-2) for (a) tropical forest evapotranspiration on 15 January, (b) tropical forest sensible heat on 15 January, (c) tundra evapotranspiration on 15 July, (d) tundra sensible heat flux on 15 July, (e)Cabauw evapotranspiration on 15 April and (f) Cabauw sensible heat flux on 15 April.

Although very different in nature (stand-alone vs. fully interactive) these off-line results are in general agreement with the global responses at constant values of CO2 seen in Figure 15. Both for 1 x CO2 and 2 x CO2 there is a Stevenson's screen temperature increase (<0.2¡C) and an evaporation decrease (<3% of global mean conditions in the control case). Thus both off-line and global model results indicate that there is likely to be sensitivity to altering the value of the stomatal resistance.

These off-line simulations are included here only as indications of the type of surface changes likely to occur in response to imposed stomatal resistance doubling. The Penman Monteith relation (Equation (2)) shows that the largest response will occur in forest areas where the aerodynamic resistance is small and latent heat fluxes are likely to be reduced while sensible heat and runoff increase. The annually averaged decreases in latent heat flux found in the off-line simulations of about ~15 W m-2 conceal much larger differences in monthly and daily fluxes. These disturbances could be large enough to prompt regional scale atmospheric changes [Henderson-Sellers et al., 1993; Nobre et al., 1991]. In the global climate model experiments described in the next two sections, the feedbacks to the atmosphere are incorporated so that the effects of doubled stomatal resistance only and doubled stomatal resistance in a greenhouse-warmed climate can be assessed.

Global model response to doubling stomatal resistance for the present-day climate

Using CCM1-Oz the stomatal resistance was "doubled" in the same manner as in the off-line experiments. Table 4 lists the globally averaged and land-only averaged responses in selected fields. The globally and annually averaged impact of this for the present day climate (i.e. at 1 x CO2) can be seen in Figure 15: Stevenson's screen temperature is increased by +0.13¡C and evaporation (and precipitation) decreased by -0.82% of the control climate values.

Table 8(a): Response to doubling stomatal resistance at 1 x CO2 for January, July and the annual mean for the globe and land areas only.

As can be seen from Table 8(a), the general effect of doubling stomatal resistance is to decrease evaporation and precipitation and to increase the Stevenson's screen temperature. Table 8(b) shows the differences as percentages of the control values for the land areas only. With the exception of the sensible heat flux and the total runoff, all the values are largest in July. There seem to be no seasonal changes of sign i.e. evaporation and precipitation always decrease, although the latter to a lesser extent, and all the other variables increase following the imposition of doubled stomatal resistance.

Table 8(b): Land only differences at 1 x CO2 for January, July and the annual mean as a percentage of the 1 x rs value.

Zonal plots (Figure 21) and zonal differences (Fig. 22) show two locations of up to 25 W m-2 decreased evapotranspiration around 5¡N and 45¡N (Fig. 21(a)); sensible heat fluxes increase particularly north of ~30¡N (Fig. 21(b)); and screen temperatures tend to increase roughly north of 30¡N (Fig. 21(c)). Geographical distributions shown in the global maps (Fig. 23) highlight these areas and, especially, the boreal forest vegetation type (Fig. 23(a)). Indicate that the decreased evapotranspiration and increased air temperatures are statistically significant across most of the land masses centered around 45¡N (Figs. 22(a) and (b)). There are also smaller areas of similar changes in eastern Brazil and tropical Africa. Precipitation differences are noisy (Fig. 23(c)) and show no significant changes (Fig. 24(c)) but there are statistically significant increases in the total soil moisture in tropical South America and boreal North America and parts of Eurasia-Asia (Fig. 24(d)). The changes in runoff seen in Figure 23(e) are therefore the combined response to increased soil moisture and variable precipitation changes. Only in western Brazil do these result in statistically significant changes (an increase) in surface runoff following the doubling of stomatal resistance (Fig. 24(e)).


Figure 21: (a) Zonally-averaged continental evapotranspiration (W m-2) for July at 1x CO2 showing standard and doubled stomatal resistance (b) as (a) but for sensible heat (W m-2) (c) as (a) but for Stevenson screen temperature (K).

The response to doubling the stomatal resistance in the global model is roughly commensurate with the off-line experiments described in Section 3. As anticipated (Fig. 14(a)), the largest climatic changes occur where trees predominate. In these areas (tropical and, especially the boreal, forests) a large decrease of evapotranspiration increases surface temperatures, sensible heat fluxes and total soil moisture and can lead to increased runoff. The largest areal responses seem to occur in the latitude band 44¡-58¡N.

