The Classification Indices-Based Model for NPP According to the Integrated Orderly Classification System of Grassland and Its Application

Grassland accounts for 25% of the world’s land area and has global significance for climatecarbon feedback [1]. As one of the most widespread ecosystem types, grassland makes a large contribution to carbon sequestration and plays a significant role in regional climates and the global carbon cycle [2]. Grassland net primary productivity (NPP) refers to carbon accumula‐ tion (photosynthesis less respiration) in grassland per unit of time, and can be used as an indicator of the capacity of the grassland ecosystem to accumulate carbon and to support grazing animals. Studies of the NPP have contributed significantly to our understanding of the pattern, process and dynamics of grassland ecosystems. Hence, the accurate estimation of the NPP of grasslands is very important not only for scientifically guiding grassland manage‐ ment but also for understanding the global carbon cycle. Reliable long-term data on grassland NPP are urgently needed to estimate carbon fluxes. Due to their timeand labor-intensive nature, ground measurements of grassland NPP are very limited in temporal and spatial coverage. The precision evaluation of NPP estimated in regional or global scales is always very difficult. Direct measurement of grassland NPP is tedious and not practical for large areas, so it is therefore appropriate to use computer models, calibrated with existing data, to study the spatial and temporal variations of grassland NPP [3].


Introduction
Grassland accounts for 25% of the world's land area and has global significance for climatecarbon feedback [1]. As one of the most widespread ecosystem types, grassland makes a large contribution to carbon sequestration and plays a significant role in regional climates and the global carbon cycle [2]. Grassland net primary productivity (NPP) refers to carbon accumulation (photosynthesis less respiration) in grassland per unit of time, and can be used as an indicator of the capacity of the grassland ecosystem to accumulate carbon and to support grazing animals. Studies of the NPP have contributed significantly to our understanding of the pattern, process and dynamics of grassland ecosystems. Hence, the accurate estimation of the NPP of grasslands is very important not only for scientifically guiding grassland management but also for understanding the global carbon cycle. Reliable long-term data on grassland NPP are urgently needed to estimate carbon fluxes. Due to their time-and labor-intensive nature, ground measurements of grassland NPP are very limited in temporal and spatial coverage. The precision evaluation of NPP estimated in regional or global scales is always very difficult. Direct measurement of grassland NPP is tedious and not practical for large areas, so it is therefore appropriate to use computer models, calibrated with existing data, to study the spatial and temporal variations of grassland NPP [3].
A number of different methods of modelling the dynamics of grassland NPP have been reported. In climate-vegetation models, empirical relationships are developed, allowing NPP to be estimated as a function of climate variables such as temperature, precipitation, and evaporation [4][5][6][7][8][9][10][11][12]. Process models for estimating NPP simulate a series of plant physiological processes and their response to changes in environmental factors, including photosynthesis, autotrophic respiration and transpiration [13,14]. In spite of this, the body of knowledge of basic biological mechanisms to simulate grassland NPP is still limited. Furthermore, process models cannot be used in developing countries because long-term data required for the parameters of the models is not available. Algorithms for estimating NPP by remote sensing may be poorly parameterized if the 'calibration' estimates of NPP are based upon an inappropriate method. In this respect, climate-vegetation models have been drawing much attention and have been widely applied internationally, owing to their simplicity [4]. These models have been shown to yield 'reasonable estimates' of regional or global NPP distribution [4,9,[14][15][16].
Vegetative types and their distribution pattern can be related to certain climatic types in a series of mathematical forms. Thus, the climate can be used to predict vegetative types and their distribution, or the reverse [17]. A growing number of research efforts have demonstrated the importance of climate-vegetation interaction in understanding climate sensitivity and climate change [18]. However, current climate-vegetation classification models mostly simulate the equilibrium state of vegetation types, and do not include NPP [19]. Furthermore, at present there is no vegetation-climate model based on a classification system including the NPP function. Therefore, development of climate-vegetation classification models coupled with NPP is urgently needed in order to evaluate the possible impacts of global change on terrestrial vegetation types [19]. So the focus of future research should be to master the link between NPP and grassland class, and to develop pertinent NPP models under a consistent classification system.

The classification indices-based model according to the Integrated Orderly Classification System of Grassland (IOCSG)
A grassland classification system, named the Integrated Orderly Classification System of Grassland (IOCSG), uses the factors Growing Degree Days (GDD), MAP (mean annual precipitation) and the moisture index (K value) to classify grassland diversity [17]. The theory behind the IOCSG has been developed in the last 60 years since it was first put forward in the 1950s, and it has achieved widespread use in China [17,20]. Liang et al. [20] compared the change in potential vegetation distributions from 1911 to 2000 between the IOCSG, the Holdridge Life Zone [5] and BIOME4 [21]. Their results show that the IOCSG has the advantage of being simple and operational in its simulation of grassland classes, making it by far the best grassland classification system.
The Classification Indices-based Model, dubbed the Holdridge life-zone system [5], and the IOCSG [17] were originally built using eco-physiological features and a regional evapotranspiration model with the elimination of the common variable RDI (radioactive dryness index) by the chain rule. It results in a value for NPP (Mg DM ha -1 yr -1 ) as a function of GDD and the moisture index (K value). Its ecological base is the IOCSG [17,20,[22][23][24][25]. The method of integrating the classification indices of IOCSG to estimate the NPP is of the form: NPP = 0.1GDD ⋅ L 2 (K ) ⋅ e − 13.55+3.17K −1 −0.16K −2 +0.0032K −3 ⋅ (K 6 + L (K )K 3 + L 2 (K )) (K 6 + L 2 (K )) ⋅ (K 5 + L (K )K 2 ) Where GDD is defined as the mean of positive unit-period temperatures with the substitution of zero for all unit-period values below 0 °C. The K value is the ratio of MAP to GDD, which provides an index of biological humidity conditions. K = MAP / 0.1GDD, L (K ) = 0.58802K 3 + 0.50698K 2 − 0.0257081K + 0.0005163874 The Classification Indices-based Model not only has the advantage of being simple, operational, and especially compatible with the IOCSG used to derive NPP from its class, it also responds to dynamic vegetation-climate relationships.

