HTML
--> --> -->In the past several decades, rapid climate change has resulted in considerable changes in terrestrial ecosystems. One widespread dynamic change is the so-called "greening" of arctic ecosystems——a feature considered to be majorly contributed by boreal shrubs (Fraser et al., 2011; Myers-Smith et al., 2011). Both observation- and model-based studies have shown an increase in shrub biomass, coverage and abundance in arctic ecosystems in recent decades (Myers-Smith et al., 2011). Another key region is the tropics, which is ecologically sensitive to climate variability and shows amplified responses compared to other regions (Seddon et al., 2016). Many tropical ecosystems are more sensitive than other regions to environmental perturbations and external disturbances, and are highly likely to cross a threshold to an alternative state (Holling, 1973; Scheffer et al., 2009).
These vegetation changes can in turn significantly affect the climate. Over high latitudes, higher shrub abundance warms the winter soil temperature by trapping more snow via their branches (Sturm et al., 2001). During early spring, taller shrubs that expand into arctic tundra ecosystems also tend to systematically warm the soil because of their lower albedo than snow (Bonfils et al., 2012). In boreal summer, however, the shading imposed by shrubs is known to reduce the soil temperature and consequently the thickness of the active layer (Blok et al., 2010). Many studies have also reported that shrub canopies can alter arctic nutrient cycling, biodiversity and ecosystem services (Myers-Smith et al., 2011). Conversely, in the tropics, the complete deforestation of Amazonia may result in warmer, drier conditions at the local scale and lead to extratropical changes in temperature and rainfall through teleconnections (Lawrence and Vandecar, 2015). Reduced plant cover in the Sahara, meanwhile, may cause a decline in precipitation because of the resultant increase in albedo (Charney et al., 1975), and the vegetation dynamics of West Africa have been shown to play a crucial role in regulating present-day climate, and probably future climate, via influences on precipitation, evapotranspiration and energy balance (Erfanian et al., 2016; Yu et al., 2016).
At present, Dynamic Global Vegetation Models(DGVMs) are the best available tool to represent vegetation dynamics at the global scale (Quillet et al., 2010). They can simulate transient and potential responses of vegetation to past and future climate change via parameterizing physical and biogeochemical processes of vegetation (Peng, 2000). With the help of DGVMs, it is possible for global climate models (GCMs) to include the bidirectional interactions between vegetation and climate (Quillet et al., 2010). Therefore, the coupling of DGVMs and GCMs is an important approach to assess the influences of climate change on vegetation dynamics and their feedbacks to climate change.
However, there are still many uncertainties in reproducing and forecasting vegetation dynamics with DGVMs because of the complicated physical and biogeochemical processes. In response to climate change, most DGVMs simulate an increase in woody coverage over high latitudes, while others suggest a gain in herbaceous vegetation or no changes (Falloon et al., 2012). In terms of the Amazon ecosystem, the predictions from different models also vary widely, which is dominated by the differences in large-scale forest dieback and forest resilience (Betts et al., 2004; Friedlingstein et al., 2006; Baker et al., 2008; Restrepo-Coupe et al., 2017). (Falloon et al., 2012) suggested that these responses of DGVMs to climate change are strongly associated with their simulation of present-day vegetation cover. Therefore, an important step to reduce these uncertainties in DGVMs is to systematically evaluate and understand their present-day performance, which is the main focus of the present study.
CoLM-DGVM is the default DGVM of the Common Land Model (CoLM). It was developed from the Lund-Potsdam-Jena DGVM (LPJ-DGVM; Sitch et al., 2003), and combined an early version of a temperate shrub sub-model of the DGVM developed at the Institute of Atmospheric Physics (IAP-DGVM; Zeng et al., 2008). In recent years, a boreal shrub sub-model, a process-based fire parameterization and a new establishment parameterization scheme have been further developed in IAP-DGVM. These developments have resulted in considerably improved reproductions of the present-day vegetation distribution and carbon cycle by IAP-DGVM (Zeng et al., 2014). At present, IAP-DGVM has been coupled to CoLM, and both are important components of the current version of the Chinese Academy of Sciences' Earth System Model (CAS-ESM). This work assesses the performance of IAP-DGVM, through comparison with that of CoLM-DGVM, within the framework of CAS-ESM, which is a necessary step to using CAS-ESM for investigating vegetation-climate interactions. In the next section, the model, experimental design and observational data are described. The results are presented and discussed in section 3, followed by conclusions in section 4.
