SCOPE 56 - Global Change: Effects on Coniferous Forests and Grasslands

16.1 INTRODUCTION

In this more general chapter, we will present different methodologies to delimit the distributions of ecosystems under current and future climate. This is important, because large shifts in vegetation patterns can occur under a changing climate (Cramer and Leemans 1993). Current grasslands and coniferous forests can be replaced regionally by other ecosystems. If these shifts are not taken into account, impact assessment for specific ecosystems under changed climate will be misleading. The objective of this chapter is to present a global framework for impact assessment, which could define the boundary conditions of specific ecosystem assessments, such as those presented in earlier chapters.

Terrestrial ecosystems play a major role in the global C cycle. The total C content of both vegetation and soil is about three times as high as that of the atmosphere. The exchange of C between the terrestrial biosphere and the atmosphere is about 20 times larger than the anthropogenic emissions resulting from fossil fuel use. This exchange of C is influenced by a multitude of feedback processes, such as CO₂-fertilization (e.g. Bazzaz 1990), climatic change on both plant growth (e.g. Fitter and Hay 1981; Larcher 1980) and soil respiration (e.g. Parton et al. 1987) and vegetation distribution (Leemans 1992; Cramer and Leemans 1993). The C dynamics of ecosystems are mainly determined by net primary productivity (NPP) in plants, followed by C partitioning over different compartments and losses through respiration and decomposition. Every ecosystem has its characteristic C budget and dynamics, and this is often used to parameterize C cycle models (e.g. Emanuel et al. 1981; Goudriaan and Ketner 1984; Esser 1991; McGuire et al. 1992; Smith et al. 1992a; Melillo et al. 1993; Klein-Goldewijk et al. 1994).

An adequate description of vegetation and its global patterns is important for the initialization of C cycle models. The earlier C cycle models (e.g. Goudriaan and Ketner 1984) used a simple ecosystem-specific C density, which combined with ecosystem extent, allowed for a straightforward characterization of its C budget. The extents of different ecosystems types were typically taken from statistical, highly aggregated sources. Shifts in vegetation patterns, driven by climate or changing land use, were simply prescribed or simulated using transition probabilities. Recently, several modelling groups have taken a more realistic approach by using geographically explicit data bases to drive their C models (e.g. Esser 1991; McGuire et al. 1992; Klein-Goldewijk et al. 1994). Feedback processes are implemented in these models in such a way that they account for local environmental differences, such as heterogeneity in topography, climate and soils. The geographic explicit ecosystem patterns are often based on global vegetation data bases, such as Matthews (1983), Olson et al. (1985), Melillo et al. (1993) and Küchler (1949, in Espenshade and Morrison 1991) or different climate classifications (Leemans 1992; Prentice et al. 1992; Cramer and Leemans 1993). One of the currently most widely used classifications in C cycle modelling is the life zone classification by Holdridge (1967; e.g. Prentice and Fung 1990; Smith et al. 1992a; Smith and Shugart 1993).

Our approach is to review the different global vegetation, ecosystem or land cover data bases that have been used for C cycle modelling. We will distinguish between data bases derived from 'observational records' and different types of models. All data bases will be compared with each other and the major differences will be explained in terms of their underlying assumptions, limitations and origins. We will attempt to rank the global cover data bases according to their applicability to global C models. We will further review a series of applications of these data bases and models to analyse different aspects of global change issues, such as the missing C sink, feedbacks in the C cycle and vegetation response to climatic change.

16.2 REVIEW OF GLOBAL LAND COVER DESCRIPTIONS

16.2.1 Global land cover classifications and data bases

Land cover classifications are often regarded as synonymous with vegetation classifications and several institutes are devoted (e.g. The International Institute of Vegetation Mapping in Toulouse, France) to create such classifications and prepare maps. Many different aspects of vegetation have been used, either individually or in combination (Mueller-Dombois and Ellenberg 1974). The criteria used in most schemes can be characterized by:

• Structure and physiognomy. Vegetation attributes such as height, growth form, cover, deciduousness and leaf form are used as a basis of the classification. Examples of this approach are Küchler (1967), Küchler and Zonneveld (1988), and also the hierarchical UNESCO classification (1973). The life form classification by Raunkiær (1934) also fits in this category.
  
• Floristics. The taxonomic affinities of plants can also be used as the basis of a classification. The lower levels of the UNESCO (1973) classification are based on floristics. Other examples include Braun-Blanquet (1964) and Tahktajan (1973). Most of these classification systems are based on key plant taxa which occur in that vegetation type. Grabherr and Kojima (1993) recently stated that the floristic approach allows for the solid basis for a globally applicable and implementable land cover classification system. However, although ecological function is implicit in most taxonomic units, the lack of detailed information on the species of the world and on their functions, makes it impossible to apply the floristic approach directly to global land cover mapping.
  
• Bioclimatic characterization. Bioclimatic schemes are not based on actual vegetation, but on the climatic regime that prevails in that region. Many of these classification schemes are based on the observation that both vegetation physiognomy and species composition are a function of climate. The close correlation was recognized in the early nineteenth century (von Humboldt 1807) and many vegetation and climate maps were published on the basis of this relationship. Even in recently published vegetation maps (e.g. Bartholomew et al. 1988) one can tell that the major vegetation patterns are derived from climate, rather than vegetation observations. The relation, however, is only obvious for the broad scale, global vegetation patterns. At the regional and local scales, other environmental (e.g. soil and topography) and historical (e.g. disturbance and succession) factors also strongly influence the actual vegetation at any place (Leemans 1992).
  

The approach is frequently used because comprehensive data on climate are more often available than good comparable vegetation data. Examples of this approach are developed by Box (1981), Budyko (1986), Köppen (1936), Holdridge (1967), Prentice et al. (1992), Thornthwaite(1948), Walter (1985) and Whittaker (1975). The different approaches are reviewed by Tuhkanen (1980). This type of classification is most frequently used for climate change impact assessments and global change studies.

• Hybrid classification schemes. Several hybrid classifications schemes that include elements of structural, physiognomic, floristic and bioclimatic attributes have been developed. This is apparent in many of the commonly used names for vegetation categories: 'secondary tropical moist montane rain forests' or temperate short grass steppe'. Unfortunately, almost no standardization and harmonization in defining these categories is followed among different classifications schemes, so that large differences can occur between different sources (UNEP/GEMS 1993).

The most well-known hybrid classification is the UNESCO (1973) classification. The UNESCO classification was developed for the description of the potential vegetation at a climax stage. This has probably been one of the major reasons that the classification has never resulted in a global assessment of land cover. The different categories often do not refer to actual vegetation cover and this has led to major confusion. This confusion becomes very obvious in the global, highly aggregated implementation of the classification by Matthews (1983). Despite its problems, several UNESCO vegetation maps have been produced for different regions (e.g. White 1983). Another regional example of a successful hybrid classification is presented by Whitmore (1984) for tropical forests.

Many more land cover classification schemes have been developed and implemented for specific ecosystems (e.g. Whitmore 1984), large regions (e.g. Bailey 1980; White 1983; Anderson et al. 1976) or political entities (e.g. Whitmore 1984; CORINE 1989). A recent review of different approaches to vegetation classification and mapping is presented in Küchler and Zonneveld (1988). Currently there are also several global land cover data bases available that describe the current patterns. However, the origin of the data included in the compilations is not always clear. Furthermore, the data bases are often developed for a specific purpose, such as C cycling or land surface parameterizations of climate models. Here we will describe the most frequently used data bases. The legends of these data bases are given in Appendix 16.1.

16.2.1.1 The physiognomic vegetation data base ( Küchler 1949 )

This data base is derived from structural characteristics of potential vegetation (trees; shrubs; grasses; deserts) and major physiognomic features, such as deciduousness, and needle versus broad-leaved. Additional attributes used in the classification involve the percentage canopy cover (no, sparse, open and closed). A global map as presented in Espenshade and Morrison (1991) has been digitized at a resolution of 1° longitude and latitude. The data base consists of 34 classes. Although the data base gives a good representation of the potential vegetation cover, it has not (yet) been used for global change studies. This is mainly due to its coarse resolution and its incompatibilities of legends with other global tabular data sets, such as UN-ECE/FAO (1992), which are mainly based on mixed classifications. For example, the southern pine forests in Florida are not distinguished from the boreal forests. To make these necessary distinctions for linkage with other data sets, additional information on climate, soil and topography is needed.

