We used a regionally modified global vegetation model (BIOME4-Tibet) to simulate biome distribution on the Tibetan Plateau under current climate conditions derived from regional meteorological observations. The bioclimatic limits (mean temperatures of the coldest and warmest months, minimum temperature, growing degree-days on 5 °C and 0 °C bases) for some key alpine plant functional types (temperate deciduous and conifer trees, boreal deciduous and conifer trees, desert woody plants, tundra shrubs, cold herbaceous plants, and lichens/forbs) were redefined based on regional vegetation-climate relationships. Modern vegetation maps confirmed that the BIOME4-Tibet model does a better job of simulating biome patterns on the plateau (gridcell agreement 52%) than the original BIOME4 model (35%). This improved model enhanced our ability to simulate temperate conifer forest, cool conifer and mixed forest, evergreen taiga, temperate xerophytic shrubland, temperate grassland and desert, and steppe and shrub tundra biomes, but made a negligible or reduced difference to the prediction of temperate deciduous forest, warm-temperate mixed forest, and three tundra biomes (erect dwarf-shrub tundra, prostrate dwarf-shrub tundra, and cushion forb, lichen, and moss tundra). Future modification of the vegetation model, by increasing the number of shrub and herb plant functional types, re-parameterization of more precise bioclimatic constraints, and improved representation of soil, permafrost, and snow processes, will be needed to better characterize the distribution of alpine vegetation on the Tibetan Plateau.
Introduction
The Tibetan Plateau, with an average elevation of ca. 4000 m and often described as a third pole, is a unique region for climate and vegetation studies. High altitude and seasonal monsoonal circulation systems control the regional climate and land surface properties on the Tibetan Plateau (Chang, 1983; Lau and Li, 1984). The Neogene-Quaternary uplift of the plateau has significantly affected the vegetation distribution patterns in East Asia (Chang, 1983; Chen et al., 2005). In turn, the land surface, particularly the vegetation, affects regional atmospheric circulation and monsoonal systems (Liu et al., 2003; Yasunari, 2006). The species composition and diversity of Tibetan vegetation is highly sensitive to warming and grazing (Klein et al., 2004). Large-scale vegetation distribution is also vulnerable to climatic warming and enhanced atmospheric CO2 concentrations (Ni, 2000, 2011). Despite its potential significance for environmental studies, the impacts of climate change on Tibetan vegetation still are not sufficiently explored.
In order to quantify the impacts of global change on terrestrial ecosystems, a modeling approach is necessary. This is because ecosystem models, both statistically and mechanically, simplify yet synthesize the ecological processes, patterns, and structures of the biosphere at the appropriate temporal and spatial scales for simulating interactions with atmosphere and anthropogenic disturbances. Dynamic and equilibrium global vegetation modeling leads to a better understanding of vegetation distribution and feedbacks under changed land cover, climate, and atmospheric compositions in the past and future. It provides a useful tool for predicting vegetation distribution and carbon and water processes at regional to global scales and for various time periods in response to large-scale environmental changes (Peng, 2000; Prentice and Raynaud, 2001; Prentice et al., 2007).
The modeling of global vegetation has been conducted from simulations by equilibrium biogeographical (e.g. BIOME1: Prentice et al., 1992) or biogeochemical models (e.g. TEM: Melillo et al., 1993; CENTURY: Parton et al., 1993), coupled biogeography and biogeochemistry models (e.g. MAPSS: Neilson and Marks, 1994; BIOME3: Haxeltine and Prentice, 1996; BIOME4: Kaplan et al., 2003), and dynamic models of ecosystem processes (e.g. IBIS: Foley et al., 1996; LPJ: Sitch et al., 2003; five dynamic global vegetation models: Sitch et al., 2008; LPJ-Why: Wania et al., 2009a, 2009b). With an aim of working towards Earth System modeling, dynamic global vegetation models have been coupled with atmosphere-ocean dynamic models (Prentice and Raynaud, 2001; Prentice et al., 2007; Smith et al., 2011). However, equilibrium models are still useful for simulating past vegetation change and future vegetation development under particular simulated climate conditions (Prentice and Raynaud, 2001).