Model validation and model inter-comparison
Model inter-comparison has been used as an alternative method for indirect validation and identification of model weaknesses and inconsistencies where ground observations are lacking [26]. However, a simultaneous comparison of these climate-vegetation models to estimate grassland NPP, using a consistent classification system, has not been attempted on a regional, China-wide or global scale.
The following three performance indicators were used as the model evaluation and performance criteria: (1) the mean bias error (MBE) [24,27,28]; (2) the coefficient of variation of the rootmean-square error (CVRMSE), which is defined as the RMSE normalized to the mean of the observed values; and (3) the forecast efficiency, E [29]. The prediction is considered excellent with the CVRMSE < 10%, good if 10-20%, fair if 20-30%, poor if >30% [30]. CVRMSE is the relative difference between the simulated and observed data, while the Nash-Sutcliffe model efficiency statistic (E) [29], regarded as a measure of the overall fit between observed and predicted values, is the primary criterion.
For the purpose of comparison with model simulations, the observed or predicted NPP was expressed in g C m -2 per year, where 1 g carbon is equivalent to 2.2 g oven-dried organic matter [31][32][33][34].
from plots to estimate landscape-scale effects, something that so far has proven to be a challenge to ecologists worldwide [37].
This study continues our efforts to develop a robust estimation capability of NPP for grassland and, specifically, verifies the grassland NPP estimates from the climate-vegetation models along an altitudinal gradient on Helan Mountain and in the surrounding desert rangeland. The specific research objectives are to: 1) compare field-measured NPP based on field-observed data among different grassland classes along an altitudinal gradient in the Helan Mountain range and the surrounding desert; 2) compare and evaluate NPP estimated from the Miami Model [6], Thornthwaite Memorial Model [38], Schuur Model [15], Chikugo Model [8], Beijing Model [12], Synthetic Model [10,11,39], and the Classification Indices-based Model [22,24,25,27,28] with NPP derived from measurements at eight sites in the region; and 3) evaluate the applicability and reliability of the seven models among these grassland classes.

Study area and site descriptions
The study area is located along the northern slopes of the Helan Mountain and the surrounding Alxa Desert (long. 105°32′ to 105°51′, lat. 38°39′ to 39°19′; alt. 3556 m to1360 m a.s.l.), in western Inner Mongolia (Figure 1). Eight study sites were selected following annual rainfall gradient variations with respect to grassland classes at three altitudes that represent the full range of grassland vegetation, environmental, terrain and soil conditions ( Figure 1). Based on the historical meteorological data among the eight study sites, GDD and the moisture index (K value) were calculated. According to the IOCSG, the grasslands at the eight sites were classified into five classes: Alpine meadow (site 1), Cold temperate-humid montane meadow (site 2), Cool temperate-sub-humid meadow steppe (site 3), Cool temperate-semiarid temperate typical steppe (site 4) and Cool temperate-arid temperate zonal semi-desert (sites 5-8), respectively ( Figure 1).

Plots-based NPP collection
The total observed NPP dataset was the sum of above-and below-ground NPP, following standard methods [3,40,41], and it was based on averages of five duplicated plots at each sampling site from April to October once every month for three consecutive years (2003)(2004)(2005). One hundred and twenty plot-based NPP sets from the eight study sites were obtained [25].

Comparison of the seven models in terms model performance values
The performance of each of the seven models compared and evaluated in this study is also shown graphically by plotting predicted NPP values as a function of measured values, as presented in Figure 2. The biases that each model introduces are reflected at the relative positions of the 1:1 line.

Discussion
Seven different methods were compared in terms of their performance in predicting grassland NPP. The relatively high model efficiency in predicting grassland NPP using the Classification Indices-based Model and Chikugo Model indicates that these models outperform the others. The comparison of models involves the input data requirements from each model into account, as these factors affect the value of model efficiency (E), if they are optimized to maximize E. Generally, models with a large number of input variables and parameters would result in higher values of E than models with a small number of input variables and parameters. When comparing the Chikugo Model with the Classification Indices-based Model, it is evident that the Chikugo Model is more complicated in structure and highly intensive in its data requirements. The Chikugo Model provides insight into the relative importance of individual environmental variables in the evapotranspira-tion process [7,11,42,43]. The model is a mechanistic approach to estimate grassland NPP and always slightly overestimated the actual NPP. Furthermore, the Classification Indicesbased Model is based on the IOCSG and used the classification indices as independent variables. It not only takes into account dynamic grassland classes [17], but also simulates NPP of corresponding grassland classes [22,24,25,27,28]. Thus it can be very easy to implement and understand. These factors support the Classification Indices-based model as a more accurate prediction among grassland classes in this study. So the Classification Indices-based Model was found to be the best choice for the given grassland classes. The model is especially suitable for grassland NPP research and application in many developing/ undeveloped countries or regions that generally lack the detailed and complex observation data required by other models (i.e., BIOME4).
The results presented in this study were not only specific to this region, but more importantly, were specific to the given grassland classes according to the IOCSG approach, which can be scaled up from plots to estimate landscape-scale effects. The ascending order of grassland NPP and its class may have significant implications for grassland succession in Chinese grassland ecosystems with predicted changes in spatio-temporal patterns of precipitation under the influence of global climate change. So there is a need to extrapolate in order to calculate the grassland's response to climate change (for instance changing temperature and rainfall). Continued climate change is expected to result in changes in temperature and precipitation, the same as the changes along the altitudinal gradient in this study area. The changing climate will in turn affect the growth of plants, and result in changes to grassland classes and subsequently NPP. The combined affect of precipitation and temperature, that is, the moisture index (K), is to some extent most important to NPP spatial distribution. Thus, the Classification Indices-based Model has the potential to evaluate the possible effects of climate change on grassland classes and their NPP in the future, in order to improve the accuracy of NPP prediction and reduce the evaluation uncertainty of the possible effects of climate change in the future.

Model validation and modelling the potential net primary productivity of grassland in China
China is located in the southeast monsoon climate of the Eurasian continent, and covers a vast territory, with complicated terrain and distinctive distribution patterns of temperature and precipitation with different climatic zones, ranging from cold temperate to tropical and from moist to extreme drought. In addition, the ascent of the Qinghai-Tibetan Plateau(the so-called third polar region of the world) leads to a special plateau climate and thus forms a special pattern of vegetation distribution, termed as 'plateau zonality of vegetation' to distinguish horizontal zonality and vertical zonality of vegetation. The vegetation distribution of China can be characterized by the latitudinal zonality, longitudinal zonality, vertical zonality, and plateau zonality [44]. Thus, China's climate is unique and forms a special pattern of vegetation distribution. Grasslands account for 41.7% of land area in China [17]. Traditionally, China's grasslands have been economically productive areas and have provided important natural resources for the nation, including meat, milk, wool and animal hides [35]. Grassland NPP can be used as an indicator of the capacity of the grassland ecosystem to accumulate carbon and to support grazing animals. The ability to accurately estimate grassland NPP is critical to our understanding of grassland dynamics [45]. But evaluation of the precision of grassland NPP estimates in China is difficult.
Since it is a developing country, some mechanistic models cannot be used because longterm data required for the parameters of these models are not available in China [16]. But the required data are available for NPP-climate models, and these models have been shown to yield 'reasonable estimates' of global patterns of productivity [4,9,14,16]. A growing number of research efforts [18] are focused on developing NPP-climate relationships. The Miami Model [38], the Schuur Model [15] and the Classification Indices-based Model [22,24,25,27,28] are examples of such efforts. But a simultaneous comparison using a consistent classification system has not been attempted in China.
The purpose of this section is: 1) to compared NPP estimated using the Miami Model [38], Schuur Model [15], and the Classification Indices-based Model [22,24,25,27,28] with NPP derived from measurements at 3767 sites in China, to evaluate the applicability and reliability of the Classification Indices-based Model; and 2) to simulate the spatial distribution patterns and its NPP characteristics of China's potential grassland under current climate scenarios using the IOCSG approach and NPP-climate models.