16.2.1.2 The major ecosystems of the world (Olson et al. 1985)

This is a global data base with a resolution of 0.5° longitude and latitude, that was developed primarily to describe the C content of the major ecosystems of the world. Its documentation (Olson et al. 1985) is comprehensive with short descriptions of each class with a list of dominating species, ecosystem structure, data sources and C densities. The data base consists of 48 classes. This data base is the only land cover data base that includes explicitly natural vegetation categories, such as taiga and tropical montane rain forests, non-vegetated land categories, such as ice and stony desert, and land use categories, such as arable land, irrigated drylands and paddylands. An updated version of this data base was recently developed by Olson (included in Kineman 1992), but it lacks the detailed documentation which made the earlier version so valuable. The improvements were made only for a few regions of the world.

The C values given in this data base have frequently been used to parameterize C budget models using bioclimatic schemes to allow for shifting vegetation zones under a changed climate (e.g. Prentice and Fung 1990; Smith et al. 1992a). The IMAGE 2 model uses the data base to initialize the current land cover patterns for its dynamic simulations of changing future land use and C cycling (Alcamo et al. 1994; Leemans and van den Born 1994).

16.2.1.3 The global vegetation data base (Matthews 1983)

The global vegetation data base uses the UNESCO (1973) classification. Only the highest hierarchical levels of this classification are used here, so that the data base uses mainly the functional attributes. The data base has a relatively coarse resolution of 1° longitude and latitude and consists of 32 cover classes. The vegetation data base can be linked to a set of compatible data bases on cultivation intensity and albedo (Matthews 1983). These have been used in many different global change assessments of the NASA-GISS group (e.g. Matthews and Fung 1989; Kaufman et al. 1990; Bouwman et al. 1993). One of the problems of this data base is that it is not clear how the many different data sources (ca. 70 atlases), from which it actually was developed, are used and translated into the UNESCO system. Furthermore, the legend is a difficult to interpret combination of actual, potential and man-induced vegetation (cf. Appendix 16.1).

16.2.1.4 The global potential vegetation data base (Melillo et al. 1993)

This data base is an extension and improvement of Matthews' (1983) global vegetation data base. First, the resolution was increased to 0.5° longitude and latitude. Next, the data base was overlaid with regional implementations of the UNESCO vegetation classification, such as the vegetation map of Africa (White 1983). Ambiguous classes have been removed, which resulted in a more comprehensive data base. Despite the improvements achieved by this approach, there are large regional differences within the data base, because regional data bases that are compatible with the UNESCO classification do not exist for all parts of the world. The data base has been used to initialize an equilibrium global C cycle model (Melillo et al. 1993).

From this short review of global vegetation data bases, it becomes apparent that no satisfactory and comprehensive data base on global land cover has yet been developed. However, several international research programmes aim to produce improved data sets with modern technologies, such as remote sensing (e.g. Townshend 1992) and several of these, probably more reliable, data bases will become available to the global change research community within this decade. These approaches have already led to an improved assessment of deforestation patterns in Brazil (Skole and Tucker 1993).

16.2.2 Using environmental characteristics to predict vegetation distributions

Due to the low reliability of the current global land cover data bases, other approaches have been applied to define the global land cover patterns (including coniferous forests and grasslands). The most important one to assess changes in land use and cover has been the use of tabular statistical data to determine the extent of each vegetation category within political boundaries. Straightforward transition probability matrices were applied to determine the impact of changing patterns (e.g. Houghton et al. 1983; Goudriaan and Ketner 1984). The main limitation of this approach is that georeferenced determinants of vegetation patterns are not used to their full extent, which has resulted in large discrepancies between different analyses.

Observation-based land cover data bases suffer from scarcity of observation and from problems associated with the classifications. The first of these problems can be overcome by predicting the dominant ecosystem types from descriptors of basic physical habitat characteristics, such as climate, soils or hydrology. This, however, cannot reflect the overwhelming influence of human land use on land cover.

These habitat predictors allow for the development of land cover change scenarios driven by climate change in a geographically comprehensive way. Previously, such changes have been subscribed as one-to-one changes (e.g. Holten 1990). Many of such studies use only climatic parameters that can be computed from readily available weather station data (such as Müller 1982) and global climate data bases (such as Leemans and Cramer 1991). Some studies use simple climatic parameters directly to delimit specific vegetation types (e.g. Hulme et al. 1992) or agricultural crops (e.g. Parry 1992), but most studies use existing bioclimatic classifications or have developed more adequate, new classification schemes.

Among the first researchers who used this approach globally was Walter (1970). The patterns of monthly temperature and precipitation in his climate diagrams are used to define vegetation. These climate diagrams are popular, because they give a straightforward visual impression of seasonality and the moisture balance at a given site. They are problematic, however, because the intersections of the temperature and precipitation 'curve' are only remotely related to evapotranspiration. The beginning and end of the dry season is therefore not precisely defined. Further, it requires expert judgement to link such a climate diagram to the proper vegetation type. Therefore, this approach has not been used for global change studies.

The more frequently used approaches are comprehensive climate-vegetation classifications. The classifications are usually defined by the climatic categories only. Therefore, they can be implemented on a computer using climate data bases and geographic information systems (GIS). When the spatial pattern of current climate has been captured by the GIS, anomalies in climate change scenarios can be overlaid and new boundaries for the bioclimatic classes can be derived. This approach has mainly been used for impact studies on natural vegetation (e.g. Leemans 1992) and terrestrial C models (e.g. Prentice and Fung 1990; Smith et al. 1992a; Melillo et al. 1993).

Here we will present and discuss some of the most frequently used bioclimatic classifications (Appendix 16.2). We have implemented all these classifications on a global grid of 0.5° longitude and latitude using a global data base with climatic normals for the period 1931-60 (Leemans and Cramer 1991). The classifications are further used to analyse the impacts of a changed climate. We have used the simulated climate anomalies for an equilibrium climate for an atmosphere with doubled CO₂ atmospheric conditions. The approach taken to define a future climate is to overlay current climate with the anomalies. The precise methodology is given in Leemans (1992) and conforms to the standardized IPCC approach (Carter et al. 1992).

Figure 16.1 The life zone classification (Holdridge 1967) is determined from biotemperature and annual mean precipitation. Potential evapotranspiration is a linear function of biotemperature. The hexagons delimit the different life zones

16.2.2.1 Life zone classification ( Holdridge 1967 )

One of the most frequently used climate-vegetation classifications is the Holdridge life zone classification (Holdridge 1967, Figure 16.1). This classification is based on two climatic indices: annual precipitation and biotemperature. The latter is the average annual positive temperature. Although an axis marked PET is added, no additional information on this climatic parameter is needed, because it is simply a linear function of biotemperature, and only the balance between the two main indices determines moisture conditions. Different logarithmic combinations of the indices are used to delimit different life zones or biomes (Figure 16.1). A considerable limitation of this model is that it is based on only two annual indices. The moisture balance is therefore not properly described and seasonal aspects are lacking completely. Biomes characterized by strong seasonality, such as monsoonal forests, cannot be captured satisfactorily.

Nevertheless, the life zone classification was the first scheme to be used for the analysis of the impacts of climatic change on global vegetation patterns (Emanuel et al. 1985). Large shifts in vegetation zones can be observed when the classification is combined with different climate scenarios (Figure 16.2). The largest changes occur in high latitude areas. This is not only due to the higher temperature increase in these regions by the scenarios (for a discussion see Mitchell et al. 1990), but also to the specific sensitivities of the life zone classification.