Focusing on Tibetan biome simulation, Ni (2000) improved the biome classification of the BIOME3 equilibrium global vegetation model (Haxeltine and Prentice, 1996) in terms of alpine vegetation, and updated climatic parameters, yielding a good agreement between modern and predicted vegetation on the Tibetan Plateau. However, the modified model only used very simple climate factors such as growing degree days on the 0 °C base and annual mean precipitation (Ni, 2000). The successor to BIOME3, the BIOME4 model (Kaplan et al., 2003) has been modified to simulate both modern and mid-Holocene Tibetan vegetation (Song et al., 2005). Biomes and their plant functional types (PFTs) have been re-classified (Song et al., 2005) according to an old Chinese vegetation classification at 1∶4 million scale (Hou et al., 1982), but regional biome classification and bioclimatic parameters are not comparable with global biome and bioclimate systems.
In this paper, we modify the BIOME4 model by redefining the bioclimatic parameters of key PFTs on the Tibetan Plateau. The modified BIOME4-Tibet model is applied to simulate modern biome distribution on the Tibetan Plateau driven by local climatic observations. The aims of this study are (1) to test the performance of the original global vegetation model in simulating high-altitude biome distribution and to improve the global vegetation model to better predict regional biomes, particularly tundra; (2) to conduct a detailed assignment of regional vegetation types to biomes; and (3) to compare potential and simulated biomes. Reasons for incorrect simulations and potential improvements that could be made to the model will be further discussed.
Study Area
The Tibetan Plateau defined here (27°–40°N, 73°–105°E) includes the Karakorum and Kunlun Mountains in the southern Xinjiang Autonomous Region, the Qaidam Basin and surrounding mountains in Qinghai Province, the whole Xizang Autonomous Region, and the western mountainous areas of Yunnan and Sichuan Provinces. The high elevation and associated upper atmospheric circulation has resulted in the development of a series of so-called ‘high-cold’ vegetation types on the Tibetan Plateau (Chang, 1983). The western part of the plateau receives very little precipitation (ca. 50 mm y−1) and is therefore dominated by semi-shrubby desert and steppe desert vegetation, especially Ceratoides latens. In the northwest, the climate is very dry and cold due to the high altitude and higher northern latitude. A sparse high-cold desert of low semi-shrubby and cushion-like shrub (Ceratoides compacta) occurs there. Annual precipitation increases to ca. 200 mm in the center of the plateau where high-cold steppe vegetation (Stipa purpurea) occupies large, flat areas. In the east, where cold and wet climates dominate (annual precipitation reaches 600 mm), there is a high-cold meadow, dominated by several Kobresia species mixed in with low scrubs comprised of Potentilla and Caragana (Chang, 1983; ECVC, 1980). Temperate desert and steppe vegetation are also distributed on the northern margin of the plateau and around the Qaidam Basin in the northeast where elevations are lower and the climate is relatively dry. Subalpine coniferous (Abies and Picea) and mixed forests as well as Rhododendron shrublands are found on the southeastern and eastern margins. Therefore, the vegetation of the plateau follows a transitional gradient from the southeast to the northwest, graduating from subalpine forests, high-cold meadow and scrub, through high-cold steppe and desert to high-cold desert (ECVC, 1980).
Methods
BIOME4 AND THE REGIONAL MODIFICATION
The BIOME4 model (Kaplan, 2001; Kaplan et al., 2003) is a global vegetation model with coupled biogeographical and biogeochemical processes to predict steady-state vegetation distribution, structure, and carbon and water fluxes, and interactions among these processes. The primary simulated unit is the plant functional type (PFT), which represents broad, physiognomically and physiologically distinct plant groups, ranging from cushion forb to tropical rainforest tree. The computational core of BIOME4 is a coupled carbon and water flux scheme (Haxeltine and Prentice, 1996), which determines the seasonal maximum leaf area index (LAI) that maximizes net primary productivity (NPP) for any given PFT, based on a daily time step simulation of soil water balance and monthly process-based calculations of canopy conductance, photosynthesis, respiration, and phenology. The model is sensitive to changes in atmospheric CO2 concentration because of the responses of NPP and stomatal conductance to CO2 and the differential effects of CO2 on the NPP of C3 and C4 plants (Kaplan, 2001).
Thirteen PFTs were used to assign 28 biomes throughout the world (Kaplan, 2001). Three PFTs (cold shrub, cold graminoid/forb, and cushion forb) were used to distinguish among five tundra biomes. Six non-tundra PFTs (cold deciduous tree, cold evergreen tree, temperate broadleaf deciduous tree, temperate needleleaf evergreen tree, xerophytic shrub, and temperate grass) were used to simulate high-latitude vegetation types including four boreal forests, three temperate forests, temperate grassland, and xerophytic shrubland. Other PFTs including tropical/warm-temperate grass and desert woody C3 and C4 plant types were also used (Kaplan, 2001).