Observed NPP database
A large reference data set (n = 3767) of observations of grassland NPP with paired climatic variables was compiled for this study. This data set came from surveys conducted by the National Animal Husbandry and Veterinary Service of the Ministry of Agriculture of the People's Republic of China from 2004 to 2005. Total observed NPP dataset from the 3767 sites was the sum of above-and belowground NPP following the standard methods [3,40] and it was based on the average of five duplicated plots at each sampling site. Locations of 3767 sampling sites were plotted using their associated geographic coordinates, as shown in Figure 3.

Climate data
Meteorological data for annual rainfall and GDD at 2348 meteorological stations, including longitude, latitude and altitude data of each meteorological station, were obtained from a database of the China Meteorological Data Sharing Service System from 1961 to 1990 (downloaded from the website: http://cdc.cma.gov.cn). The locations of the 2348 meteorological stations are shown in Figure 4. The Digital Elevation Model (DEM) was derived from the global 1km×1km DEM issued by the U.S. Geological Survey (USGS) (downloaded from the website: http://edcdaac.usgs.gov/gtopo30/ README. asp#h17). Regional data were derived from a database of 1:4 million Chinese regional data points, which was last updated in May 2004, and were used to confirm the Chinese boundary.

Estimating grassland NPP and its distribution patterns in China
Spatialization of GDD and K value was created by introduction into the interpolation analysis of meteorological elements. The 1km×1km grid data of the moisture index averaged over years was obtained by the Kriging interpolation method ( Figure 5). The inverse distance squared method (IDS) [46], modified by DEM in the smaller regions, was used to spatialize GDD data and to improve the interpolation precision of GDD in China [47]. By mosaicking the grid data of GDD on the actual ground of eight smaller areas and clipping the integrated interpolation area by the Chinese regionalism, we finally obtained the 1km×1km grid data of GDD ( Figure  6). The 1km×1km grid data of the moisture index and GDD were used to predict the potential classes recognized by the IOCSG (Figure 7), and were then used to drive the Miami Model, Schuur Model, and the Classification Indices-based Model, respectively. To more explicitly reflect the spatial distribution patterns of potential biomes in China, and for ease of comparison and application, we merged the 42 classes of IOCSG to Grassland and Forest categories. The Grassland category consists of seven grassland super-class groups (biomes): Tundra and alpine steppe, Frigid desert, Semi-desert, Steppe, Temperate humid grassland, Warm desert and Savanna. The Forest category includes Temperate forest, Sub-tropical forest and Tropical forest ( Table 2). The summed NPP has a unit of g C for pixels or Tg C (1 Tg = 10 12 g) for each grassland super-class group (biome) or Pg C (1 Pg = 10 15 g) for Grassland or Forest category in China.        Comparison of the predicted with the observed grassland NPP indicated that the CVRMSE value using the Classification Indices-based Model was <10%, demonstrating that the prediction of the Classification Indices-based Model is considered excellent, while the output of the Miami Model is considered good, with CVRMSE 13.0%. As the CVRMSE value using the Schuur Model is 23.9%, its predictive ability is considered fair. The results of linear regression showed that the Classification Indices-based Model explained, on average, 65.3% (R 2 ) of the variation in observed NPP, while the Miami Model explained 63.0% (R 2 ) of the variation. With the highest E value, the Classification Indices-based Model ranked first for predicting NPP, followed by the Miami and Schuur Models, respectively, indicating that the Classification Indices-based Model can estimate grassland NPP more precisely (Table 3).

Distribution patterns of grassland NPP in China
Almost all classes (42 classes) occur in China, except for tropical-extra-arid tropical desert (VIIA7) under current climate conditions ( Figure 7). The spatial distribution of grassland biomes ( Table 2) under current conditions is shown in Figure 9a. The Grassland category was predicted to be distributed primarily in the arid and semi-arid regions of Northern China, the Qinghai-Tibetan Plateau, and scattered throughout warm-temperate and tropical regions ( Figure 9a). From north to south, grassland biomes were predicted to be mainly distributed as follows: Tundra and alpine steppe (accounting for 20.42% of Chinese land area); Cold desert (13.1%); Semi-desert (8.1%); Steppe (8.0%); Temperate humid grassland (6.0%); Warm desert (0.1%) and Savanna (0.1%) ( Figure 9a and Table 4). These percentages correspond to the actual grassland distribution.
Compared with the Classification Indices-based Model and Miami Model, the Schuur Model estimated substantially lower TNPP, especially in Tundra and alpine steppe, Cold desert, Semi-desert, Steppe, Temperate humid grassland and Savanna areas (Table 4), but the simulated spatial pattern for grassland NPP is very similar (Figure 9b, c, d). Comparing the NPP maps (Figure 9b, c, d) with the data in Table 4 shows highest TNPP of grassland biomes in Tundra and alpine steppe, followed by Steppe and Temperate humid grassland, respectively. The Tundra and alpine steppe, which covers the largest area, was predicted to be mainly distributed in the Tibet Autonomous Region (94.4 million hectares), Qinghai Province (49.1 million hectares) and Xinjiang Uygur Autonomous Region (32.1 million hectares). Although Tundra and alpine steppe has lower NPP values than Temperate humid grassland on a unit area basis, the TNPP for Tundra and alpine steppe, estimated by the Classification Indicesbased Model, was predicted to be nearly twice as large as that of Temperate humid grassland because Tundra and alpine steppe was predicted to cover a greater area ( Table 4). The total area of Steppe and Temperate humid grassland was predicted to be 130.75 million hectares, accounting for 25.0% of the total grassland area, with nearly half of distributed on the Inner Mongolian Plateau. Savanna and Warm desert super-classes were predicted to cover less than 1.3 million hectares. Savanna was predicted to be located in southern China, particularly in Hainan Province, and Warm desert was predicted to be located in the Tarim Basin in Xinjiang Uygur Autonomous Region and the Qaidam Basin in Qinghai Province ( Figure 9a). Warm desert and Savanna were predicted to have the lowest TNPP, and Semi-desert and Cold desert to have moderate TNPP values ( Figure 9 and Table 4). The Semi-desert was predicted to be mainly distributed in Xinjiang Uygur Autonomous Region ( Table 3. The mean bias error (MBE), coefficient of variation of the root mean square error (CVRMSE), coefficients of determination (R 2 ), and model efficiency (E) for the comparison of the predicted with the observed grassland NPP in China.
The Classification Indices-Based Model for NPP According to the Integrated Orderly Classification System of... http://dx.doi.org/10.5772/57297 In conclusion, under current climate conditions, the main parts of China's grassland are predicted to be the Tundra and alpine steppe and Steppe, and they account for 50.88% China's total grassland and 63.77-63.90% China's grassland TNPP (Table 4).