Figure 16.2 Shifts in life zones under several doubled CO₂-derived climates as determined by the life zone classification (adapted from Leemans 1992). The left part of the histogram presents a decrease in extent, while the right part presents an increase, both in respect to the current extent

16.2.2.2 Global climate classification (Köppen 1936)

Köppen (1936) tried to capture the annual cycles of temperature and precipitation in a climate classification. He designed his classification so that the categories approximately resemble global vegetation patterns. It is the first empirical and objective climate classification that is established with a limited set of climatic parameters, defining five major climate categories (Table 16.1; Strahler and Strahler 1987). The strength of the Köppen (1936) climate classification is that it includes both the different latitudinal zones (based on extreme temperatures) and seasonality in both precipitation and temperature. It therefore should theoretically perform better than the life zone classification by Holdridge (1967). However, the extent of arid climates is largely underestimated due to an inadequate description of the hydrological cycle by only comparing extreme temperatures with precipitations. The classification system neglects evapotranspiration as an important limiting factor for plant growth (Thornthwaite 1943). Its advantage, however, is that it does not use land cover terms to label climatic zones, but a hierarchy of symbols (Table 16.1). The possibility of misinterpreting the nature of these climatic zones by less experienced investigators is therefore reduced.

The Köppen (1936) climate classification has been used by the US Forest Service to delineate ecosystem regions (Bailey 1983, 1989). Bailey (1983) defines ecoregions as large ecosystems of regional extent that contain a number of smaller ecosystems. They define major geographical zones that represent associations of similarly functioning vegetation or potential land covers. Bailey's (1983) purpose was to develop a land cover classification that divided the landscape into variously sized ecosystem units that have significance both for resource development or environmental conservation. The major problem with such an ecoregion approach is that only climatological parameters are emphasized and that the resulting cover classification will strongly focus on potential cover class and neglect human use. A slightly modified Köppen (1936) climate classification (Trewartha 1968) that addressed some of the shortcomings of the aridity definitions has been used by Guetter and Kutzbach (1990) to analyse the impacts of changing climate on land cover patterns during the last 18000 years. They clearly illustrate the large changes in land cover, especially in mid and high latitudes, that have occurred since the last glaciation. Figure 16.3 illustrated the potential shifts of the climate classes under several doubled CO₂ climates.

16.2.2.3 Biogeographical zones (Budyko 1986)

The biogeographical zones (Budyko 1986) are based on an ordination of moisture- and temperature-related indices. As such it has strong similarities to the classifications of Thornthwaite (1948) and Whittaker (1975). However, the latter two are based on climatic indices that are less functional in their definition, and therefore less likely to reflect the major correlations between climate and vegetation. Budyko (1986) uses an approach that delimits vegetation classes through the computed energy (or radiation) and moisture balance (Figure 16.4). The moisture balance is characterized by a dryness index, which is based on an elaborate evapotranspiration scheme, that accounts for latitude, humidity and energy provided by the local radiation balance. If climate data are available, this approach is more reliable than the empirical evapotranspiration indicators developed by Holdridge (1959) or Thornthwaite and Mather (1957). It clearly separates tundra, forest, steppes, semi-deserts and deserts of the main zones. Within each zone there are large differences in the energy balance for the forested zone, but these become much smaller with increasing aridity (Figure 16.4).

The Budyko scheme (1986) has frequently been used by investigators from the former Soviet Union in their global change impacts studies (e.g. Izrael et al. 1990) and has recently been coupled to climate change scenarios derived from climate models, by Tchebakova et al. (1993a, b; Figure 16.5). They distinguished 16 land cover classes world-wide and illustrated the potential shifts of vegetation patterns under a changed climate. Although the results compare well with other studies and the performance of the ordination is relatively good (see below), the approach still has some major disadvantages for global change studies. First, the radiation balance is used to compute some vegetation characteristics, such as albedo. Such characteristics should not be used for vegetation prediction, because of the circularity in the approach. Second, seasonality is not adequately covered, so that no linkages can be made to the more physiognomic and structural vegetation types like those of Küchler (1949). Third, the boundaries are not very clearly defined. The table containing limits for the different zones (Budyko 1986: p. 94) has to be modified subjectively to be implemented unambiguously on global climate data bases. Finally, the classification depends on a series of not readily available climatic and physical parameters (such as humidity and surface albedo). The lower quality of data bases with these parameters limits the suitability of the biogeographical zones for global change studies.

Figure 16.3 The ordination for biogeographical zones as defined by the energy balance (R, W m^-2) and the relative dryness index (RjLp, L = the latent heat of evaporation, p = annual precipitation) and the delineations of the major biogeographical zones (Budyko 1986)

16.2.2.4 The BIOME plant functional type model

A different approach to model global vegetation patterns has recently been developed by Prentice et al. (1992) with the BIOME model. Their aim was to develop a conceptually simple model which primarily defined the climatic limits of the most important plant types, rather than biomes. Hence, the model captures aspects of ecological function in limiting the distribution of plant types. The bioclimatic limits are defined such that they can be based on physiological processes and their physical limits. The approach was based on the model of plant-forms (Box 1981). The BIOME approach also has similarities with the rule-based model for the North American vegetation types (Neilson et al. 1992; Neilson 1993).

Figure 16.4 Shifts in biogeographic zones (Budyko 1986) under several doubled CO₂-derived climates. The left part of the histogram presents a decrease in extent, while the right part presents an increase, both in respect to the current extent.

Plant functional types (PFTs) are defined for distinct temperature zones (tropical, temperate and boreal or warm, cool and cold). These zones coincide with the major latitudinal zones, but the labels for these zones are just for convenience and do not imply a rigid boundary. Secondly, for each zone the major physiognomic adaptations relating to limiting climatic factors such as evergreen vs deciduous, broad-leaved vs needle-leaved, and woody vs herbaceous are described. The PFTs show similarities to Küchler's (1949) physiognomic classification. Although the list (cf. Table 16.2) represents an oversimplification of the variety of plant types that exist, it can capture the major features of major biomes and their transient zones rather well.

The BIOME model relates the distributions of the PFTs to climate indices such as growing degree days, mean temperature of the coldest and warmest month, and the a-moisture index (Figure 16.6). The temperature-based indices defined the cold tolerances, chilling and heat requirements for each PFT. The a-moisture index is defined as the ratio between actual and potential evaporation and determined from a plain bucket-type soil water balance model. This model computes actual evapotranspiration by accounting for precipitation, potential evapotranspiration and a soil-specific water supply to plants. It explicitly includes soil characteristics and is able to carry forward moisture into dry seasons (Prentice et al. 1992). Prentice et al. (1993) give a complete description of the algorithm. Only a small number of the critical values for the climatic variables are given (based on ecophysiological principles. Table 16.2). The criterion for use of certain limits was that it proved to be necessary to match the vegetation patterns given by the major ecosystems of the world data base (Olson et al. 1985). For example, tropical evergreen trees are assumed to tolerate no frost and to have a high moisture requirement. Based on a world-wide regression of annual minimum temperature against mean coldest-month temperatures, 'no frost' implies a coldest month temperature of >15.5ºC. Based on map comparisons, a 'high moisture requirement' means an a-coefficient of at least 0.8. Limits for all other plant types are determined in a similar way. Finally, a 'dominance hierarchy' in which PFTs dominate over others (e.g. trees dominate over grasses) was defined. This hierarchy was strictly applied so that only PFTs from the highest level present were retained.

Figure 16.5 Shifts in climatic zones under several doubled CO₂-derived climates as determined by the modified global climate classification (Köppen 1936; Trewartha 1968). The left part of the histogram presents a decrease in extent, while the right part presents an increase, both in respect to the current extent

Figure 16.6 Structural diagram of the BIOME model (Prentice et al. 1992)

At each location, an array of PFTs is determined to occur. Unique combinations of PFTs define implicitly different biomes, which therefore emerge from the analysis, rather than being defined a priori. The model should be quite capable of producing novel combinations under a changed climate. The model generated 17 different combinations for current climate (Table 16.3). The BIOME model has been used to assess C dynamics in past, current and future climates (Leemans 1992; Prentice et al. 1994). Because it is driven only by climatic and soil characteristics, the BIOME model cannot simulate any effect of changing land use. In an attempt to assess changes in global C storage caused by changes in available land for agriculture, Cramer and Solomon (1993) overlaid the biome vegetation patterns with a category 'climatologically suitable for agricultural land', which matched fairly well with global maps of non-irrigated crops. This approach is further elaborated upon by linking BIOME with a potential agricultural model (Leemans and Solomon 1993). Both are now an essential part of the terrestrial environment system of the IMAGE 2 model (Alcamo et al. 1994; Leemans and van den Born 1994), which determines global vegetation response to changing land use, atmospheric conditions and climate.