Figure 25 shows the evaporation and surface air temperature for the continental grid points between 44¡ and 58¡N for all four combinations of single and double stomatal resistance and atmospheric CO2 . The evaporation rates differ in the summertime and are primarily controlled by the imposed stomatal resistance: the two cases with 2x being 1015 W m-2 less than for 1 x (Fig. 25(a)). In the case of Stevenson's screen temperature, however, doubled CO2 warms throughout the year while the effect of imposing differing stomatal resistance influences temperatures (via evaporation) only in the summertime (Fig. 25(b)). In summer, the highest temperatures occur with 2xCO2 and 2x and the lowest with 1xCO2 and lx whereas the largest evaporation rates occur for 2xCO2 and lx and the smallest for 1xCO2 and 2x .

Figure 22:As for Figure 16 except differences between doubled and standard stomatal resistance at 1x and 2x CO2 .

Figure 26 shows four 5-year time series of evaporation from the land areas of the globe as monthly means. It can be seen that doubling stomatal resistance at 1xCO2 decreases the peak evaporation and, marginally, the values through the year although the lowest values are not really affected. At doubled CO2 , the evaporation is greater than at lxCO2 . The impact of doubling stomatal resistance seems also to be larger than at 1xCO2 . This is particularly true for the lowest values through the year, which occur in the Northern Hemisphere winter-time, although the peaks are also markedly decreased.

Global model response to doubling stomatal resistance for doubled CO2

As can be seen in Figures 13 and 25, CCM1-Oz has a relatively small sensitivity to doubling atmospheric CO2 . Table 9a summarizes the global and land-only sensitivities for January, July and the annual average to doubling the stomatal resistance. It can be seen that, overall, the responses are similar to those seen at lxCO2 (Table 8(a)). Evaporation decreases, and the decrease is larger than at 1xCO2 ; precipitation decreases also, but not in July, Stevenson's screen temperature increases as does sensible heat; and total runoff increases except in January. The major difference seen in these land-area averages at 1xCO2 (Table 8(b)) and 2xCO2 (Table 9(b)) is that the total soil moisture decreases in the enhanced greenhouse climate but increases in present-day conditions following the doubling of stomatal resistance.

FIGURE 23 a,b,&c
FIGURE 23 d&e

Figure 23. (a) Geographical distribution of the difference (double - standard stomatal resistance) evapotranspiration (W m-2) for July for 1x CO2 . (b) As (a) for Stevenson screen temperature (K) (c) As a for total precipitation (mm d-1) (d) As (a) for total soil water (mm). (e) As (a) for a total runoff (mm).

Figure 24: As Figure 23 except T values at 95% significance levels.

Table 9(c) shows the percentage changes between 2xCO2 and 2x minus 1xCO2 and 1x : the pair of simulations which most closely represents the likely global changes because, as CO2 increases in the atmosphere, it is likely that stomatal resistance will also increase [Leuning et al., 1993]. The combined result of doubling CO2 and stomatal resistance is to decrease evaporation (marginally) but to increase precipitation over the land areas. The Stevenson's screen temperature rises as does sensible heat and total runoff but total soil moisture is decreased. The decrease in soil moisture is in agreement with the IPCC results [Houghton et al., 1990; Houghton et al., 1992] although the land surface scheme used here (BATS) is very much more complex than those employed in the IPCC models [Pitman et al., 1993b]. The increase in total runoff is also in agreement with earlier studies that reviewed the possible impact on stream flow of increased CO2 and increased stomatal resistance [Wigley and Jones, 1985]. Idso and Brazel [1984] found that a previously calculated reduction in CO2 of 40-70% following doubling of atmospheric CO2 could be changed to an increase of 40-60% in runoff if the direct effects of increased stomatal closure were also considered.

Figure 25: Seasonal variation in evapotranspiration (a) and surface temperature (b) for the land points in the zone 44.44¡ N to 57.77¡ N for 1xCO2 and 2x CO2 and standard and doubled stomatal resistance.

Figure 26: Five year time series of monthly values of evaporation from the land areas s a function of the amount of CO2 in the atmosphere and the imposed stomatal resistance.