Model-data comparison
The performance of the different models can be compared on the basis of their error frequency distributions. Figure 10 shows the distribution of errors for the three models. Even though the pattern is to some extent controlled by the heterogeneous distribution of the validation points, the direct influence of the different models on the error structure remains evident. In the Miami Model, the error distribution ( Figure 10a) shows a unimodal trend with a peak centered on a negative value (average local error of about -48 g C m -2 per year) and an error range from -419 to 492 g C m -2 per year. The largest underestimated values occur with the Schuur model, although a few overestimated predictions are also present in the same areas. In the histogram of error frequency (Figure 10b), a large single maximum is clearly centered on values of approximately 132 g C m -2 per year, indicating that grassland NPP is underestimated almost everywhere within the test area ( Figure 8). The fact that the Schuur Model produced a model efficiency value of 0.58 is an indication that any predictions arising from this model must be treated cautiously. The Classification Indices-based and Miami models have more consistent behavior throughout the validation points, and Figure 10c indicates that the results of the Classification Indices-based Model are in good agreement with observed data (mean error 23 g C m -2 per year with a standard deviation of 105.6 g C m -2 per year) and the mean absolute error is lower than that obtained from the Miami Model. The Classification Indices-based Model performs slightly better than the Miami Model, with a narrower error range and an error histogram that approximates a bell-shaped distribution ( Figure 10). In general, the distribution of residuals is homogeneous, although a few larger errors occur in the Classification Indices-based model, which can compromise its reliability. Comparing the Miami Model with the Classification Indices-based Model, the Miami Model determines NPP for a particular location by comparing the minimum value of either temperature or precipitation functions. Hence, it only considers a single factor, and there are no interactions between the two variables. In contrast, the Classification Indices-based Model considers three meteorological measurements and their interactions in its formulation. The reduction in errors produced by the Classification Indices-based Model should lead to significant improvements in applications as a newer NPP-climate relationship model ( Figure 10).

Grassland carbon budgets
The Classification Indices-based Model has two advantages: 1) it needs few input parameters; and 2) remote sensing data can be incorporated easily. Consequently, the Classification Indices-based Model is effective and practical for large-area application. But the simulated NPP by the Classification Indices-based Model was ideally potential value and land-use practices were not taken into account. The potential grassland, as a final state of succession which achieves the balance with its habitat, is the most stable and mature climax grassland class of the habitat without human interference, and is the trend of the regional grassland development. The study of potential grassland can substantially reveal the impact of climate on the changes of grassland patterns. It is the starting point of the vegetation-environmental classification and relationship study, as well as the key point of the global change and grassland ecosystem study [27]. and1978 and also between 1979 and 1990. This wide range in the reported potential TNPP could be due to their use of different estimation methods, inconsistent classification systems, and/or different time periods with changing grassland areas.

Human Appropriation of Net Primary Production (HANPP)
One of the central issues of sustainable development is the question of how to meet the globally growing demand for biomass in an ecologically sound way. Land use activities, primarily for agricultural expansion and economic growth, have transformed one-third to one-half of our planet's land surface from forest or grassland clearance, agricultural practices and urban expansion, which make profound impacts on ecosystem service, food production and the environment [52]. China, occupying only 7% of the world's total arable land, supports nearly 21% of the world's total human population. Agriculture is the major sector of growth in the Chinese economy and with 80% of Chinese living in villages, a large proportion of these people are still dependent upon agriculture. However, with rapid industrialization, urbanization is accelerating and leading to changes in grassland and forest land uses. With increasing scientific and political interest in regional aspects of the global carbon cycle, there is a strong impetus to better understand the carbon balance of China. This is not only because China is the largest emitter of fossil-fuel CO 2 into the atmosphere, but also because it has experienced regionally distinct land-use histories and climate trends, which together control the carbon budget of its ecosystems [53,54]. Changes in terrestrial ecosystems through human land use and management are also important in land degradation and carbon balance studies. Environmental degradation is a major concern for China and research aimed at understanding the linkages between land use changes, socioeconomics and the biophysical variables governing environmental degradation is needed. Despite the vast amount of research conducted independently on land use changes and social indicators in China, an integrated assessment combining them to study the socioeconomic metabolic flows and their impact on land resources on a large spatial scale is absent. In this context, the concept of HANPP (human appropriation of net primary production) is an important tool for assessing the impact of anthropogenic forces through land use changes on biomass.
HANPP is an indicator of the extent to which ecosystem processes are altered by human activities. HANPP equates to the difference between the NPP of the potential natural vegetation and the amount of NPP remaining in ecosystems. The notion of HANPP represents the food supply for humans, providing the free energy for humans to sustain their metabolic activities and allowing them to perform physical work. For example, agriculture and forestry harness biomass energy for socio-economic purposes and thereby reduce the amount of NPP remaining in ecological food chains [52,[55][56][57][58]. HANPP alters the composition of the atmosphere, levels of biodiversity, energy flows within food webs and the provision of important ecosystem services [59][60][61]. This removal of carbon naturally comes at a cost in that increased intensities of human appropriation should directly result in less vegetative growth compared to its natural state, if all other factors are held constant, such as nutrient availability. In this study, we present a comprehensive assessment of Chinese HANPP based on vegetation modelling using our model, agricultural and forestry statistics, and geographical information systems data on land use, land cover, and soil degradation that localizes human impact on ecosystems. We find, overall, that the HANPP values suggest that more than 43.2% of the available NPP has already been modified and exploited in China. This value is comparatively higher than the range of global estimates of 20% to 40% for HANPP [62,63]. Furthermore, compared with HANPP estimates cited for Austria [57,59], where HANPP declined from 1980 to 1995, HANPP has increased in China. Increases in HANPP may lead to carbon fluxes from biota to the atmosphere, they may contribute to biodiversity loss, and they may result in diminished resilience of ecosystems [62]. As Chinese human population grows and with mounting political interest in increasing GDP, there is likely to be an increased requirement for more agricultural areas and further increases in HANPP over the present value. These analyses suggest policy options for slowing future growth of HANPP. It also offers a means to aid people in using the aforementioned carbon sinks to fulfill China's commitment of reducing greenhouse gases.

Model-data comparison and modelling the potential net primary productivity of global grassland
Grassland biomes account for 25% of the world's land area and have global significance for climate-carbon feedback [1], containing approximately 30% of global soil carbon stocks [64].
Ranging from the savannas of Africa to the North American prairies and the converted grasslands of Latin America and Southeast Asia, grassland ecosystems support the majority of the world's livestock and large mammals [65], besides playing a very important role in regulating the global carbon cycle and providing meat and milk for human beings.
As one of the most widespread ecosystem types, natural grassland plays a significant but poorly recognized role in the global carbon cycle [1,2]. In order to effectively manage grassland ecosystems and maintain their sustainability, large-scale analysis and modelling of grassland NPP are needed to develop a better grasp of the spatial distribution of grasslands and their productivity [3,4,13,20,66]. Continuous monitoring of global grassland productivity has never been possible because of technological limitations. So it is necessary to use computer models, calibrated with existing data, to study the spatial and temporal variations of grassland NPP [3]. Recently, climate-vegetation models have been drawing much attention and have been widely applied internationally [4,9,18], having been shown to yield 'reasonable estimates' of global patterns of productivity [4,9,14,16]. The Miami Model [38], Schuur Model [15] and Classification Indices-based Model [22,24,25,27,28] are examples of these climate-vegetation models. But model inter-comparison under a consistent classification system simultaneously has not been attempted. In particular, the quantitative and spatial descriptions of global grassland classes and their NPP values have not been reported.
The purpose of this section is: 1) to compare NPP estimated from the Miami Model [38], Schuur Model [15], and the Classification Indices-based Model [22,24,25,27,28] with NPP derived from measurements at 37 sites around the world, to evaluate the applicability and reliability of the Classification Indices-based Model; and 2) to simulate the spatial distribution patterns and associated NPP characteristics of global potential grassland under recent past climate scenario using the IOCSG approach and NPP-climate models.