The BIOME4 model uses widely accepted bioclimatic and ecophysiological parameters to define PFTs and for the assignment of biomes (Woodward, 1987; Prentice et al., 1992; Kaplan, 2001; Kaplan et al., 2003). However, some of these parameters are not suitable for PFTs on the Tibetan Plateau. In this study we did not modify the rules for assigning Tibetan biomes, but instead re-calibrated some of the parameters for key PFTs. The phenological type of C3/C4 temperate grass has been defined as ‘raingreen’ in the global vegetation classification (Kaplan, 2001), but is ‘summergreen’ in the temperate regions of China (ECVC, 1980; EBVAC, 2001). The phenology of tundra shrub and lichen/forb on the plateau is ‘summergreen’ in most cases rather than ‘evergreen.’ On the basis of Chinese vegetation, the expected leaf longevity was modified from 8 to 6 months in the case of cold herbaceous plants, from 12 to 10 months for C4 tropical grass and from 8 to 7 months in the case of C3/C4 temperate grass plants (ECVC, 1980). The limits of some climatic indices, such as the mean coldest month temperature and the growing-degree days on 5 °C base, were redefined (Table 1) based on established Tibetan vegetation-climate relationships and published literature (ECVC, 1980; Ni, 2000; EBVAC, 2001; Song et al., 2005). The basic structure and subroutines remained the same (Kaplan, 2001; Kaplan et al., 2003).
TABLE 1
Bioclimatic parameters in BIOME4-Tibet. Tcm and Twm are the mean temperatures of the coldest month and the warmest month, respectively. Tmin is the minimum temperature, and GDD5 and GDD0 are the growing degree days on the basis of 5 °C and 0 °C, respectively. * indicates that the original parameters have been changed. Numbers in parentheses are the original parameters from Kaplan (2001).
INPUT DATA OF BIOME4-TIBET
The regional climate data set consisted of four variables (monthly temperature, precipitation and cloudiness, and minimum temperature) that were averaged from long-term records at 1814 meteorological stations across China (China Meteorological Administration, unpublished data in 2003). Climate data obtained from 703 standard stations were averaged between 1971 and 2000, and from 1111 local stations were averaged between 1981 and 1990. These data were interpolated into 10 km × 10 km gridcells using a thin plate smoothing spline surface-fitting technique (Hutchinson, 2006; ANUSPLIN version 4.36) that takes the impact of elevation into account on the basis of the STRM digital elevation model (Farr et al., 2007). An atmospheric CO2 concentration of 360 ppmv was used for the modern baseline simulation. In addition, BIOME4 utilizes soil properties of the water-holding capacity and percolation rate of two layers (top: 0–30 cm; bottom: >30 cm). Data regarding these two soil properties are not available in regional resource, although there is a soil database of China at 1∶1 million scale (Shi et al., 2004). The soil data used in this paper was read from a global digital soil map at a 0.5° resolution (FAO, 1995; Kaplan, 2001).
VEGETATION AND BIOMES
The Tibetan vegetation map, derived from the Vegetation Atlas of China at a scale of 1∶1 million, was originally comprised of 269 vegetation formations and sub-formations which belong to 37 vegetation types and 10 non-vegetation categories (EBVAC, 2001). The most common Tibetan vegetation are alpine steppe (grass and Carex high-cold steppe), alpine meadow (Kobresia and forb high-cold meadow), and alpine sparse vegetation. They are distributed on flat areas and account for 60% of the total area of the whole plateau. Montane conifer forest, subalpine evergreen broadleaved shrubland, and subalpine deciduous broadleaved shrubland occupy ca. 14% of the total area and are distributed mainly in southeastern and eastern Tibet. Subalpine dwarf-shrub desert, alpine cushion dwarf-shrub desert, and areas with no vegetation account for an additional 14% of the plateau and are distributed in the north and northwest, and in the Qaidam Basin. Other vegetation types account for the remaining 12% of the plateau.