Comparison between the 1xCO2 and 2xCO2 responses to doubling the stomatal resistance can also be seen in Figure 22. The response is much larger at 2xCO2 in the latitude zone 44¡N to 58¡N than elsewhere. This suggests that the only location/vegetation type where the impacts of doubling atmospheric CO2 and doubling stomatal resistance are synergistic is in the boreal forests. Elsewhere the effects can act in opposite directions.

Table 9a: Response to doubling stomatal resistance at 2 x CO2 for January, July and the annual mean for the globe and land areas only.

Table 9b: Land only differences at 2 x CO2 for January, July and the annual mean as a percentage of the 1 x rs value.

Table 9c: 2 x CO2 and 2 x rs minus 1 x CO2 and 1 x rs land only differences for January, July and the annual mean as a percentage of 1 x CO2 and 1 x rs control climate.

The geographical distributions for differences prompted by doubling stomatal resistance at 2 x CO2 for July (Fig. 28) can be compared with their 1 x CO2 counterparts (Fig. 23). The evapotranspiration response is slightly weaker at 2xCO2 (Fig. 28(a)) and the areas of statistically significant decrease are much less extensive than at 1 x CO2 (Fig. 27(a) vs. Fig. 24(a)). In contrast, the areas of temperature increases are somewhat larger at 2xCO2 and the temperature changes slightly greater (Fig. 28(b)). The regions of statistically significant increased temperatures (Fig. 29(b)) are shifted pole wards in the northern hemisphere but the tropical continents show no statistically significant temperature changes (Fig. 29(b) vs. Fig. 24(b)). As for 1xCO2 , the precipitation changes are noisy and not statistically significant. However, although soil moisture exhibits both increases and decreases (Fig. 28(d)), the regions of statistically significant increases are more widespread than at 1xCO2 and statistically significant decreases also occur (Fig. 29(d) vs. Fig. 24(d)). There are few coherent regions of surface runoff change at 2xCO2 (Figs. 26(e) and 21(e)).

Summary and implications of RICE for greenhouse simulations

Sensitivity to increased stomatal resistance

Pollard and Thompson [1994] found for the present-day, that the climatic effects of a uniform doubling of stomatal resistance in the GENESIS model were largest in equatorial South America and in certain regions of northern boreal forests in summer (south of Hudson Bay and in central and eastern Siberia). These regions experienced a surface warming of ~2¡C, soil-moisture drying of ~0.1 relative to saturation, precipitation increases of ~5 mm day-l in the tropics and an order of magnitude increase in runoff. Moreover, changes were found to be statistically significant (several times the uncertainty in the estimates of the means due to interannual variability).

In the experiments reported here, the largest responses were found in the boreal and, to a lesser extent the tropical, forests. The latitude zone from about 44¡N to 58¡N shows very large, and statistically significant, responses in the surface energy and hydrological budgets following an imposed doubling of stomatal resistance. This relative response was anticipated from the Penman-Monteith relationship (Equation (2)) which, on differentiation with respect to , shows a much greater sensitivity of E/Ep for small rather than for larger ra: i.e. forests, which have, relatively, smaller aerodynamic resistances exhibit greater sensitivities to imposed changes in stomatal resistance as shown in Figure 14(a).

Figure 27:As Figure 21 but for 2x CO2 .

The 1xC02 results were consistent with the off-line experiments conducted with BATS: evapotranspiration decreased by up to ~15 W m-2; temperatures increased by over +2K; sensible heat fluxes increased by +15 W m-2. In addition, the soil moisture was increased in some locations which was often associated with increased runoff. The most noticeable difference between the off-line and GCM results is the smaller response in the case of tropical forests in the climate model simulations [Henderson-Sellers, 1992]. Although the impact of doubling stomatal resistance can be seen in parts of South America, these changes are very much less extensive than in the boreal forests.

FIGURE 28 a,b,&c
FIGURE 28 d&e

Figure 28. As Figure 23 but for 2x CO2

The combined impact of greenhouse gas and stomatal resistance increases

Although there have been many attempts to use global climate models to predict future climates as CO2 and other greenhouse gas concentrations increase [Houghton et al., 1990; Houghton et al., 1992], the results reported here are the first results from a global climate model in which both CO2 concentrations and plant stomatal resistance were increased. When the imposition of doubled stomatal resistance was combined with an instantaneous doubling of atmospheric carbon dioxide, the global response was rather small (Fig. 15): total annual precipitation was slightly reduced c£ the "standard" greenhouse experiment but annual mean temperature increase was little affected. However, comparison of the continental responses in Tables 8(b) and 9(b) with those in Table 9(c) show that the combined effect of CO2 and increases may be different from either imposed separately. In all cases, evaporation decreases and temperatures, sensible heat and runoff increase but the magnitude of the responses differ and this combines to produce different impacts on, for example, total soil moisture.