Table 16.2 The parameters for each plant type of the BIOME model: 1. Growing degree days, base 0°C; 2. Growing degree days, base 5 °C; 3. Mean temperature of the coldest month; 4. Mean temperature of the warmest month; 5. µ moisture index. The last column (6) gives the dominance hierarchy for each plant functional type

16.2.3 Constraints of the different climate-vegetation classifications

The presentation of the different climate classifications and their implementations gives an overview of the progress that has been made during the last decade. The models have achieved significant improvement concerning the mechanisms of the physical relationships between the atmosphere and biosphere. The more recent models (e.g. Box 1981; Woodward 1987; Prentice et al.1992) are all derived from physiological considerations, rather than correlations. This is a very important development for global change studies, because the more mechanistic models are likely to be more robust under changed climatic conditions.

Table 16.3 Combinations of plant functional types generated by the BIOME model. The area (1000 km²) is the global extent of each combination

The major problem for climate change impact assessments based on the first three classifications (life zones, global climate and the biogeographical zones) is their high sensitivity to small changes along the boundaries. These boundaries are very sharp on maps with implementations of these classifications (see e.g. Cramer and Leemans 1993), and small changes could lead to large shifts in the global patterns. In reality, the boundaries are often not so clear, and transient zones or ecotones are abundant across many landscapes. Ecotones are probably also more resilient against climatic change, because they include elements of several zones, which allows for a larger adaptive capability.

The most important improvement of the different approaches is the use of plant functional types. This approach allows for a more realistic response of vegetation to a changing environment. Palaeoecological studies clearly demonstrate that plants react to climate change as individual taxa. Biomes have formed, dissolved and re-formed throughout the Quaternary period (Huntley and Webb 1988). The life zone, global climate and the biogeographical zones classifications are therefore less suitable, because their basic unit is biomes, not a taxon. These models could lead to an inadequate description of vegetation patterns under climatic change, because novel, no-analogue vegetation types per definition cannot occur within these models. The more suitable models, however, are still strongly limited in the dynamic responses of vegetation, taking into account processes like migration, competition and succession.

16.2.4 Comparison of the different data sets and climate classifications

Global land cover databases and bioclimatic classifications or models have different shortcomings. Nevertheless, a pairwise comparison between data bases from either of these groups may lead to insights into the limitations of either the data base or the model. If so, we can start to derive some general principles which should be included in a model used for global change studies. The comparison presented here is based on the results of all the different data bases and global classifications, implemented into a GIS (Leemans 1992), which is linked to an array of different spatial statistical techniques to compare, overlay and plot different data sets.

The creation of comparable sets from all nine data bases (Table 16.4) for such comparison was not a trivial task. First, we defined a common grid for the comparison of all nine available data sets. We defined this grid as those cells that in all data sets were designated as land cells. We did not consider the large, ice covered land masses, Antarctica and Greenland. The coarser data sets (physiognomic vegetation classification (Küchler, 1949) and world vegetation data base (Matthews 1983)) were overlaid onto the finer common grid of 0.5° longitude and latitude. Incomparable cells, such as those with a specific land use (especially in the ecosystems of the world data base (Olson et al. 1985), the category 'cultivation' in the global vegetation data base (Matthews 1983) was empty), had further to be removed from all data sets (cf. Table 16.4). This common grid is used for further analysis and consists of 44 335 cells and includes a surface of 99 239 km² (75% of the total terrestrial ecosystems).

Secondly, all data sets had to be aggregated into a comparable and compatible legend by reclassifying and aggregating the original data bases. Legends appear to be compatible using similar labels, but these could mean structurally very different categories. A further difficulty was related to the distinction between actual, potential, or human-induced vegetation in some data bases. This was difficult because the necessary documentation on definitions and sources of the data bases was not always available. We constructed a target classification with only 18 different classes (Table 16.4), which incorporated the major biomes that are needed to obtain an adequate resolution for different C cycle models (e.g. Melillo et al. 1993; Leemans and van den Born 1994). The original classes are sometimes split using secondary information on climate, topography or location. If large discrepancies occurred due to peculiar emerging patterns, the whole process was repeated. All rules for the final aggregation are listed in Table 16.4.

Table 16.4 The aggregation for the different land cover classifications (the numbers listed correspond with the different legend items - see Appendices 16.1 and 16.2)

The resulting compatible data sets were finally analysed by comparing the extent of different classes separately as well as by analysing the differences in the spatial patterns using the Kappa statistic (Cohen 1960; Monserud and Leemans 1992). This statistic compares cell-to-cell agreement for each category and for the data set as a whole. The Kappa statistic ranges from -1 (total disagreement) to 1 (total agreement) and is very suitable to rank similarities and differences between complex spatial patterns. Monserud and Leemans (1992) suggested that values <0.4 are to be considered poor, 0.4-0.55 fair, 0.55-0.7 good, 0.7-0.85 very good and 0.85-1.0 excellent. A limitation of the original Kappa statistic is that it treats each cell independently and the resolution of the data bases is not suitable to define all scales of interactions. Some patterns emerge on larger scales and there the original Kappa statistic was also applied by Prentice et al. (1992) to blocks of aggregated cells. This generalized Kappa index displays an overall fit over larger blocks of cells, taking into account the dependency of neighbouring cells.

The Kappa statistic (Table 16.5) illustrates only a poor to fair agreement between all different maps. This is also indicated by the low correspondence between all extents. The location of the biomes that all data sets agree upon are those with a large extent, such as hot desert, boreal and tropical rain forests. Ecotones and the smaller biomes are characterized by large differences among the different data bases. The biogeographical zones (Budyko 1986) do not translate into the whole range of categories. The mixed categories (Table 16.4) are lacking and their cells are merged with the often broader neighbouring categories. This could lead to an unrealistically high agreement between the major biomes and the overall Kappa statistic and extent figures for the biogeographical zones.

It is clear that the different observational land cover databases are heterogeneous. The Kappa statistic indicated only poor to fair agreement between them (Table 16.5). Up to 60% of all cells may be classified differently. This difference is most visible in the comparison between the global vegetation data base (Matthews 1983) and the major ecosystems of the world data base (Olson et al. 1985). The other data bases display a somewhat better agreement among themselves. The discrepancy between the different data sets can perhaps be explained by the large differences in the actual classifications, but we have tried to be as careful as possible in creating compatible data sets. Our conclusion from this comparison is that we actually have only a very approximate comprehension of actual global land cover. This conclusion is further strengthened by the fact that even in the statistical, tabular data bases based on country censuses, such as FAO agricultural and forestry yearbooks, large differences occur between the data.

There is a clear pattern for the comparison between the different models based on the climate-vegetation classifications and global land cover data bases (Table 16.5). The BIOME model generally provides the more adequate simulation of global vegetation patterns, closely followed by Budyko's biogeographical zones. The global climate classifications display an intermediate fit. The frequently used life zone classification has the poorest fit. Both the Kappa statistic and the extent figures present these trends. The land cover categories that are simulated most accurately by all models are hot desert and tundra, closely followed by tropical rainforest and taiga. All other categories are simulated with a much reduced accuracy.

We have drawn the following conclusions for this emerging pattern of data base comparisons. The life zone classification (Holdridge 1967) is based only on annual mean climatic parameters that do not capture the major driving forces of climate upon vegetation patterns. Major physiognomic patterns are adaptations to specific seasonality in temperature and precipitation regimes. The global climate classifications (Köppen 1936; Trewartha 1968) explicitly include seasonality and therefore give a somewhat better performance. The modified Köppen classification (Trewartha 1968) shows a closer match with most observed distributions of biomes because of the close correlation with the boundaries between arid and moist zones. The BIOME model (Prentice et al. 1992) and the biogeographical zones (Budyko 1986) both include a more realistic parameterization for the moisture balance. The biogeographical zone model is only based on potential evapotranspiration, while BIOME incorporates a more elaborate moisture availability scheme, including soil characteristics. However, both models do not strongly consider seasonal aspects of moisture availability and can be improved in this respect. For example, inclusion of seasonality by using the characteristics of a dry or growing period (cf. Leemans and Solomon 1993) could already enhance performance.