Observed NPP database
A reference data set (n = 113) of grassland NPP field observations with paired climatic variables was compiled for this study. Total observed NPP data were gathered from the Oak Ridge National Laboratories (ORNL) Net Primary Production database (http://www-eosdis.ornl.gov/NPP/npp_home. html), and were the sum of aboveground and belowground NPP following standard methods [9]. The grasslands including the Tundra and alpine steppe, Cold desert, Semi-desert, Steppe, Temperate humid grassland, Warm desert and Savanna, or other intensively managed sites were omitted from this study. Locations of sampling sites were plotted using their associated geographic coordinates, as shown in Figure 11.

Monthly precipitation and mean temperature grid data
The monthly precipitation and mean temperature grid data, with spatial resolution of 30 arc seconds (i.e., about 1 km), for global land areas excluding Antarctica over 50 years from 1950 to 2000, were generated using the software package ANUSPLIN Version 4.3 (Australian National University, Canberra, ACT, Australia) [67,68]. We chose this data set because the method that created it has been used in other global studies [69,70], and it has performed well in comparison with the other multiple interpolation techniques [68,71].

IOCSG approach and NPP-climate model operation
A common approach to extrapolating grassland NPP to the biosphere is to use vegetation maps together with models of plant productivity at a biome level. The IOCSG potential biome maps and its NPP were created and processed by the ArcGIS software (Esri Inc., Redlands, CA, USA) under current climate . In the IOCSG biome (super-class group) classification map, lake water, permanent snow and ice were excluded by using the Moderate Resolution Imaging Spectroradiometer-International Geosphere Biosphere Program (MODIS-IGBP) land-cover classification dataset in 2001, found at http://wist.echo.nasa.gov. The MODIS-IGBP land-cover classification dataset was used to calculate the area of potential grassland biome. The simulated NPP was an ideal potential value, and land-use practices were not taken into account. The total NPP has a unit of Tg C (1 Tg = 10 12 g) for biome or Pg C (1 Pg = 10 15 g) for world.

Spatial distribution patterns and characters of grassland biomes over 1950-2000
The spatial distribution of grassland biomes during the period of 1950-2000 is shown in Figure  13a. From Figure 13a, the potential grassland is characterized by a significant distribution pattern in latitudinal and altitudinal directions. From the Equator to the North Pole, grassland biomes were mainly distributed in sequence as follows: (1)   Comparisons with the Classification Indices-based Model and Miami Model, Schuur Model estimated substantially higher TNPP, especially for Temperate humid grassland, Steppe and Warm desert (Table 6), but the simulated spatial pattern for global NPP is very similar ( Figure  13bcd). The NPP increases from the South and North poles toward the Equator, corresponding to changes in precipitation and temperature. Comparing the NPP maps (Figure 13bcd) with the Table 6 shows highest TNPP of grassland in Savanna, followed by Tundra and alpine grassland and Steppe, respectively. Although Warm desert grassland covers the second greatest area among grassland super-classes, TNPP for Warm desert grassland is still lower due to lower NPP.
In conclusion, under recent past climate conditions, the main parts of global grassland are the Savanna and Tundra and alpine grassland, which together account for 42.15% of total global grassland area and 63.76-68.91% of global grassland TNPP (Table 6).  The Classification Indices-Based Model for NPP According to the Integrated Orderly Classification System of... http://dx.doi.org/10.5772/57297

Model-data comparison
GDD, MAP and the moisture index (K value) are used as criteria to determine the class in the IOCSG. The Classification Indices-based Model connected with the IOCSG by using the classification indices as independent variables and considers three meteorological measurements and their interactions in its formulation. The use of these variables allows the Classification Indices-based Model to more accurately predict changes of NPP among grassland classes or super-classes [17,22,24,25,27,28]. The model's relative simplicity and ability to make reasonable estimates of the patterns of NPP in global grasslands is attractive. Despite being a newer NPP-climate relationship model, the Classification Indices-based Model provides more accurate conclusions and may also reduce the uncertainty of climate change impacts on grasslands on a global scale. Furthermore, it has the potential to rapidly advance grassland NPP research and application in developing/undeveloped regions or countries that generally lack the detailed and complex data required by other models (i.e., BIOME4).

Grassland carbon budgets
The Classification Indices-based Model estimates global TNPP at 54.51 Pg C compared to 55.38 Pg C and 61.52 Pg C using the Miami model and the Schuur Model under recent past climate conditions ( Table 6). The Classification Indices-based Model estimate is closer to the estimates of 46 Pg C, 48 Pg C and 55 Pg C by the National Center for Ecological Analysis and Synthesis (NCEAS) model [4], the Carnegie-Ames-Stanford approach (CASA) Biosphere model [72], and the MODIS NPP [73], respectively. This wide range in the reported potential global TNPP could be due to their use of different estimation methods, inconsistent classification systems, and/or different time periods [74]. The potential carbon accumulated in plants was calculated based on simulated NPP. This calculation does not account for the potential impacts from human disturbances, but assumes that the full potential NPP is reached in the grassland.  [75] and assess the human appropriation of net primary productivity.