Zonal responses to doubling stomatal resistances show similarities at 1 x and 2xCO2 (Fig. 22) with the most marked responses occurring between about 44¡-58¡N. In this latitude zone, which is predominantly boreal forest in the BATS vegetation distribution, surface air temperature is highest when both CO2 and are doubled (Fig. 25(b)) but the combined effect of 2xCO2 and 2 x is to greatly reduce summer evaporation as compared with the standard stomatal resistance in either the current or greenhouse climates (Fig. 25(a)).

Figures 28(a) and (b) show the statistically significant areas of response in July for 2xCO2 and 2x minus 1xCO2 and 1x . The most noticeable difference between 1xCO2 and 2xCO2 is that the boreal response to increased stomatal resistance is increased but the tropical response decreased. Although increasing CO2 intensifies the global hydrological cycle producing more precipitation and evaporation (Fig. 15), the land areas of the northern mid-latitudes are so strongly affected by the increase in stomatal resistance that there are large regions of evaporation decrease (Fig. 30(a)). The same latitude band (44¡-58¡N) also shows statistically significant increases in total soil moisture despite the global decrease (Table 5(c)). Apparently, incorporating a plausible increase in stomatal resistance in a complex land surface scheme in a greenhouse simulation can reverse a major conclusion of the IPCC Science Report [Houghton et al., 1990; Houghton et al., 1992] that soil moisture "decreases over northern mid-latitude continents in summer" [Mitchell et al., 1990].

Figure 29: As Figure 24 but for 2 x CO2.

Soil moisture is a particularly difficult land surface characteristic to simulate for a variety of reasons including lack of observations, poor parameterizations and scale-dependent processes [Shao et al., 1994]. Predicting changes in soil moisture is even harder as the result is the difference of differences in much larger components of the surface hydrology, i.e. evaporation, precipitation and runoff plus drainage, and their potential feedbacks into regional circulation's. It is clear from the experiments described here that increasing stomatal resistance alters the surface hydro-climate and does so differentially in different regions, partly because of the relative importance of the aerodynamic and stomatal resistances and partly as the result of other factors, such as the partition of precipitation between evaporation and runoff plus drainage (e.g. Fig. 30(c)).

To date, most predictions of the impact of increasing greenhouse gases have been conducted with rather simple "bucket" land surface schemes which have shown significant decreases in soil moisture in the summertime in northern mid latitudes. It is possible that the inclusion of a more complex representation of the land surface and/or the inclusion of increased stomatal resistance may modify the conclusions drawn to date about changes in soil moisture. Thompson and Pollard (1995) compared a bucket and a more complex model (LSX) at 2xCO2 and find that although the sign of the responses in soil moisture are similar the magnitudes of the changes are much smaller with LSX (e.g. their Fig. 15). We have found that with both 2xCO2 and 2 x , BATS predicts increased soil moisture in northern mid latitudes and note that in a series of off-line simulations of a single catchment BATS partitioned total annual precipitation between runoff and evaporation in almost exactly the same proportions (Fig. 30(c)).

Figure 30: (a) Evapotranspiration t values at 95% significance levels for July differences between 2 x CO2 and 2xrs minus 1 x CO2 and 1 x rs. (b) As (a) but for total soil moisture. (c) Range of partitioning of total annual precipitation between evaporation and runoff plus drainage for a series of off-line simulations. The total evaporation comprises between about 60% and 90% of the precipitation of 865 mm, depending upon the land surface scheme employed. Interestingly, the simplest (bucket) scheme (I) and BATS (A) are almost coincident on this graph. The other schemes are intentionally not identified.

Overall, although the impact of doubling stomatal resistance can be as large as the effect of doubling atmospheric CO2 , the effects are not global (being confined to the continents and, preferentially, to forested regions). Conclusions drawn from "greenhouse" simulations which fail to take account of the likely changes in stomatal resistance or to the land surface scheme employed may give rise to results of different sign (soil moisture) and different geographical distribution (surface air temperature) than those with simple prescriptions.

IV. Conclusions

The goals of the "Regional Interactions of Climate & Ecosystems" (RICE) have been to:

- ascertain the regional effects of vegetation and soils on climates simulated by global models;

- establish the sensitivity of vegetation and ecological schemes to regional climates derived from global models; and

- facilitate the integration of new vegetation/ecological schemes into global models.