Observed vegetation patterns are needed to evaluate the model simulation. Vegetation classification based on observations must be translated into the biomes used in the vegetation model. This is always a complicated task because of the difficulty in matching observed vegetation with modeled biomes and where some vegetation definitions and biome classifications are ambiguous. Previous assignments of vegetation to biomes in China (e.g. Ni et al., 2000) and on the Tibetan Plateau (e.g. Ni, 2000; Song et al., 2005) were based on bioclimatic features and spatial distributions of observed vegetation from published maps. A particular vegetation type (a named category of plant community or vegetation in vegetation science, including one to several vegetation formations or sub-formations) was assigned to one biome, although some vegetation types span very broad horizontal and/or vertical ranges. However the practice of assigning a vegetation type to only one biome on the Tibetan Plateau is questionable because vegetation distribution is largely controlled by the altitudinal gradient. The Tibetan vegetation types often span a large elevational range where bioclimates are different and various biomes are likely to occur. Therefore in this study we attempted to assign vegetation types which span large elevation ranges to different biomes. Based on information of vegetation regionalization (division) of the Tibetan Plateau (ECVC, 1980; CITTP, 1988), for example, Abies forests (A. faxoniana, A. squamata, A. georgei, A. georgei var. smithii, A. delavayi, A. delavayi var. motuoensis, and A. densa) and Picea forests (P. likiangensis var. balfouriana, P. likiangensis var. linzhiensis, and P. spinulosa) were assigned to evergreen taiga if elevation was >3000–4000 m (varied depending on latitude), and to cool conifer forest if elevation was ≤3000–4000 m. Pinus forests (P. densata and P. yunnanensis) were assigned to cool mixed forest or cool conifer forest or cold mixed forest if elevation was >2000–3000 m, and to temperate conifer forest or warm mixed forest if elevation was ≤2000–3000 m. Rhododendron scrubs (R. fastigiatum, R. heiliolepis, R. adenogynum, R. racemosum, R. nivale, and R. thymifolium) and other alpine and subalpine evergreen or deciduous scrubs were assigned to shrub tundra if elevation was >3500–4000 m, and to cool/cold mixed forests and shrubland if elevation was ≤3500–4000 m. Kobresia meadows (K. pygmaea, K. humilis, and K. capillifolia) and Stipa steppes (S. purpurea) were assigned to shrub tundra if they are in mosaics distributed with shrubs and otherwise to graminoid and forb tundra (ECVC, 1980; CITTP, 1988).
Biomes simulated by the vegetation model represent potential natural vegetation, so cultivated vegetation types must be assigned to a potential biome in the same bioclimatic zone. Additionally some vegetation types, particularly azonal (local habitat-controlled) shrublands and coniferous forests, need to be assigned to zonal (broad bioclimate-controlled) biomes (Table 2).
TABLE 2
Assignment of vegetation types (EBVAC, 2001) to biomes on the Tibetan Plateau.
We used the original biome classification of the BIOME4 model for simulation and mapping rather than combined vegetation and mega-vegetation types as per Song et al. (2005), meaning that the biomes are comparable with the global biome scheme. All of the vegetation formations and sub-formations were assigned to 18 biomes (Table 2; Fig. 1, a) used in the BIOME4 simulation. The most widely distributed biomes are steppe tundra and shrub tundra (Fig. 1, a) which occupy ca. 73% of the total area of the Tibetan Plateau. Steppe tundra (38%) is distributed across the whole plateau but mainly in central, northwestern, and northern Tibet. Shrub tundra (35%) is concentrated in southern, southeastern, and eastern Tibet (Fig. 1, a). Temperate desert (9.5%, in the northern margin and the Qaidam Basin), evergreen taiga/montane forest (4.5%, in southeastern Tibet) and prostrate shrub tundra (4%, in northwestern Tibet) biomes account for ca. 18% of the entire plateau. The erect dwarf-shrub tundra and cushion forb, lichen, and moss tundra only occupy a small area of the plateau (Fig. 1, a).
Results
OVERALL BIOME SIMULATIONS
The original BIOME4 simulation produced 18 biomes (Fig. 1, b). The model failed to simulate tropical evergreen forest and land ice. Instead it predicted cool mixed forest and barren areas, which do not exist in the observational data (Fig. 1, a). The model over-predicted three biomes: evergreen taiga/montane forest in large areas of northeastern, eastern, central western, and northern Tibet; erect dwarf-shrub tundra in the central plateau; and deciduous taiga/montane forest in northeastern and southern plateau. Temperate desert, steppe tundra, and shrub tundra biomes were under-estimated in ca. 40–50% (based on the number of gridcells) of instances (Fig. 2, a and b; Table 3). Agreement between observed and simulated biomes was only 35.1% (Fig. 1, a and b), with particularly large disagreement in the central, southern, and eastern areas of the plateau (Fig. 2, a and b).