Introduction
Global climate change will seriously affect terrestrial ecosystems. In global climate change research, modelling is a key method for the study of climate-vegetation, which is an important component in the area of global change research. The current period of climate change and high CO 2 concentrations in recent years has resulted in warm summers and longer growing seasons, to a change in vegetation patterns can be expected. This change would cause changes in vegetation classes and their NPP. For example, the dominance of some cold-season C 3 grasses would decrease, while that of warm season C 4 grasses could be expected to increase. Global warming also causes changes in precipitation patterns and the total amount of precipitation. The far-reaching consequences of climatic phenomena such as La Niña and El Niño receive widespread attention and appear to be occurring more frequently in recent decades. Zhang et al. [26] recently showed that global warming is already affecting the world's rainfall patterns, bringing more precipitation to northern Europe, Canada and northern Russia, but less to sub-Saharan Africa, southern India and Southeast Asia. They explained that human activity is causing alterations to the water cycle, moving more water vapor away from the warmest parts of the planet and pushing it toward the poles. As a result, wet areas are becoming wetter and dry areas are becoming drier. Thus, there has been an increase in the demand by the broader scientific community and policymakers for better projections of regional impacts of future greenhouse-gas-induced climatic change. Global environmental change, signified by 'global warming' and its possible effects on ecosystems have been drawing increasing attention from the scientists and governments of every country and from all circles of the world [9,13,76]. Study of responses of terrestrial NPP to climate changes will help us understand the feedback between climate systems and terrestrial ecosystems and the mechanisms of increased NPP in the northern, middle and high latitudes [77]. NPP is not only used for assessing the carbon balance on regional and global scales, it also plays an important role in demonstrating compliance with the Kyoto Protocol on greenhouse gas reduction [11]. So, accurate estimates of NPP are very important not only for scientifically guiding ecosystem management but also for the study of global climate change [53]. Therefore, study of NPP and its response to global change is one of the key focuses for the global scientific community [9,78].
Climate change has been identified as having far-reaching implications for the world's grasslands [16,[79][80][81][82][83]. Therefore, understanding the sensitivity of grassland to climate change and the effect of these changes on grassland ecosystems is a key issue in global carbon cycling. NPP in grassland is a key variable in our understanding of carbon exchange between the biosphere and atmosphere, both currently and under climate change conditions [3,4,13,84]. The mechanisms and modelling of responses in grassland NPP to global change will enhance understanding of how grassland ecosystems will respond to global change and the projection of grassland NPP to global change. This will enable the development of countermeasures to maximize the grassland NPP under conditions of global change, especially to provide sustainable development of pasturing in arid and semiarid areas, by using these theoretical and technological methods. The IPCC report [85] states that climate change will present challenges for future grassland use by livestock and other grazers, and for the resulting public policy designed to manage grasslands. Given their vast area and diversity of classes, China's grasslands play an important role in both regional and global carbon cycling [33]. In order to effectively manage grassland ecosystems and maintain their sustainability, a deeper understanding of how these ecosystems will respond to growing pressures is needed. Large-scale analysis and modelling is needed to develop a better grasp of the spatial distribution of grasslands, their productivity, and potential variations in response to climate changes [3,4,13,20,23,66].
Climate change will have profound effects on the class level, which in turn will influence vegetation types and their NPP. Clearly, this requires improved projections of regional climate and understanding of how vegetation types will respond to climate change. Climate is a major driver of variation in NPP, but a clear understanding of the impact of climate change on NPP is lacking [13]. Not only are grassland researchers concerned with the performance of NPP simulation models under contemporary climate, but also they want to know the behavior of these models under extrapolated future environmental conditions [3,4,13]. A growing number of research efforts have demonstrated the importance of climate-vegetation interaction in understanding climate sensitivity and climate change [18]. The study of climate-vegetation interaction is the basis for research on terrestrial ecosystems' responses to global change and mainly comprises two important components: climate vegetation classification and NPP of natural vegetation. From the view of generic relationships among all the vegetation types, the IOCSG can be used to predict climate-linked spatial or temporal succession from an original class to a new class as these climatic factors change. Hence, there is potential to develop future scenarios based on possible changes in vegetation type and its NPP in response to climate change in coming years. The IOCSG is a suitable tool to incorporate the effects of both climate change and vegetation management measures on plant growth activity into NPP estimation at the same resolution as satellite data. The Classification Indices-based Model connects with the IOCSG by using the classification indices as independent variables. The use of these variables allows the Classification Indices-based Model to project changes of NPP among grassland classes or super-classes more accurately [17,22,24,25,27,28]. The section on Model Validation and Model Inter-comparison has shown that the Classification Indices-based Model is suitable to predict grassland NPP at regional, Chinese or global scale. In addition, the Classification Indices-based Model presented the closest values to the observed in situ meteorological variables. Furthermore, its relative simplicity and ability to generate reasonable patterns of NPP is attractive. Also, GDD and the moisture index (K value) are used as proxies to represent the water and thermal properties of the growing season, and are effective drivers for modelling NPP. The Classification Indices-based Model not only takes into account dynamical classes [17], it also simulates NPP of corresponding classes. As a new NPP-climate relationship model, the Classification Indices-based Model has the potential to evaluate the possible effects of climate change by improving the accuracy of the NPP prediction and reducing the evaluation uncertainty of the possible effects of climate change [22,24,25,27,28].
In order to effectively model climate change impacts on global grassland distributions and associated NPP, it is important to understand climate dynamics in the recent past  as well as climatic projections for the future (2001-2050). Hence, in this section we: (1) simulate the spatial distribution patterns and associated NPP characteristics of the world's and China's potential grassland under a future climate scenario using the IOCSG approach and the Classification Indices-based Model; and (2) estimate future trends in response to climate change in the first half of the 21st century by comparing the variation in distribution of NPP with the potential total NPP (TNPP) of grassland between the recent past and the projected future climate scenario. Such findings should improve knowledge about changes in global grassland primary productivity under global warming.

Climate data
Two different global climatic datasets were used in this study. The first was the monthly precipitation and mean temperature grid datasets over 50 years from 1950-2000, described at section of 3.3.1.2. The second dataset was the global monthly precipitation and mean temperature prediction data set (also with 30 arc-second resolution) excluding Antarctica from 2001 to 2050 under the A2a scenario (see below), which was simulated by Australia's Commonwealth Scientific and Industrial Research Organisation (CSIRO) and is available through the website http://www.worldclim.org /futdown.htm [86]. The A2a scenario takes into consideration the following land-use changes: (1) high rate of population growth; (2) slow technological change; and (3) increased energy use, and describes a highly heterogeneous future world in light of regionally-oriented economies. The gridded baseline climate (recent past) and gridded future (2000-2050) A2a climate scenarios were used to estimate changes in mean annual temperature (MAT), growing degree-days (GDD) and mean annual precipitation (MAP).

IOCSG approach and NPP simulation
The IOCSG potential super-class group (biome) maps and its NPP were created and processed by the ArcGIS software (ESRI Inc., Redlands, CA, USA) under the baseline climate  and future climate A2a scenario (2000-2050) on average level. In the IOCSG biome (super-class group) classification map, lake water, permanent snow and ice were excluded by using the Moderate Resolution Imaging Spectroradiometer-International Geosphere Biosphere Program (MODIS-IGBP) land-cover classification dataset in year 2001, found at http://earthdata.nasa.gov. The MODIS-IGBP land-cover classification dataset was used to calculate the area of potential grassland biome. The simulated NPP by the Classification Indices-based Model was an ideal potential value and land-use practices were not taken into account. The recent past map was chosen as a baseline and was used for comparison with projected future climate to estimate trends in changes of class or grassland biome distribution and grassland NPP distribution in response to climate change. Both the IOCSG and the Classification Indicesbased Model were applied to analyze the changes in grassland biomes and to measure the change in TNPP of grassland biomes from the recent past  to the future A2a scenario (2001-2050). The total NPP has a unit of Tg C (1 Tg = 10 12 g) for biome or Pg C (1 Pg = 10 15 g) for China or the world.
In conclusion, the main grassland biomes found in the world were the Savanna, Steppe and Semi-desert, occupying 50.86% of total global grassland areas and 78.06% of global grassland TNPP under future A2a scenario (Table 7).