These are important and challenging goals and ones which cannot be quickly achievedin their entirety. Nonetheless, the results presented here demonstrate that two-way interactions between vegetation and soils and climate are imporant. We have found significant feedbacks and unexpected sensitivites in the parameterization schemes we have studied.

The simulation of soil moisture controls the short-term response of soils and vegitation to precipitation and evaporation demand. It also controls the very long-term capacity for a regional climate to support an ecosystem. The formulation of soil moisture in climate and ecosystem models demands a great deal more study.

Coupling of biome models to climate models begins the development of Biospheric Global Climate Models (BGCM's). The studies presented here and elsewhere show that the biome models can be tuned to produce realistic results for the present climate, but that true coupling is as difficult as that between the physical components of a GCM: atmosphere, ocean, and sea-ice.

ACKNOWLEDGEMENTS

The Regional Interactions of Climate and Ecosystems (RICE) project was made possible by generous financial support from:

Australian Research Council (ARC)

Model Evaluation Consortium for Climate Assessment (MECCA)

National Oceanographic & Atmospheric Administration (NOAA) with the University of Arizona

 

Operating funds for the GAIM office have been provided by:

US National Science Foundation (NSF)

US National Oceanographic and Atmospheric Administration (NOAA)

US Department of Energy (DOE)

US Environmental Protection Agency (EPA)

 

References:

Bazzaz, F.A., and E.D. Fajer, Plant life in a CO2 -rich world, Sci. Amer., 266(1), 68-74, 1992.

Blondin, C., Research on land surface parameterizaton schemes at ECMWF, ECMWF, UK, 1989.

Botkin, D.B., J.F. Janak, and J.R. Wallis, Some ecological consequences of a computer model of forest growth, J. Ecol., 60, 849-872, 1972.

Box, E.O., Macroclimate and Plant Forms: An Introduction to Predictive Modeling in Phytogeography, The Hague, 1981.

Ciret, C., and A. Henderson-Sellers, Sensitivity of ecosystem models to the spatial resolution of the NCAR Community Climate Model CCM2, Clim. Dyn. submitted, 1997a.

Ciret, C., and A. Henderson-Sellers, Sensitivity of global vegetation models to present-day climate simulated by global climate models, Global Biogeochemical Cycles, 11, 415-434, 1997b.

Ciret, C., J. Polcher, and X. LeRoux, An approach to simulate the phenology of savanna ecosystems in the LMD general circulation model, Global Biogeochem. Cycles, in review.

Claussen, M., Coupling global biome models with climate models, Climate Res., 4, 203-221, 1994.

Claussen, M., and M. Esch, Biomes computed from simulated climatologies, Clim. Dyn., 9, 235-243, 1994.

Cure, J.D., Carbon dioxide doubling responses: a crop survey, U. S. Dept. of Energy, Washington, D. C., 1985.

Desborough, C.E., A.J. Pitman, and P. Irannejad, Analysis of the relationship between bare soil evaporation and soil moisture simulated by 13 land surface schmes for a simple non-vegetated site, Global Planet. Change, 13, 47-56, 1996.

Dickinson, R.E., Modeling evapotranspiration for three-dimensional global climate models, Climate Processes and Climate Sensitivity, Geophysical Monograph 29, 58-72, 1984.

Dickinson, R.E., The force-restore model for surface temperature and its generalizations, J. Clim., 1, 1086-1097, 1988.

Dickinson, R.E., and A. Henderson-Sellers, Modelling tropical deforestation: a study of GCM land-surface parameterizations, Quart. J. Roy. Meteor. Soc., 114(B), 439-462, 1988.

Dickinson, R.E., A. Henderson-Sellers, and P.J. Kennedy, Biosphere-Atmosphere Transfer Scheme (BATS) Version 1e as coupled to the NCAR Community Climate Model, 80 pp., Boulder, CO, 1993.

Dickinson, R.E., A. Henderson-Sellers, P.J. Kennedy, and M.F. Wilson, Biosphere-Atmosphere Transfer Scheme (BATS) for the NCAR Community Climate Model, 69 pp., Boulder, CO, 1986.

Doyle, T.W., The role of disturbance in the gap dynamics of a montane rain forest: An application of a tropical forest succession model, in Forest Succession: Concepts and Application, (D. C. West, H. H Shugart, D. B. Botkin, eds.), Springer-Verlag, New York, 56-73, 1981.