FIGURE 2
Differences between (a) observed and (b) original BIOME4 simulated biomes, and between (c) observed and (d) modified BIOME4 simulated biomes. Correct matches are shown in gray.
TABLE 3
Comparison between observed and original BIOME4 simulated biomes (gridcell number) on the Tibetan Plateau.
The improved model with re-parameterized PFTs predicted a better biome distribution on the Tibetan Plateau (Fig. 1, c) with an overall match rate to observational data (Fig. 1, a) of 52.4%. The extent of evergreen taiga/montane forest (Fig. 1, c) was significantly reduced compared to the original simulation and there was no simulation of deciduous taiga/montane forest (Fig. 1, b), although the amount of evergreen taiga vegetation was still over-predicted (Fig. 2, c and d). The accuracy of correct predictions for desert, shrub tundra, and steppe tundra increased (Fig. 1, c) but the prediction of erect dwarf-shrub tundra was still poor, and cold mixed forest, temperate xerophytic shrubland, and cushion forb, lichen, and moss tundra were all over-predicted (Fig. 2, c and d; Table 4).
TABLE 4
Comparison between observed and regional-improved BIOME4 simulated biomes (gridcell number) on the Tibetan Plateau.
DETAILED COMPARISON OF BIOME PREDICTIONS
Although an agreement of 52.4% between observed and modified model simulated biomes is still not high, the difference of 17% in the number of accurate predictions compared to the original model is a substantial improvement. Under closer examination (Tables 3 and 4), biomes were often mismatched to others with similar bioclimatic characteristics (Fig. 2).
Tundra is the most extensive biome on the Tibetan Plateau. Low- and high-shrub tundra (hereinafter shrub tundra) and graminoid and forb tundra (steppe tundra) were both correctly simulated in 58% of instances (Fig. 1, a and c; Table 4), which is better than the original model simulation of 43.7% and 30.4%, respectively (Fig. 1, a and b; Table 3). However, they were also wrongly simulated as each other in many cases, and also as three other tundra biomes (erect dwarf-shrub tundra, prostrate dwarf-shrub tundra, and cushion forb, lichen, and moss tundra—hereinafter cushion forb tundra), barren areas, temperate xerophytic shrubland, temperate grassland, temperate desert and cool- and cold-temperate, and boreal forests (Fig. 2, c and d; Table 4). Other tundra biomes, especially erect dwarf-shrub tundra, prostrate dwarf-shrub tundra and steppe tundra, and temperate desert and land ice were incorrectly predicted to be shrub tundra. The shrub tundra, temperate grassland and desert, and two forest biomes (evergreen taiga/montane forest and cold-temperate mixed forest) were also incorrectly predicted to be steppe tundra in many sites (Fig. 2, c and d; Table 4). Both the original and modified models could not accurately predict erect and prostrate dwarf-shrub tundra, and only predicted cushion forb tundra correctly in a few cases. Barren areas and land ice were also not simulated well (Tables 3 and 4).
Temperate shrubland was correctly predicted in 56% of cases and was wrongly simulated as temperate desert and cold-temperate mixed forest. Temperate grassland was wrongly simulated as temperate shrubland, steppe tundra, evergreen taiga/montane forest, and cold-temperate mixed forest, and was only correctly predicted in 16% of instances. Temperate desert was largely correctly simulated (65%) and the wrong predictions came mainly from shrub tundra and steppe tundra (Table 4). These temperate shrubland, grassland, and desert biomes were all better predicted than the original model (Tables 3 and 4).
Cool-temperate conifer forest was correctly simulated by the modified model in 56% of cases and was mainly wrongly assigned to cold-temperate mixed forest. Cold-temperate mixed forest was correctly predicted in only 33% of cases and was largely incorrectly predicted to be evergreen taiga/montane forest (39%). It was also wrongly assigned to steppe tundra, temperate grassland, and shrubland (in total ca. 19%). In addition to this, many other forest sites (73%) of evergreen taiga/montane forest and temperate conifer forest, grassland, desert, and tundra (steppe and shrub tundra) were incorrectly predicted to be cold-temperate mixed forest (Table 4). All these cold- and cool-temperate forests were better simulated than the original model (Table 3). Evergreen taiga/montane forest was correctly simulated by the modified model in 46% of cases (Table 4) which is slightly worse than the original model prediction (Table 3), but the modified simulation reduced the number of sites from steppe tundra, shrub tundra, temperate grassland, and cold-temperate mixed forest which were wrongly predicted as evergreen taiga/montane forest (Tables 3 and 4). Non-deciduous taiga/montane forest was largely correctly simulated by both models. Temperate deciduous forest and temperate conifer forest were slightly better predicted, but warm-temperate mixed forest was more poorly simulated by the modified model (Tables 3 and 4).