16.3 APPLICATIONS OF GLOBAL CLIMATE-VEGETATION MODELS

16.3.1 Predicting future biome redistribution caused by climate change 

The earliest approaches that determined the impacts of climate on ecosystems used climate-vegetation classifications. Emanuel et al. (1985) used the Holdridge life zone classification to determine the extent of forests and grasslands under different climatic conditions. Their analysis clearly showed that with climatic warming the broad-scale vegetation patterns could shift considerably polewards. Besides these (largely) latitudinal shifts, they predicted that some regions might shift from forested to grassland ecosystems. On a global scale, this scenario seems unreasonable today, because changes in precipitation were assumed to be zero. Consequently, warming could lead to a global decrease in moisture, and the global water balance would not be stable.

Table 16.5 Kappa statistic and extent in common (10³ km²; upper half of the matrix) between the different global land cover maps. Wetlands and agricultural lands are excluded from the analysis. The total extent of land cover used for this assessment was 99239 x 10³ km²

Since the time of this far-reaching study, many other impacts studies using climate-vegetation classifications have been conducted. A thorough review is given by Cramer and Leemans (1993), who reimplemented both the life zone classifications and the first computerized plant functional type model (Box 1981 ). The sensitivities of these models were different under the climate change scenarios using anomalies for both temperature and precipitation. The life zone classification showed large shifts in vegetation zones for most high latitude regions, while in Box's model, tropical forests also showed large decreases in extent. This latter decline may be an artefact of the strong temperature sensitivity in Box's model. In his setting of the climatical limits, tropical trees could not survive at temperatures over 30 °C.

The potential shifts in life zones were used to assess the impact on large nature reserves (Leemans and Halpin 1992). Depending on the actual scenario, between 30 and 60% of all reserves could be severely affected with strong negative consequences for biodiversity.

We have repeated the climate-change impacts analysis for the different climate-vegetation classifications (except for the original global climate classification) with several climate scenarios based on both temperature and precipitation anomalies produced by four general circulation models for the Atmosphere (GCMs): the Oregon State University model (OSU, Schlesinger and Zhao 1989); the Goddard Institute for Space Studies model (GISS, Hansen et al. 1988); the Princeton General Fluid Dynamics Laboratory model (GFDL, Manabe and Wetherald 1987) and the United Kingdom Meteorological Office model (UKMO) of Mitchell (1983). The scenarios are listed in order of their global annual mean temperature increase under doubled CO₂ atmospheric conditions. All four scenarios differ in the details of the geographic distribution of climate changes they predict. In general all models predict a temperature increase, which is larger in the winter season and high latitude regions. The simulated precipitation change is geographically more complex, but all models, except OSU, show a significant increase of precipitation (e.g. Mitchell et al. 1990). The methodology for defining a future climate scenario is described by Leemans (1992).

Figure 16.2 illustrates the potential shifts in vegetation zones using the life zone classification under a changed climate. Even for the most modest scenario (OSU), few areas remain unaffected. Similar changes can be observed for the other climate-vegetation models (Figures 16.3, 16.5 and 16.7). From these figures it becomes clear that all climate-vegetation classifications show similar responses.

Figure 16.7 Shifts in biomes (Prentice et al. 1992) under several doubled CO₂-derived climates. The left part of the histogram presents a decrease in extent, while the right part presents an increase, both in respect to the current extent

Most shifts occur in high latitude regions. The extent of tundra is largely reduced, while the southern part of the tundra is replaced by boreal forests, the southern part of the boreal forest by temperate forests in more maritime climates and by steppe in the more continental climate. The changes in the tropics are less easy to generalize. Deserts and tropical rain forests remain relatively constant, while the vegetation types in more seasonal climates change considerably. Here the largest differences occur between different GCM scenarios.

Looking at these different impact assessments, it is striking how similar the changes unfold for the different GCM scenarios and climate-vegetation models, especially if one accounts for the uncertainties and differences in all models and their simulations. The four GCM scenarios considered here actually span the whole range of 1.5-5.0ºC temperature increase given by IPCC (Houghton et al. 1990). The reason for the convergence in these impact assessments can partly be explained by the methodology of creating a scenario. We have used a baseline climatology with monthly values (Leemans and Cramer 1991) and overlaid them with the appropriate climatic anomalies. The underlying baseline pattern is always part of the scenario, and pronounced climatic gradients will remain, especially with a coarser GCM resolution (3-7°) than the baseline (here 0.5°). This method could therefore lead to a greater similarity between scenarios than when the doubled CO₂ GCM simulations are compared together. In interpreting the results, one has to be aware of these limitations, but there are currently no better approaches which are widely accepted (Carter et al. 1992).

However, we believe that the climatic anomalies of the different GCMs are much more similar, as experienced by vegetation. The temperature signal between the different GCMs varies considerably, but mostly in the temperature ranges that do not affect plants very much. This is the main explanation for the relatively similar patterns of change in bioclimatic patterns of change in bioclimatic boundaries between different GCM scenarios. For example, the largest differences in the four scenarios occur in January temperatures at high latitudes with increases ranging from 5 °C (OSU) to 25 °C (GFDL). The large temperature increase significantly influences the mean annual temperature increase. Plants are not strongly affected when temperatures remain well below freezing, as is the case for all scenarios.

There are still considerable limitations to such a vegetation-climate model approach. First, it is implicitly assumed that vegetation is in equilibrium with climate. However, the response, adjustment and redistribution of plants require time. Simulations with vegetation-climate models are probably valid for longer time spans, such as millennia (Prentice et al. 1991), but they are surely not valid in the decade to century time span for which the climatic change due to the enhanced greenhouse effect is foreseen. Several dynamic models (e.g. Prentice et al. 1993; Smith et al. 1992b) have been used to determine the time-dependent vegetation response simulating species-specific life-histories and succession processes. From these analyses, it becomes obvious that time lags up to several centuries could easily occur.

Such impact studies further assume that species can reach suitable localities instantaneously. Using palaeoecological data, Davis (1981) has clearly demonstrated that this is not true and that maximum historical migration speeds for trees range from 10 to 100 km per century. This slow rate could be further reduced in the current fragmented landscape, but others have argued that humans also could facilitate higher migration rates. However, with a relatively fast rate of climatic change, conditions could be suitable for any species during the seedling or sapling stage, but not any more for more mature stages. This is especially true for long-lived species, such as trees, and would make an adequate adaptation in sectors such as forestry, extremely difficult.

Secondly, the parameters used in most climate-vegetation models are not the ones that define the actual distribution of vegetation. (We have already discussed that species and not biomes are the actual unit of change.) Together with climate and soil, vegetation history strongly determines the actually occurring vegetation in a given locality and these vegetation patterns should be used for defining a more realistic transient response, not the potential patterns. Determining the response of actual vegetation could be achieved by linking 'plant functional type'-schemes with more dynamic models. The principles of such an approach were recently presented by Smith and Shugart (1993), but the results of their approach were less reliable because they still used the life zone classification.

16.3.2 Determining the global C budget

One of the most important applications of global vegetation or land cover models and data bases is the determination of the role of the terrestrial biosphere in the global C cycle. A straightforward method is to use climate data to derive the C contents of each vegetation type. The first model that used this approach was the Miami model (Lieth 1975). This model is based on a regression between climate indices and net primary productivity (NPP) and is still used in some of the current C cycle models (Esser, 1991). This regression method was further improved by Seino and Uchhijima (1992) who used the indices developed for the biogeographical zones (Budyko 1986) to define NPP and produce global productivity maps. This model, however, has not been used in global change studies.

Prentice and Fung (1990) used the life zone classification (Holdridge 1967) to define the potential C content of the terrestrial biosphere. This is done by: (1) assigning a C density to each life zone; (2) computing the extents of each life zone and (3) multiplying (l) and (2). They did this analysis for the last glacial maximum (LGM: 18 000 BP), current climate and doubled CO₂ climate conditions. The distribution of life zones for the different climate was determined (cf. Figure 16.2) and the specific C content computed. Their conclusion was that the terrestrial biosphere had been acting as a C sink since the LGM and would continue to do so under climatic change. Smith et al. (1992a) have redone this analysis using the life zone classification at its full resolution and added soil C densities using the values given by Post et al. (1982). Aboveground C densities were assigned using the values given by Olson et al. (1985). They did the analyses for several GCM scenarios and concluded that the sink strength of the terrestrial biosphere was not as large as initially predicted by Prentice and Fung (1990). Similar approaches using different models were taken by Adams et al. (1990) and Prentice et al. (1994; Figure 16.7).