Eamus, D., and P.G. Jarvis, The direct effects of increase in the global atmospheric CO2 concertration on natural and commercial temperate trees and forests, Adv. Ecol. Res., 19, 1-55, 1989.

Emanuel, W.R., H.H. Shugart, and M.P. Stevenson, Climatic Change and the broad-scale distribution of terrestrial ecosystem complexes, Clim. Change, 7, 29-44, 1985a.

Emanuel, W.R., H.H. Shugart, and M.P. Stevenson, Response to Rowntree, Clim. Change, 7, 457-460, 1985b.

Goutorbe, J.P., A critical assessment of the SAMER network accuracy, in Land surface Evaporation (T. Schmugge and J. Andre, eds.), Springer-Verlag, N.Y., 403-410, 1991.

Heal, O.W., J.C. Menaut, and W.L. Steffen, Towards a global terrestrial observing system(GTOS): detecting and monitoring change in terrestrial ecosystems, MAB Digest 14 and IGBP Global Change Report 26, UNESCO, Paris and IGBP, Stockholm, 1993.

Henderson-Sellers, A., Assessing the Sensitivity of a Land-Surface Scheme to Parameters Used in Tropical Deforestation Experiments, J. Roy. Meteor. Soc, 118, 1101-1116, 1992.

Henderson-Sellers, A., Continental Vegetation as a Dynamic Component of a Global Climate Model - A Preliminary Assessment, Clim. Change, 23, 337-377, 1993a.

Henderson-Sellers, A., A Factorial Assessment of the Sensitivity of the BATS Land-Surface Parameterization Scheme, J. Climate, 6, 227-247, 1993b.

Henderson-Sellers, A., Soil moisture simulation: Achievements of the RICE and PILPS intercomparison workshop and future directions, Global Planet. Change, 13, 99-116, 1996a.

Henderson-Sellers, A., Soil Moisture: A critical focus for global change studies, Global Planet. Change, 13, 3-9, 1996b.

Henderson-Sellers, A., and R.E. Dickinson, Intercomparison of land-surface parameterizations launched, EOS, 73(17), 195-196, 1992.

Henderson-Sellers, A., R.E. Dickinson, T.B. Durbidge, P.J. Kennedy, and others, Tropical Deforestation - Modeling Local-Scale to Regional-Scale Climate Change, J. Geophysis. Res., 98, 7289-7315, 1993.

Henderson-Sellers, A., W. Howe, and K. Mcguffie, The MECCA Analysis Project, Glob. and Planet. Change, 10, 3-21, 1995.

Henderson-Sellers, A., K. McGuffie, L. Zhang, and C. Gross, Sensitivity of global climate model simulations to increased stomatal resistance and CO2 increases, J. Clim., 8, 1738-1756, 1995.

Holdridge, L.R., Life Zone Ecology, 206 pp., Tropical Science Center, San Jose, Costa Rica, 1967.

Houghton, J., G. Jenkins, and J. Ephraums, Climate Change, 362pp., Cambridge University Press, Cambridge, 1990.

Houghton, J.T., b. Callander, and S.K. Varney, Climate Change 1992: The Supplementary Reprot to the IPCC Scientific Assessment, 200 pp., Cambridge University Press, 1992.

Houghton, R.A., Tropical deforestation and atmospheric carbon dioxide, Climatic Change, 19, 99-118, 1991.

Idso, S.B., and A.J. Brazel, Rising atmospheric carbon dioxide concentrations may increase streamflow, Nature, 312, 51-53, 1984.

Jarvis, P.G., The interpretation of the variations inleaf water potential and stomatal conductance found in canopies in the field, Phil. Trans. Roy. Soc. London, Ser. B 273, 593-610, 1976.

Leemans, R., Possible changes in natural vegetation patterns due to a global warming, in Der Treibhauseffet: Das Problem-Mogliche Folgen-Erforderliche Maanahmen, (Hackl, ed.), Akademie fur Umvlet and Enrgie, Laxenburg, 105-121, 1989.

LeRoux, X., Modelling the savanna radiation balance, water balance and primary production, in Lamto, a savanna ecosystem. Thirty years of ecological studies in West Africa, (J.-C. Menaut, L. Abbadie and M. Lepage, eds.), Springer Verlag , N.Y., 1997.

Leuning, R., Y.P. Wang, D. de Pury, O.T. Denmead, and others, Growth and water use of wheat under present and futre levels of CO2, J. Agr. Meteor, 48(5), 807-810, 1993.