Discussion and Conclusions
REASONS FOR INCORRECT SIMULATIONS
The modified BIOME4-Tibet yielded a 17% improvement in the correct simulation of Tibetan biomes compared to the original BIOME4 model. The simulation of two tundra biomes (steppe tundra and shrub tundra), temperate grassland, desert, and several forest biomes such as cool conifer forest and evergreen taiga was improved (Fig. 1; Tables 3 and 4). The model benefited from restrictions to the bioclimatic limits of key PFTs (Table 1) compared to the original model setup (Kaplan, 2001; Kaplan et al., 2003). The narrowed bioclimatic range ensured that the biome boundaries were more accurate at regional scales, although these parameters may not be accurate at global and continental scales.
No model is perfect, and the global vegetation model BIOME4 has its limitations. The BIOME4 model uses 13 PFTs to characterize 28 biomes throughout the world (Kaplan, 2001), but this number of PFTs is insufficient to properly characterize non-forest biomes. Several key shrub types on the Tibetan Plateau exist: subalpine evergreen/deciduous broadleaved and coniferous shrubs, semi-tree shrub, semi-shrub, dwarf shrub, and cushion dwarf shrub in subalpine and alpine environments (ECVC, 1980; EBVAC, 2001) (Table 2). Conversely, the BIOME4 model only has one tundra shrub PFT and one woody desert PFT, which do not sufficiently separate different shrub vegetation on the Tibetan Plateau. Different herbaceous plant types also occur on the plateau, such as the dominant species of Stipa, Festuca, Carex, Kobresia and cushion forbs from various steppe, meadow, and sparse vegetation (Table 2). Three herbaceous PFTs (temperate grass, cold herbaceous, and lichen/forb type) were used in BIOME4, but the limited numbers of herb PFTs did not accurately characterize the key bioclimatic features of different herb types (e.g. precipitation limits of graminoids mentioned above), and so they could not successfully separate different herb types-based biomes.
Sharing key PFTs among biomes is a key reason for mismatches among certain biomes, for example among the five tundra biomes. This is mainly due to the mixture of shrub, grass, and forb PFTs among these tundra biomes, and the model could not differentiate between them under similar bioclimatic conditions. Desert and tundra biomes contain similar grass and shrub PFTs (temperate grass, desert woody shrub, tundra shrub, and cold herb). They were also wrongly simulated to each other in some places. Several conifer and mixed forests had the same situation because they share conifer PFTs and they are also distributed in a transitional zone between forest and grassland at high altitudes. Incorrect simulations of temperate, cool-, cold-, and warm-temperate forests are mainly due to the shared deciduous and conifer PFTs.
The BIOME4 model uses broad climate constraints to differentiate between biomes (Kaplan, 2001). Previous work on arctic biomes resulted in incorrect predictions of forest and dwarf-shrub tundra biomes due to problems with regional climate data and bioclimatic constraints (Kaplan et al., 2003). We improved the climate parameters of some key PFTs in BIOME4-Tibet, but the rules for assigning biomes based on their differences in bioclimate and optimal PFT-based LAI and NPP (Kaplan, 2001; Kaplan et al., 2003) were not changed. This may have resulted in some of the incorrect simulations.
Permafrost and snow cover set important limits on vegetation distribution, especially in arctic and alpine regions (Vaganov et al., 1999; Wang et al., 2006). The BIOME4 model has built-in mechanisms for simulating permafrost and snow processes, but improvements to these modules are needed to more precisely characterize the permafrost and snow–vegetation relationships in Tibet. In addition, soil properties derived from the global data set at 0.5 degree resolution are relatively coarse. Song et al. (2005) used a digitized soil texture data set from an old version of Soil of China (Xiong and Li, 1987), but estimations of soil texture-based water-holding capacity and percolation rate are not available. More detailed and accurate data from regional resources, for example the new Chinese soil database (Shi et al., 2004), are required for future study.