Numerical simulations of potential grassland dynamics
Both the IOCSG approach and the Classification Indices-based Model were applied to analyze the succession of classes and super-classes and to measure the change of super-class TNPP from recent past  to future A2a scenario (2001-2050).

Change of class and super-class area
The class called Frigid-extrarid frigid desert, alpine desert (IA1), mainly distributed in America and Asia, continuously decreased from recent past to A2a scenario and the rate of decrease per decade was 4.72%. If decreasing rates persist, Frigid-extrarid frigid desert, alpine desert (IA1) could potentially disappear by the second half of this century, while the class called Cold temperate subhumid montane meadow steppe (II D23), mainly distributed in Asia and America, continuously increased from recent past to A2a scenario, by 18.40%.
Climate change also influences the spatial distribution of super-classes. The distribution area for various super-classes in these two time periods shows differing dynamic characteristics from recent past  to future A2a scenario (2001-2050) ( Tables 6 and 7). The area of Tundra and alpine steppe will decrease by 53.26 million hectares compared with its size during 1950-2000. Likewise, Warm desert will decrease by 29.97 million hectares. However, trends show that Savanna and Temperate humid grassland will strongly increase, to nearly 123.16 million hectares. Overall, the Forest category will decrease by 29.29 million hectares (Tables  6 and 7).

Shift trend of super-classes
The A2a scenario predicts changes in global temperature and rainfall patterns, making wet areas wetter, and dry areas drier (Figure 14), resulting in class conversions from an original class to a new class due to changes in climate conditions. For example, the Frigid perhumid rain tundra, alpine meadow (IF36) class will change to Cold temperate-humid montane meadow (IIE30) with climate warming, or Cold temperate perhumid taiga forest (IIF37) with climate drying. The conversion of classes from recent past to future A2a scenario will result in changes in area and TNPP distribution patterns among biomes (Table 8 and Figure 15a). For example, in the super-class of Tundra and alpine steppe, an area of 656.09 million hectares will be converted to Forest (Legend 18 in Figure 15a and Table 8) in this long-run sequence, which will mainly take place in Asia (accounting for 58.96%) and America (32.21%), while another 182.92 million hectares will be converted to Temperate humid grassland (Legend 15 in Figure  15a and Table 8), a change that will mainly take place in Asia (65.90%). An area of 590.06 million hectares of Forest under recent past climate conditions will be converted into the Temperate humid grassland under A2a scenario (Legend 85 in Figure 15a and Table 8), mainly taking place in Asia (51.14%), the Americas (26.22%) and Europe (22.58%). An area of 389.29 million hectares of Forest will be converted into Savanna (Legend 87 in Figure 15a and Table 8), mainly taking place in the Americas (49.62%) and Africa (35.52%). An area of 253.28 million hectares of Temperate humid grassland under recent past climate conditions will be converted into Steppe under the A2a scenario (Legend 54 in Figure 15a and Table 8), mainly taking place in Asia (48.76%), the Americas (28.89%) and Europe (22.22%). An area of 243.37 million hectares of Steppe will be converted into Semi-desert (Legend 43 in Figure 15a and Table 8), mainly taking place in Asia (49.72%) ( Table 8).
An area of 856.54 million hectares of Grassland category will be converted to Forest category and an area of 1089.77 million hectares of Forest category will be converted into Grassland category (Figure 15b. and Table 8). There is a clear increasing trend in the Grassland category. As a whole, areas of Grassland category will increase by 31.76 million hectares, while areas of Forest category will decrease by 29.29 million hectares. The relative area of Grassland category will increase by 0.43% over this period , whereas the relative area of Forest category will decrease by 0.51% over the same period (Tables 6 and 7).

Change of TNPP
The Classification Indices-based Model shows NPP either neutral or increasing for most global grid cells during the periods from recent past to A2a scenario ( Figure 16).
Note: The legends show the differentiated NPP and negative values indicate C loss for NPP.  Note: The class name refers to Figure 7 Explanation.Unit of TNPP is TgC.

Biome conversions
The changing climate will affect the growth of plants, alter the species composition of plant communities and ultimately change community structure [87]. Under these conditions, the classes will shift [3,5] and result in biome conversions [3,5]. Our analysis indicated that Steppe will experience the largest variation in biome conversions. The result is the same as the conclusion of Liu et al. [88], who confirmed that under climate warming, the temperate steppe in the arid and semiarid regions of northern China will act as a net C source. The Savanna and Temperate humid grassland will have the highest percentage of growth. The Tundra and alpine steppe biome will have the greatest change in total area, followed by Forest (Tables 6, 7 and 9).

Trends in grassland NPP in response to climate change
The Classification Indices-based Model estimates global potential grassland TNPP at 16.47 Pg C and global TNPP at 57.67 Pg C under future climatic A2a scenario (Table 7). Climate change is expected to affect the TNPP of grassland. Understanding climate and NPP feedback cycles in grasslands requires knowledge of the changes in spatial patterns on the NPP of grasslands. The dynamic change characteristics of potential biomes (Table 8 and Figure 15) reflect further evidence of increasing global warming since the early 1980s (Intergovernmental Panel on Climate Change, 2007). The Classification Indices-based Model estimated an increase of 1.09 Pg C in TNPP for global potential grassland from recent past to future A2a scenario (Tables 6, 7 and 9). It will bring an increase of 3.91 Pg in annual sequestration of CO 2 and 2.88 Pg of annual contribution to the atmospheric O 2 cycle respectively (1.63 g of CO 2 is absorbed and 1.2 g of O 2 is released to generate 1 g of DM according to photosynthesis stoichiometry; [89]), which will impose a new issue for future grassland research to support sustainable development and to provide relevant knowledge to meet the challenge of climate change [90].

Characters of potential grassland over 2001-2050 under A2a scenario
In the A2a scenario, the Tundra and alpine steppe, Cold desert, Semi-desert, Steppe, Temperate humid grassland, Warm desert, and Savanna were projected to occupy 8.9%, 8.8%, 11.1%, 12.0%, 5.9%, 6.2%, and 1.8% of land area of China, respectively (Table 10). Under scenario A2a, the Classification Indices-based Model projected the highest TNPP to be in the Forest category. Table 10 shows that the highest TNPP values of grassland biomes were projected to be in the Steppe, followed by Semi-desert, Tundra and alpine steppe, and Temperate humid grassland, respectively. In conclusion, it was projected that the two main grassland biomes found in China will be in the Steppe and Semi-desert, and they will occupy 42.2% of China's total grassland areas and 58.3% of China's Grassland TNPP in the future climatic A2a scenario (Table 10).