Malanson, G.P., Comment on modeling ecological response to climatic change, Climatic Change, 23, 95-109, 1993.

Matthews, E., Global vegetation and land use: New high-resolution data bases for climate studies, Journal of Climate and Applied Meteorology, 22, 474-487, 1983.

McGuire, A.D., L.A. Joyce, D.W. Kicklighter, J.M. Esser, and others, Productivity response of climax temperate forests to elevated temperature and carbon dioxide: a North American comparison between two global models, Climatic Change, 24, 287-310, 1993.

McMurtie, R.E., and Y.P. Wang, Mathematical models of the photosynthetic response of tree stands to rising CO2 concentrations and temperatures, Plant Cell and Environment, 16, 1-13, 1993.

McMurtrie, R.E., H.N. Comins, M.U.F. Kirschbaum, and Y.P. Wang, Modifying existing forest growth models to take account of effects of elevated CO2 , Aust. J. Bot., 40, 657-677, 1992.

Mitchell, J.F.B., S. Manabe, T. Tokioka, and V. Meleshko, Equilibrium climate change, Climate Change The IPCC Scientific Assessment, 131-172 pp., Cambridge University Press, 1990.

Monserud, R.A., N.M. Tchebakova, and R. Leemans, Global vegetation change predicted by the modified Budyko model, Climatic Change, 25, 59-83, 1993.

Morison, J.I.L., and R.M. Gifford, Plant growth and water use with limited water supply in high CO2 concentrations, I. Leaf area, water use and transpiration, Aust. J. Plant Physiology, 11, 361-374, 1984.

Mahfouf, J.-F., C. Ciret, A. Ducharne, P. Irannejad, J. Noilhan, Y. Shao, P. Thronton, Y. Xue, and Z.L. Yang, Analysis of transpiration results from the RICE and PILPS Workshop, Global Planet. Change, 13, 73-88, 1996.

Noble, I.R., and R.O. Slatyer, The use of vital attributes to predict successional changes in plant communites subject to recurrent disturbances, Vegetatio, 43, 5-21, 1980.

Nobre, C., P. Sellers, and J. Shukla, Amazonian Deforestation and Regional Climate Change, J. Climate, 4, 957-988, 1991.

Ojima, D.S., K.A. Galvin, and B.L. Turner, The global impact of land-use change, BioScience, 44, 300-304, 1994.

Olson, J.S., J.A. Watts, and L.J. Allison, Carbon in live vegetation of major world ecosystems, Oak Ridge Nat'l Lab., Oak Ridge, TN, 1983.

Pacala, S.W., and G.C. Hurtt, Terrestrial vegetation and climate change: integrating models and experiments, in Biotic Interactions and Global Change, (P. M. Kareiva, J. G. Kingsolver, R. B. Huey, eds.), Sunderland, MA, ch. 5, 1993.

Parton, W.J., A. Haxeltine, P. Thornton, R. Anne, and M. Hartman, Ecosystem sensitivity to land-surface models and leaf area index, Global Planet. Change, 13, 89-98, 1996.

Pitman, A., A. Henderson-Sellers, and Z.I. Yang, Sensitivity of Regional Climates to Localized Precipitation in Global Models, Nature, 346, 734-737, 1990.

Pitman, A.J., T.B. Durbidge, A. Henderson-Sellers, and K. Mcguffie, Assessing Climate Model Sensitivity to Prescribed Deforested Landscapes, Int. J. of Clim., 13, 879-898, 1993a.

Pitman, A.J., A. Henderson-Sellers, F. Abramopoulos, R. Avissar, G. Bonan, and others, Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS). Results from off-line control simulations (phase 1A), GEWEX Report, IGPO Publication Series No. 7, 47pp, 1993b.

Policy Makers Summary, IPCC Scientific Assessment of Climate Change, WMO/UNEP, Geneva, 1990.

Pollard, D., and S.L. Thompson, The effect of doubling stomatal resistance in a global climate model, Global and Planetary Change, (in press), 1994.

Prentice, I.C., W. Cramer, S. Harrison, R. Leemans, and others, A global biome model based on plant physiology and dominance, soil properties and climate, J. Biogeogr., 19, 117-134, 1992.

Prentice, I.C., M.T. Sykes, and W. Cramer, A Simulation Model for the Transient Effects of Climate Change on Forest Landscapes, Ecological Modelling, 65, 51-70, 1993.