We used regional climate data to drive the model, but gaps in the distribution of weather stations in central and western Tibet may be responsible for some incorrect simulations. There are 207 weather stations in the study area, but almost all of them are distributed in central, eastern, and the northern margins of Tibet and are mainly below 4500 m. In the central and western areas there are only three stations, and the sparsity of weather stations might lead to the incorrect interpolation of climate data. From a similar example, winter temperature was not accurately represented in Alaskan arctic climate interpolation in large valleys and basins with a sparse density of climate data, but the interpolated maps were still consistent with regional climatology (Fleming et al., 2000). We used the same interpolation method as Fleming et al. (2000), so the climate data should be sufficiently precise for regional vegetation modeling.
The BIOME4 model uses global definitions for five arctic tundra biomes (Kaplan et al., 2003). Tibetan vegetation was assigned to tundra biomes based on the global classification of tundra (Kaplan et al., 2003; Walker et al., 2005). However, there is some debate among most of ecologists in China who do not agree that tundra biomes are widely distributed on the Tibetan Plateau. The Tibetan vegetation was called “alpine” or “high-cold” vegetation (mainly alpine meadow, shrubland, steppe, and desert) rather than “tundra” (ECVC, 1980; EBVAC, 2001). There are three types of tundra vegetation in China: small shrub and moss tundra (Dryas octopetala, Phyllodoce caerulea, and Rhacomitrium canescens) and forb and moss tundra (Papaver pseudo-radicatum, Oxytropis anertii, and Polyprachasrum alpinum) which occur in the Changbai Mountains in northeastern China, and moss and lichen tundra (Cetraria nivalis) situated in the Altay Mountains in northwestern China (ECVC, 1980). They grow in cold and wet high mountains and are thought to be connected to tundra biomes that occur in boreal and arctic zones in Siberia and the Far East (ECVC, 1980). High-cold alpine vegetation on the Tibetan Plateau grows in very cold yet dry conditions except in the southeastern subalpine region, which is different from boreal and arctic habitats. However, based on the global definitions of tundra biomes (Walker et al., 2005), high-cold vegetation on the Tibetan Plateau should be assigned to tundra biomes.
The eco-physiognomy of high-cold vegetation is similar to that of tundra biomes, but their composition and structure are different. For example, the continuous, 50-cm- to 2-m-tall low- and high-shrub tundra in circumpolar regions consists of deciduous shrubs (Alnus, Betula, and Salix), evergreen pine (Pinus pumila in eastern Siberia), and Eriophorum and Sphagnum, sometimes with tussock-forming graminoids (Kaplan et al., 2003; Walker et al., 2005). However, in the eastern and southern areas of the Tibetan Plateau, this biome is mainly composed of evergreen shrubs (Rhododendron, Juniperus, and Sabina) and moss, with some deciduous shrubs (Salix, Dasiphora, Rose, Cotoneaster, and Spiraea). These are mostly 25–50 cm to 1–1.5 m tall, with very few herbs (ECVC, 1980). Steppe tundra, dominated by forbs and graminoids, includes typical taxa such as Artemisia, Kobresia, Brassicaceae, Asteraceae, Caryophyllaceae, Gramineae, and true mosses in arctic regions (Kaplan et al., 2003; Walker et al., 2005). This graminoid and forb tundra has a similar species composition on the Tibetan Plateau (Stipa, Festuca, Poa, Carex, Artemisia, and Polygonum), although there remain issues with the assignment of Kobresia meadows. Kobresia meadow, which could be assigned to graminoid and forb tundra, is normally distributed on the southern slopes between 3200 and 5200 m (ECVC, 1980). However, Kobresia meadow also occurs in conjunction with Rhododendron shrub tundra on the northern slopes at altitudes of 4200–4800 m, such as the K. pygmaea meadow which occupies the largest area on the plateau (ECVC, 1980). This area is described as a shrub-meadow complex in Chinese vegetation (ECVC, 1980). The simulated biome for this area consists mostly of shrub tundra rather than steppe tundra. The erect dwarf-shrub tundra (consisting of Betula, Cassiope, Empetrum, Salix, Vaccinium, Gramineae, and Cyperaceae in circumpolar regions, and Artemisia and Ajania on the Tibetan Plateau) and prostrate dwarf-shrub tundra (consisting of Salix, Dryas, Pedicularis, Asteraceae, Caryophyllaceae, Gramineae, and true mosses in circumpolar regions, and Ceratoides and Rhodiola in Tibet) biomes are very different, and belong to alpine desert vegetation on the Tibetan Plateau (ECVC, 1980; EBVAC, 2001). Cushion forb, lichen, and moss tundra on the plateau (Saussurea and Waldheimia) have a different composition than the same tundra in the Arctic (Papaver, Draba, Saxifragaceae, and Caryophyllaceae) and is often found alongside Kobresia meadow (ECVC, 1980). These issues made the assignment of high-cold vegetation to tundra biomes difficult and in some cases incorrect, and have affected the model-data comparison.