These approaches are all based on the equilibrium between the terrestrial biosphere, climate and the other components of the C cycle. Such an equilibrium may exist for longer time scales, but is not valid for scales ranging from decades to centuries (Prentice et al. 1991). The short-term transient dynamics could give a completely different picture of C storage in the terrestrial biosphere. Several analyses (Neilson 1993; Smith and Shugart 1993) have shown that through an increased forest dieback there could be a C pulse to the atmosphere from the biosphere under climatic warming, before the distribution of vegetation belts are readjusted and biospheric C uptake increases again. Such transient dynamics might have large implications for mitigation strategies that aim to reduce CO₂ in the atmosphere. The response of the biosphere would not likely be linear and smooth. Many surprises could occur and our understanding of most processes and interactions is rather incomplete. This is clearly illustrated by the recent sudden decline in atmospheric CO₂ build-up (Sarmiento 1993). No comprehensive explanations have been proposed yet.

Another problem with the global C budget is the inability to balance the global C cycle and most analyses cannot account for a 'missing sink' of about 2 Pg C per year. Siegentaler and Sarmiento (1993) concluded in a recent review that the uptake of CO₂ by the oceans is relatively well understood and that the excess C should be taken up by the terrestrial biosphere. Kauppi et al. (1992) reports that boreal and temperate forests are currently sinks of C. This sink function is further elaborated on by Dai and Fung (1993) who assume a large sink in these regions through slightly increasing temperatures. Houghton (1993) challenges this conclusion, stating that these large increases must have been observed in annual growth increments.

Part of this important issue of balancing the global C budget and reducing the missing sink, can well be explained by the inadequacies of the global land cover data sets and models. The current state-of-the-art global C budget models (e.g. Emanuel et al. 1984; Goudriaan and Ketner 1984; Esser 1991; Melillo et al. 1993) all use different land cover data bases to initialize and parameterize the different ecosystems' productivities and environmental feedbacks. The differences between these data sets are too large to permit a good assessment of the global C cycle. The models therefore can give insights into the processes and their interactions and be very effective in defining trends in the C dynamics under a changing climate, increasing atmospheric CO₂ concentrations, and SOM. But these models and especially their underlying data bases do not allow us to determine the actual size of sources and sinks in the terrestrial biosphere. Discussions on the actual location and characteristic of the missing sink will continue for a while.

16.4 CONCLUSIONS AND SUMMARY

Recently, several global models that simulate one or more aspects of the terrestrial biosphere have been developed (Melillo et al. 1993; Prentice et al. 1992; Alcamo et al. 1994). The analysis presented above stresses the importance of the development of a high-quality global land cover data base, which can be used to initialize, calibrate and/or validate these global models. No reliable and generally accepted data base exists today. This has led to a multitude of approaches, often leading to largely conflicting opinions on the importance of ecological processes.

The current global carbon cycle models (e.g. Melillo et al. 1993; Kindermann et al. 1993) are not yet able to simulate shifts in vegetation zones caused by climatic change and many other vegetation related feedbacks. To include such important vegetation responses, global climate-vegetation classifications should be applied. Of the currently available models, BIOME is the most appropriate candidate, because it includes the most important aspects of seasonality in temperature patterns, a quasi-realistic moisture balance and its vegetation responses are based on a series of independent plant functional types. However, BIOME still lacks an appropriate seasonality in its definition of the moisture regime. Advancements in modelling global vegetation patterns should start with improving the simple moisture balance model and include a better array of soil characteristics, especially now that an improved version of the FAO world soil resources has been released (Anonymous 1993).

Several global assessments on the impacts of climatic change on terrestrial ecosystems have been accomplished (e.g. Houghton et al. 1990, 1992; Izrael et al. 1990,1992). These studies have been limited by the sparsely available data with a comprehensive global cover. Most analyses have therefore been limited to regional and/or sectorial studies and little integration has been achieved. Many global change studies have therefore been very anecdotal. Only recently has a more global ecological theory begun to emerge (e.g. Solomon and Shugart 1993; Kareiva et al. 1993; Ehleringer and Field 1993). Currently, a series of integrated models are being developed that aim to assess the most important aspects of global change and will be used to address important policy issues (e.g. Alcamo et al. 1994). These models can now be developed because of the improved understanding of global biogeochemical cycles, ecological processes and interactions between climate, soil and land cover. It is probably this kind of model that will most clearly indicate the weaknesses of current data bases and will benefit most from future improvements.

16.5 ACKNOWLEDGEMENTS

We would like to thank I. Colin Prentice and Martin T. Sykes for providing the data of Figure 16.7. Critical readings of earlier versions of this draft by Joe Alcamo and Kees Klein-Goldewijk are appreciated. The research was funded by the Dutch National Programme on 'Global Air Pollution and Climate Change' under contracts NOP 852067 (MAP 481510 and 482507) and the Dutch Ministry of Housing, Physical Planning and the Environment under contract MAP 481507. The study further contributes to core research of the Global Change and Terrestrial Ecosystems project of IGBP.

16.6 REFERENCES

Adams, J. M., Faure, H., Faure-Denard, L., McClade, J. M. and Woodward, F. I. (1990) Increases in terrestrial carbon storage from the Last Glacial Maximum to the present. Nature 348, 711-714.

Alcamo, J., Kreileman, G. J. J., Krol, M. and Zuidema, G. (1994) Modeling the global society-biosphere-climate system, Part 1: Model description and testing. Water Air Soil Pollut. 76, 1-35.

Anderson, J. R., Hardy, E. E.,Roach,J. T. and Witmer, R. E. (1976) A Land Use and Land Cover Classification System for Use with Remote Sensor Data. Professional Paper 964. US Geological Survey, Washington DC.

Anonymous (1993) World Soil Resources. An Explanatory Note on the FAO World Soil Resources Map at 1 :25000000 scale. World Soil Resources Report 66 rev. 1. Food and Agriculture Organization of the United Nations, Rome.

Bailey, R. G. (1980) Description of the Ecoregions of the United States. Miscellaneous Publication No.1391. US Department of Agriculture, Ogden, Utah.

Bailey, R. G. (1983) Delineation of ecosystem regions. Environ. Manage. 7, 365-373. 

Bailey, R. G. (1989) Explanatory supplement to ecoregions maps of the continents. Environ. Conserv. 16, 307-309.

Bartholomew, J. C., Christie, J. H., Ewington, A., Geelan, P. J. M., Lewisobe, H. A. G., Middleton, P. and Winkleman, H. (Eds) (1988) The Times' Atlas of the World. Times Books Limited, London.

Bazzaz, F. A. (1990) The response of natural ecosystems to the rising global CO₂ levels. Ann. Rev. Ecol. Syst. 21, 167-196.

Bouwman, A. F., Fung, I., Matthews, E. and John, J. (1993) Global analysis of the potential for N₂O production in natural soils. Glob. Biogeochem. Cyc. 7, 557-597.

Box, E. O. ( 1981) Macroclimate and Plant Forms: An Introduction to Predictive Modeling in Phytogeography. Dr W. Junk Publishers, The Hague.

Braun-Blanquet, J. (1964) Pflanzensoziologie, Grundzüge der Vegetationskunde. Springer- Verlag, Vienna.

Budyko, M. I. (1986) The Evolution of the Biosphere. D. Reidel Publishing Company, Dordrecht.

Carter, T. R., Parry, M. L., Nishioka, S. and Harasawa, H. (1992) Preliminary Guidelines for Assessing Impacts of Climate Change. IPCC Working group 2 report. Environmental Change Unit, Oxford and Center for Global Environmental Research, Tsukuba.

Cohen, J. (1960) A coefficient of agreement for nominal scales. Educat. Psychol. Measurements 20, 37-46.

CORINE (1989) CORINE database manual, version 2.1. Commission of the European Communities, Brussels.