Comparing the area and TNPP of China's major terrestrial biomes of 1950-2000 with those of 2001-2050 under the A2a scenario
Climate change was projected to influence the spatial distribution of grassland biomes, with a combined distribution ranging from the northeastern plain adjacent to Mongolia to the southern Tibetan Plateau, with warm desert showing the greatest discrepancies. The distribution of the area for various grassland biomes in these two time periods was projected to change considerably from the recent past (1950-2000) ( Table 4) to the future (2001-2050) in the A2a scenario (Table 10). The area of Tundra and alpine steppe was projected to decrease by 107.9 m ha compared with its size during 1950-2000. The TNPP of Tundra and alpine steppe was projected to change from 250.7 to114.0 Tg C. This is a huge reduction of 54.5% due to global climate change. Likewise, the areas of the Cold desert and Temperate humid grassland were projected to decrease by 40.9 and 0.5 m ha, respectively, while the TNPPs of the Cold desert and Temperate humid grassland were projected to decrease by 1.4 Tg C and 14.1 Tg C, respectively. The Warm desert and Savanna super-class groups were projected to increase the most. The area of Warm desert, Steppe, Semi-desert, and Savanna were projected to increase considerably, by nearly 138.2 m ha (Tables 4 and 10).

Changes in class
The changing climate is likely to affect the growth of plants, alter the species composition of plant communities, and ultimately change plant community structure [87]. It is anticipated that the classes will shift [3,5]

Trends in super-classes
The A2a scenario projects changes in China's temperature and rainfall patterns, making wet areas wetter and dry areas drier (Figure 17), resulting in class succession from an original class to a new class if one just considers changes in climate conditions. The model outputs suggest that the conversion of classes from the recent past to the A2a scenario would result in changes in distribution patterns of the area and TNPP among biomes. Table 11 showed the projected results of the changes in areas of biome conversions between the recent past and the future climatic A2a scenario. For example, in this long time sequence, an area of 58.0 million hectares of Tundra and alpine steppe would be converted to Forest (Legend 18 in Figure 18a and Table  11), which would take place mainly in the Tibet Autonomous Region (accounting for 40.1% of the change) and Qinghai Province (35.9% of change), while another 28.9 million hectares would change to Temperate humid grassland (Legend 15 in Figure 18a and Table 11), and this change would mainly take place in the Tibet Autonomous Region (78.1%). Likewise, an area of 56.9 million hectares of Cold desert under recent climate conditions would be converted into Warm desert under the A2a scenario (Legend 26 in Figure 18a and Table 11), mainly taking place in the Xinjiang Uygur Autonomous Region (99.1%). An area of 43.7 million hectares of Temperate humid grassland would be converted into Steppe (Legend 54 in Figure 18a and  Figure 18a and (Legend 43 in Figure 18a and Table 11), mainly taking place at the Inner Mongolia Autonomous Region (44.8%).   Note: The symbols are defined in the same way as in Table 8.  Figure 18b and Tables 4, 10, 11). There is a clear trend for a decrease in the area of the Grassland category by 11.1 million hectares, whereas areas of the Forest category would increase by 11.3 million hectares (Tables 4, 10 and 11). The relative area of the Forest category would increase by 2.7% over this period , whereas the relative area of the Grassland category would decrease by 2.1% over the same period.

Change in TNPP
The Classification Indices-based Model shows NPP either neutral or increasing for most Chinese grid cells during the periods from the recent past through the A2a scenario ( Figure  19). At the class level, the largest increases would be in  Table 12. Comparison of the total net primary productivity (10 11  At the category level, the TNPP of Grassland category would increase by 14.5% to 104.7 Tg C over this period , whereas the TNPP of Forest category would increase by 10.2% to 228.0 Tg C (Tables 4, 10 and 12).

Grassland carbon budgets
The Classification Indices-based Model projects China's TNPP at 3.28 Pg C (Table 10) under the future climatic A2a scenario and is close to the projections of 3.33 and 3.34 Pg C by the Vegetation Interaction Model (AVIM) [91] and the CEVSA (Carbon Exchange between Vegetation, Soil, and Atmosphere) model [92], both derived by using long-term annual average data . The results could potentially favor China's position in supporting its effort to reduce global warming gas as outlined in the Kyoto Protocol. The model outputs suggest that the potential TNPP of grasslands in China is 0.83 Pg C in the A2a scenario (Table 10).
Climate change is expected to affect the TNPP of grassland. Understanding climate and NPP feedback cycles in grasslands requires knowledge of the changes in spatial patterns on the NPP of grasslands. The dynamic change characteristics of potential biomes (Table 12 and Figure  18) highlights the risk that China's grassland faces with a climate that is 'already hot, dry and variable' [85] under global warming. The Classification Indices-based Model estimated an increase of 104.71Tg C in TNPP for potential grasslands from recent past to the future A2a scenario. The increase in TNPP for grasslands at northern latitudes was mainly driven by increased temperature and, in lower latitudes, by changes in precipitation ( Figure 17 and Figure 19). The increase in TNPP for grasslands will bring an increase of 375.5 Tg of annual sequestration of CO 2 and 276.4 Tg of annual contribution to the atmospheric O 2 cycle respectively (1.63g of CO 2 are absorbed and 1.2 g of O 2 are released to generate 1g of dry matter according to photosynthesis stoichiometry) [89].

'Safe' livestock carrying capacity
The goal of sustainable livestock husbandry is to balance the consumption of forage by livestock with the annual production of the grassland, so that future grassland production is not diminished. The concept of a 'safe' livestock carrying capacity was developed to estimate the capacity of the grassland resource to sustainably carry livestock (and other herbivores) in the long term (>30 years) [83]. Predictions of NPP give a good estimate of the annual production of the grasslands, and therefore NPP can be used to balance the food demands of different kinds of livestock, allowing a calculation of the maximum carrying capacity of livestock that can be supported on the grassland. This maximum carrying capacity of the grassland should be used to develop appropriate policies of grazing control to preserve sustainable production on the grassland [16]. The implication of climate change projections for 'safe' livestock carrying capacity remains an open question. Global warming causes changes in temperature and precipitation patterns, resulting in fluctuations in the NPP of grasslands. Several researchers have predicted that increases in temperature will cause a shift in the ratio of C 4 to C 3 grasses, with C 4 increasing [17,93]. This may decrease carrying capacity, since C 4 grasses are generally considered to be of lower nutritive value to grazing animals than C 3 species [94]. Moreover, increasing temperature and declining precipitation are also considered to decrease dietary crude protein, protein availability, and digestible organic matter [93]. In situations where climates become warmer and drier, grazing livestock are likely to encounter a protein-limited diet. The future declines in forage quality would also produce greater methane production from ruminant livestock [93]. Thus, it is important to assess climate change projections in terms of their implications for future 'safe' livestock carrying capacity and the risk of resource degradation [83]. However, the questions of what the 'safe' livestock carrying capacity is and of how best to integrate a wide range of factors, such as grassland classes, climatic variability and animal nutrition, are unresolved. So further research and development is needed to identify the regional trends for the 'safe' livestock carrying capacity to maintain sustainable resource condition and reduce the risk of resource degradation. This important task remains a challenge for all grassland scientists and practitioners.