COMPARISONS WITH PREVIOUS SIMULATIONS
Two previous simulations of Tibetan biomes (Ni, 2000; Song et al., 2005) have three features in common: they all adopted the Chinese classification of Tibetan vegetation rather than the global biome scheme; simulated modern biomes were compared to a vegetation map of China at a scale of 1∶4 million published 30 years ago; and the baseline atmospheric CO2 concentration was set at 340 ppmv. Ni (2000) used the BIOME3 model and by improving the climatic constraints (GDD5 and annual precipitation) of seven alpine and subalpine biomes, was able to simulate 11 biomes on the Tibetan Plateau. PFT parameters were not changed. The climate data (1951–1980) and soil data at 10′ resolution came from regional observations in China (Ni, 2000). The BIOME3 model did not include any alpine PFTs such as cold herbaceous and lichen/forb types and, therefore, although there was a good agreement (ca. 62%) between the simulated and observed biomes (Ni, 2000), the simulation was less precise than the present simulations. The BIOME4 model, improved by changing the bioclimatic limits for each PFT (Tcm, Twm, GDD5, GDD0, annual moisture availability α, and dominance class D) and by using a new ranking of PFTs for biome assignment, was able to simulate nine biomes in Tibet (Song et al., 2005). The improved model was driven by PRISM long-term mean climatological data (1961–1990) at 3′ resolution. The soil data used referred to soil texture properties rather than the water holding capacity and percolation index used in the BIOME4 simulations. The two dominant biomes, alpine meadow and alpine steppe, showed reasonable agreement with vegetation maps, while there was good agreement for the alpine desert biome (Song et al., 2005).
The different models, biome classifications, and climate and soil data used in previous studies make it difficult to compare the results with our model, but the simulation using the improved BIOME4 model (Song et al., 2005) produced a general trend in biome change from subalpine conifer forest in southeastern Tibet, alpine meadow and montane shrub steppe and steppe in the eastern central region, alpine steppe in the central of the plateau, to alpine desert and montane desert in the northwest. This pattern is more similar to our simulation than the BIOME3 prediction (Ni, 2000).
CONCLUSIONS
The well-tested equilibrium global vegetation model (BIOME4) was modified by redefining the regional bioclimatic limits of key PFTs in order to simulate biome distribution on the Tibetan Plateau. A detailed comparison between the simulated biomes and vegetation map showed that the improved model (BIOME4-Tibet) did a better job of simulating several alpine biomes such as steppe tundra, shrub tundra, and evergreen taiga. The agreement between modeled and observed biome data increased by 17%, which indicates that the global model has potential for regional vegetation modeling. The modified model can be further used to illustrate the impacts of past and future climate change on alpine vegetation and to asses the role of vegetation in atmosphere–biosphere interactions, including changes in carbon and water cycles.
The large proportion of incorrect simulations mean that further improvements to the model are necessary, in addition to obtaining more accurate regional climate, soil, and vegetation data. Extra shrub and herbaceous PFTs also need to be added to BIOME4-Tibet, and more precise bioclimatic and ecophysiological parameters are required in order to constrain the distribution of PFTs and biomes. Further modification of permafrost, snow, and soil modules is also required.
Acknowledgments
The study was supported by the Priority Programme 1372 of German Research Foundation (DFG, grant no. He 3622/15): Tibetan Plateau: Formation–Climate–Ecosystems (TiP). Another two DFG projects (He 3622/11 and He 3622/6) also financially contributed to this study. The digitized Vegetation Atlas of China was provided by the “Environmental & Ecological Science Data Center for West China, National Natural Science Foundation of China” ( http://westdc.westgis.ac.cn). We thank Jed Kaplan for providing the code of BIOME4, and Han Wang for extracting the Tibetan vegetation map from the Vegetation Atlas of China. Thanks also to two anonymous reviewers for their helpful comments to improve the manuscript.