Cramer, W. and Leemans, R. (1993) Assessing impacts of climate change on vegetation using climate classification systems. In: Solomon, A. M. and Shugart, H. H. (Eds) Vegetation Dynamics Modelling and Global Change, pp. 190-217. Chapman & Hall, New York.

Cramer, W. and Solomon, A. M. (1993) Climatic classification and future global redistribution of agricultural land. Clim. Res. 3, 97-110.

Dai, A. G. and Fung, I. Y. (1993) Can climate variability contribute to the 'Missing' CO₂ sink? Glob. Biogeochem. Cyc. 7, 599-609.

Davis, M. B. (1981) Quaternary history and the stability of forest communities. In: West, D. C., Shugart, H. H. and Botkin, D. B. (Eds) Forest Succession: Concepts and Application, pp 132-154. Springer-Verlag, New York.

Ehleringer, J. R. and Field, C. B. (Eds) (1993) Scaling Physiological Processes: Leaf to Globe. Academic Press, San Diego.

Emanuel, W. R., Killough, G. E. G. and Olson, J. S. (1981) Modelling the circulation of carbon in the world's terrestrial ecosystems. In: Bolin, B. (Ed.) Carbon Cycle Modelling, pp. 335-353. Wiley & Sons, New York.

Emanuel, W. R., Killough, G. E. G., Post, W. M. and Shugart, H. H. (1984) Modeling terrestrial ecosystems in the global carbon cycle with shifts in carbon storage capacity by land-use change. Ecology 65, 970-983.

Emanuel, W. R., Shugart, H. H. and Stevenson, M. P. (1985) Climatic change and the broad-scale distribution of terrestrial ecosystems complexes. Clim. Change 7, 29-43.

Espenshade, E. B. Jr. and Morrison, J. L. (Eds)(1991) Goode's World Atlas. Rand McNally & Company, Chicago.

Esser, G. (1991) Osnabrück biosphere model: structure, construction, results. In: Esser, G. and Overdieck, D. (Eds) Modern Ecology, Basic and Applied Aspects, pp. 679-709, Elsevier, Amsterdam.

Fitter, A. H. and Hay, R. K. M. (1981) Environmental Physiology of Plants. Academic Press, London.

Goudriaan, J. and Ketner, P. (1984) A simulation study for the global carbon cycle, including man's impact on the biosphere. Clim. Change 6, 167-192.

Grabherr, G. and Kojima, S. (1993) Vegetation diversity and classification systems. In: Solomon, A. M. and Shugart, H. H. (Eds) Vegetation Dynamics and Global Change, pp. 218-232. Chapman & Hall, New York.

Guetter, P. J. and Kutzbach, J. E. (1990) A modified Köppen classification applied to model simulations of glacial and interglacial climates. Clim. Change 16, 193-215.

Hansen, J., Fung, I., Lacis, A., Rind, D., Lebedeff, S., Ruedy, R., Russell, G. and Stone, P. (1988) Global climate changes as forecast by Goddard Institute for Space Studies three-dimensional model. J. Geophys. Res. 93, 9341-9364.

Holdridge, L. R. (1959) Simple method for determining potential evapotranspiration from temperature data. Science 130, 572.

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

Holten, J. I. (1990) Biologiske og Økologiske konsekvenser av klimaforandringer i Norge. NINA Utreding 11. Norsk Institutt for Naturforskning, Trondheim.

Houghton, J.T., Callander, B. A. and Varney, S. K. (Eds) (1992) Climate Change 1992. The Supplementary Report to the IPCC Scientific Assessment. Cambridge University Press, Cambridge.

Houghton, J.T., Jenkins, G. J. and Ephraums, J. J. (Eds)(1990) Climate Change: The IPCC Scientific Assessment. Cambridge University Press, Cambridge.

Houghton, R. A. (1993) Is carbon accumulating in the northern temperate zone? Glob. Biogeochem. Cyc. 7, 611-617.

Houghton, R. A., Hobbie, I. E., Melillo, J. M., Moore, B. III, Peterson, B. J., Shaver, G. R. and Woodwell, G. M. (1983) Changes in the carbon content of terrestrial biota and soils between 1860 and 1980: a net release of CO₂ to the atmosphere. Ecol. Monogr. 53, 235-262.

Hulme, M., Wigley, T.,Jiang, T., Zhao, Z.-C., Wang, F., Ding, Y., Leemans, R. and Markham, A. (1992) Climate Change due to the Greenhouse Effect and its Implications for China. WWF, Gland, Switzerland.

Huntley, B. and Webb, T. III (Eds) (1988) Vegetation History. Kluwer Academic Publishers, Dordrecht.
Izrael, Y. A., Canziani,O., Hashimoto,M., Odingo,O. S. and Tegart,W. J. M.(Eds)(1992) Climate Change 1992: The Supplementary Report to the IPCC Impact Assessment. WMO and UNEP, Geneva.

Izrael, Y. A., Hashimoto, M. and Tegart, W. J. M. (Eds)(1990) Climate Change: The IPCC Impact Assessment. Australian Government Publishing Service, Canberra.

Kareiva, P. M., Kingsolver, J. G. and Huey, R. B. (Eds) (1993) Biotic Interactions and Global Change. Sinauer Associates Inc., Sunderland, Mass.

Kaufman, Y. J., Tucker, C. J. and Fung, I. (1990) Remote sensing of biomass burning in the tropics. J. Geophys. Res. 95, 9927-9939.

Kauppi, P., Mieliküinen, K. and Kuusela, K. (1992) Biomass and carbon budget of European forests, 1971 to 1990. Science 256, 70-74.

Kindermann, J., Lüdeke, M. K. B., Badeck, F. W., Otto, R. D., Klaudius, A., Hager, C., Wurth, G., Lang, T., Donges, S., Habermehl, S. and Kohlmaier, G. H. (1993) Structure of a global and seasonal carbon exchange model for the terrestrial biosphere- The Frankfurt Biosphere Model (FBM). Water Air Soil Pollut. 70, 675-684.

Kineman, J. J. (1992) Global Ecosystems database Version 1.0 (on CDROM) User's guide. Key to Geophysical Records Documentation No.26. USDOC/NOAA National Oceanic and Atmospheric Administration, Boulder, Color.

Klein-Goldewijk, K., van Minnen, J. G., Kreileman, G. J. J., Vloedbeld, M. and Leemans, R. (1994) Simulating the carbon flux between the terrestrial environment and the atmosphere. Water Air Soil Pollut. 76, 199-230.

Köppen, W.(1936)Das geographische System der Klimate. In: Köppen, W. and Geiger, R. (Eds) Handbuch der Klimatologie, pp. 1-46. Gebrüder Borntraeger, Berlin.

Küchler, A. W. (1949) A physiognomic classification of vegetation. Ann. Ass. Amer. Geog. 39, 201-210.

Küchler, A. W. (1967) Vegetation Mapping. The Ronald Press Company, New York. Küchler, A. W. and Zonneveld, I. S. (Eds) (1988) Vegetation Mapping. Kluwer Academic Publishers, Dordrecht.

Larcher, W. (1980) Physiological Plant Ecology. Springer-Verlag, Berlin.

Leemans, R. (1992) Modelling ecological and agricultural impacts of global change on a global scale. J. Sci. 1nd. Res. 51, 709-724.

Leemans, R. and Cramer, W. (1991) The IIASA Databasefor Mean Monthly Values of Temperature, Precipitation and Cloudiness on a Global Terrestrial Grid. Research Report RR-91-18. International Institute of Applied Systems Analyses, Laxenburg, Austria.

Leemans, R. and Halpin, P. (1992) Global change and biodiversity. In: Groombridge, B. (Ed.) Biodiversity 1992: Status of the Earth's Living Resources, pp. 254-255. Chapman & Hall, London.

Leemans, R. and Solomon, A. M. (1993) The potential response and redistribution of crops under a doubled CO₂ climate. Clim. Res. 3, 79-96.

Leemans, R. and van den Born, G. J. (1994) Determining the potential global distribution of natural vegetation, crops and agricultural productivity. Water Air Soil Pollut. 76, 133-161.

Lieth, H. (1975) Modelling the primary production of the world. In: Lieth, H. and Whittaker, R. H. (Eds) Primary Productivity of the Biosphere, pp. 203-215. Springer- Verlag, Berlin.