Riebsame, W.E., W.J. Parton, K.A. Galvin, I.C. Burke, and others, Integrated modeling of land use and cover change, BioScience, 44, 350-357, 1994.

Rosenberg, N.J., The increasing CO2 concentration on the atmosphere and its implications on agricultural productivity, Climatic Change, 3, 265-279, 1981.

Sellers, P.J., E.R. Albertazzi, J.L. Dorman, and Y.T. Hou, Modelling land surface processes at the Center for Ocean-Land-Atmoshphere interactions, COLA Univ. Maryland, UK, 1989.

Shao, Y., and A. Henderson-Sellers, Validation of soil moisture simulation in landsurface parameterisation schemes with HAPEX data, GLobal Planet. Change, 13, 11-46, 1996.

Shao, Y., R.D. Anne, A. Henderson-Sellers, P. Irannejad, and others Soil Moisture simulation. A report of the RICE and PILPS Workshop, GEWEX/GAIM Report, IGPO Publication, 130pp, 1994.

Shugart, H.H., A Theory of Forest Dynamics: The ecological Implications of Forest Succession, 278 pp., Springer-Verlag, New York, 1984.

Shugart, H.H., and D.C. West, Development of an Appalachain deciduous forest model and its application to assessment of the impact of the chestnut blight, J. Env. Manage., 5, 161-179, 1977.

Shukla, J., and Y. Mintz, Influence of land-surface evapotranpiration on the Earth's climate, Science, 215, 1498-1501, 1982.

Slingo, A., A GCM parameterization for the shortwave radiative properties of water clouds, J. Atmos. Sci., 46, 1419-1427, 1989.

Smith, T.M., R. Leemans, and H.H. Shugart, Sensitivity of terrestrial carbon storage to CO2 -induced climate change: comparison of four scenarios based on general circulation models, Climate Change, 21, 367-384, 1992.

Tegart, W.J., G.W. Sheldon, and D.C. Griffiths, Climate Change. The IPCC Impacts Assessment, Australian Gov't Printing Service, Canberra, 1990.

Tucker, C.J., H.E. Dregne, and W.W. Newcomb, Expansion and contraction of the Sahara desert from 1980 to 1990, Science, 253, 1991.

Wetzel, P.J., X. Liang, P. Irannejad, A. Boone, J. Noilhan, Y. Shao, C. Skelly, Y. Xue, and Z.L. Yang, Modeling vadose zone liquid water fluxes: Infiltration, runoff, drainage, interflow, Global Planet. Change, 13, 57-72, 1996.

Wigley, T.M.L., and P.D. Jones, Influences of precipitation changes and direct CO2 effects on streamflow, Nature, 314, 149-152, 1985.

Williamson, D., J. Kiehl, and J. Hack, Climate sensitivity of the NCAR Community Climate Model (CCM2) to horizontal resolution, Clim. Dyn., 11, 377-397, 1995.

Williamson, D.L., J.T. Kiehl, V. Ramanathan, R.E. Dickinson, and J.J. Hack, Description of NCAR Community Climate Model (CCM1), NCAR Tech. Note, 112 pp., Bouler, CO, 1987.

Williamson, G.S., and D.L. Williamson, Circulation statistics from seasonal and perpetual January and July simulations with NCAR Community Climate Model (CCM1), 112pp pp., Nat'l Center for Atmos. Res., Boulder, CO, 1987.

Wilson, M., and A. Henderson-Sellers, A global archive of land cover and soils data for use in general circulation models., J. Climate, 5, 119-143, 1985.

Wilson, M.F., A. Henderson-Sellers, R.E. Dickinson, and P.J. Kennedy, Investigation of the sensitivity of the land-surface parameterization of the NCAR Community Climate Model in regions of tundra vegetation, J. Climatol., 7, 319-343, 1987a.

Wilson, M.F., A. Henderson-Sellers, R.E. Dickinson, and P.J. Kennedy, Sensitivity of the biosphere-atmosphere transfer scheme (BATS) to the inclusion of variable soil characteristics, J. Clim. Appl. Meteor., 26, 341-362, 1987b.

Woodward, F.I., Climate and Plant Distribution, , Cambridge University Press, Cambridge, MA, 1987.

Yang, Z.-L., and R.E. Dickinson, Description of the Biosphere-Atmosphere Transfer Scheme (BATS) for the Soil Moisture Workshop and evaluation of its performance, Global Planet. Change, 13, 117-134, 1996.