McGuire, A. D., Melillo, J. M., Joyce, L. A., Kicklighter, D. W., Grace, A.L., Moore, B.III and Vrsmarty, C. J. (1992) Interactions between carbon and nitrogen dynamics in estimating net primary productivity in North America. Glob. Biogeochem. Cyc. 6, 101-124.

Manabe, S. and Wetherald, R. T. (1987) Large-scale changes in soil wetness induced by an increase in carbon dioxide. J. Atmos. Sci. 44, 1211-1235.

Matthews, E. (1983) Global vegetation and land use: new high-resolution data bases for climate studies. J. Clim. Appl. Meteorol. 22, 474-487.

Matthews, E. and Fung, I. (1989) Methane emission from natural wetlands: global distribution, area, and environmental characteristics of sources. Glob. Biogeochem. Cyc. 1, 61-68.

Melillo, J. M., McGuire, A. D., Kicklighter, D. W., Moore, B.III, Vörösmarty, C. J. and Schloss, A. L. (1993) Global climate change and terrestrial net primary production. Nature 363, 234-240.

Mitchell, J.F.B. (1983) The seasonal response of a general circulation model to changes in CO₂ and sea temperatures. Quart. J. R. Meteorol. Soc. 109, 113-152.

Mitchell, J.F.B., Manabe, S., Meleshko, V. and Tokioka, T. (1990) Equilibrium climate change-and its implications for the future. In: Houghton, J. T., Jenkins, G. J. and Ephraums, J. J. (Eds) Climate Change: The IPCC Scientific Assessment, pp. 131-172. Cambridge University Press, Cambridge.

Monserud, R. A. and Leemans, R. (1992) The comparison of global vegetation maps. Ecol. Model. 62, 275-293.

Mueller-Dombois, D. and Ellenberg, H. (1974) Aims and Methods of Vegetation Ecology. John Wiley & Sons, Chichester.

Mller, M. J. (1982) Selected Climatic Data for a Global Set of Standard Stations for Vegetation Science. Dr W. Junk Publishers, The Hague.

Neilson, R. P. (1993) Vegetation redistribution-a possible biosphere source of CO₂ during climatic change. Water Air Soil Pollut. 70, 659-673.

Neilson, R. P., King, G. A. and Koerper, G. (1992) Toward a rule-based biome model. Landscape Ecol. 7, 27-43.

Olson, J., Watts, J. A. and Allison, L. J. (1985) Major World Ecosystem Complexes Ranked by Carbon in Live Vegetation: A Database. Report NDP-017. Oak Ridge National Laboratory, Oak Ridge, Tenn.
Parry, M. (1992) The potential effect of climate changes on agriculture and land use. Adv. Ecol. Res. 22, 63-92.

Parton, W. J., Schimel, D. S., Cole, C. V. and Ojima, D. S. (1987) Analysis of factors controlling soil organic matter levels in Great Plains grasslands. Soil Sci. Soc. Am. J. 51, 1173-1179.

Post, W. M., Emanuel, W.R., Zinke, P. J. and Stangenberger, A. G. (1982) Soil carbon pools and world life zones. Nature 298,156-159.

Prentice, I. C., Bartlein, P. J. and Webb, T. III(1991) Vegetation and climate change in eastern North America since the last glacial maximum. Ecology 72, 2038-2056.

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

Prentice, I. C. and Fung, I. Y. ( 1990) The sensitivity of terrestrial carbon storage to climate change. Nature 346, 48-51.

Prentice, I. C., Sykes, M. T. and Cramer, W. (1993) A simulation model for the transient effects of climate change on forest landscapes. Ecol. Model. 65, 51-70.

Prentice, I. C., Sykes, M. T., Lautenschlager, M., Harrison, S. P., Denissenko, O. and Bartlein, P. J. (1994) Modelling global vegetation patterns and terrestrial carbon storage at the last glacial maximum. Biogeogr. Let. 3, 67-76.

Raunkiaer, C. (1934) Life Forms of Plants and Terrestrial Plant Geography. Clarendon Press, Oxford.

Sarmiento, J. L. (1993) Carbon cycle-atmospheric CO₂ stalled. Nature 365, 697-698. 

Shlesinger, M. E. and Zhao, Z.-C. (1989) Seasonal climatic changes induced by doubled CO₂ as simulated by the OSU atmospheric GCM/mixed-layer ocean model. J. Clim. 2, 459-495.

Seino, H. and Uchhijima, Z. (1992) Global distribution of net primary productivity of terrestrial vegetation. J. Agr. Met. 48, 39-48.

Siegentaler, U. and Sarmiento, J. L. (1993) Atmospheric carbon dioxide and the ocean. Nature 365, 119-125.

Skole, D. and Tucker, C. (1993) Tropical deforestation and habitat fragmentation in the Amazon: Satellite data from 1978 to 1988. Science 260, 1905-1910.

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

Smith, T. M. and Shugart, H. H. (1993) The transient response of terrestrial carbon storage to a perturbed climate. Nature 361, 523-526.

Smith, T. M., Shugart, H. H., Bonan, G. B. and Smith, J. B. (1992b) Modelling the potential response of vegetation to global change. Adv. Ecol. Res. 22, 93-116.

Solomon, A. M. and Shugart, H. H. (Eds) (1993) Vegetation Dynamics and Global Change. Chapman & Hall, New York.

Strahler, A. N. and Strahler, A. H. (1987) Modern Physical Geography. J. Wiley & Sons, New York.

Tahktajan, A. A. (1973) Origin and Distribution of the Angiosperms. Academic Press, New York.

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

Tchebakova, N. M., Monserud, R. A., Leemans, R. and Golovanov, S. (1993b) A global vegetation model based on the climatological approach of Budyko. J. Biogeogr. 20, 219-244.

Thornthwaite, C. W. (1943) Problems in the classification of climate. Geogr. Rev. 33, 233-255.

Thornthwaite, C. W. ( 1948) An approach toward a rational classification of climate. Geogr . Rev. 38, 55-94.

Thornthwaite, C. W. and Mather, J. R. (1957) Instructions and tables for computing potential evapotranspiration and the water balance. Publ. Climatol. 10, 185-310.

Townshend, J. R. G. (1992) Improved Global Data for Land Application: A Proposal for a New High Resolution Data Set. IGBP-Report No.20. International Geosphere- Biosphere Programme, Stockholm.

Trewartha, G. T. (1968) Köppen's classification of climates. In: Trewartha, G. T. (Ed.) An Introduction to Climate, pp. 395-399. McGraw-Hill, New York.

Tuhkanen, S. (1980) Climatic parameters and indices in plant geography. Acta Phytogeogr. Suec. 67,1-110.

UN-ECE/FAO(1992) The Forest Resources of the Temperate Zones: 1990 Forest Resource Assessment. UN-ECE/F AO, Rome.

UNEP/GEMS (1993) Vegetation Classification. Report of the UNEP-HEM/ WCMC/GCTE preparatory meeting, Charlottesville, Virginia, USA. GEMS Report Series No.19. United Nations Environment Programme, Nairobi.

UNESCO (1973) International Classification and Mapping of Vegetation. Report United Nations Educational, Scientific and Cultural Organization, Paris.

Von Humboldt, F. H. A. (1807) Ideen zu einer Geographie der Pflanzen neben einem Naturgemälde der Tropenländer. Tübingen.

Walter, H. (1970) Vegetationszonen und Klima. Eugen Ulmer, Stuttgart.

Walter, H. (1985) Vegetation of the Earth and Ecological Systems of the Geo-Biosphere. Springer- Verlag, Berlin.

White, F. (1983) UNESCO Vegetation Map of Africa. UNESCO, Paris.

Whitmore, T. C. (1984) A vegetation map of Malaysia at the scale 1:5 million. J. Biogeogr. 11, 461-471.

Whittaker, R. H. (1975) Communities and Ecosystems. Macmillan, New York. 

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

16.7 APPENDICES

Appendix 16.1 The legend of the different land cover data bases. The numbers in the left column correspond with the numbers in Table 16.4

Appendix 16.2 The legend of the different bioclimatic classifications. The numbers in the left column correspond with the numbers in Table 16.4