2.1. Model description
IAP-DGVM is a DGVM developed at the Institute of Atmospheric Physics, Chinese Academy of Sciences. It originates from LPJ-DGVM (Sitch et al., 2003) and the Community Land Model's DGVM (CLM-DGVM; Levis et al., 2004). Plants in IAP-DGVM are classified into 14 plant functional types (PFTs) (Table 1), of which eight are trees, three are shrubs and three are grasses. These PFTs are defined by their physical, phylogenetic and phenological parameters, and are assigned bioclimatic limits to determine their establishment and survival. At present, crops are not simulated in IAP-DGVM. More details about the model can be found in (Zeng et al., 2014).In recent years, IAP-DGVM has undergone several major developments. These include the following: (1) A shrub sub-model was established (Zeng et al., 2008; Zeng, 2010). With this sub-model, IAP-DGVM can reproduce the global distribution of temperate and boreal shrubs realistically, and distinguish shrubs from grasses effectively (Zeng et al., 2008; Zeng, 2010). (2) A process-based fire parameterization of intermediate complexity was introduced (Li et al., 2012), which comprises fire occurrence, fire spread and fire impact. This fire parameterization significantly improves simulations of global fire, including burned area and fire carbon emissions (Li et al., 2013). The fire parameterization is now also adopted in the Community Earth System Model at the National Center for Atmospheric Research (NCAR) to investigate the influences of fire on carbon balance in terrestrial ecosystems, and on global land energy and water budgets (Li et al., 2014; Li and Lawrence, 2017; Li and Lawrence, 2017). (3) A new establishment parameterization scheme was developed (Song et al., 2016). This scheme significantly improves IAP-DGVM's simulation of vegetation density by introducing soil water as an impact factor. These improvements, together with other optimized modifications, contribute to a better performance of IAP-DGVM in reproducing the present-day vegetation distribution and carbon cycle.
Figure1. Zonal average fractional coverage (units: %) of (a) trees, (b) shrubs, (c) grasses and (d) bare groundin CTL (blue), IAP (red) and CLM4 surface data (OBS; black).The land surface model used in this study is CoLM. Starting from the code of the NCAR's Land Surface Model (Bonan, 1996), (Dai et al., 2003) developed the first version of CoLM, combining the codes of the Biosphere-Atmosphere Transfer Scheme (Dickinson et al., 1993) and the IAP's land model (Dai and Zeng, 1997). Since then, CoLM has been continually improved at Beijing Normal University in many aspects and has been adopted as the land component of the Beijing Normal University Earth System Model (Ji et al., 2014).
Figure2. Differences in fractional coverage (units: %) of (a) trees, (c) shrubs, (e) grasses and (g) bare ground between CTL and observations. (b, d, f, h) as in (a, c, e, g), respectively, but between IAP and observations.2
2.2. Experimental design
Two global simulations were conducted within the framework of CAS-ESM. One, which coupled CoLM and CoLM-DGVM, is the control (CTL) experiment, while the other, which coupled CoLM and IAP-DGVM, we refer to as IAP. Both simulations were spun up from bare ground for in excess of 1200 years. Then, a further 660 years were run to approach an equilibrium state through cycling the 33-year (1972-2004) atmospheric forcing data of (Qian et al., 2006), with a T85 resolution (128× 256 grid cells). The relative humidity and associated fire parameters needed for the IAP simulation were derived from (Li et al., 2012) and fixed in 2004. The atmospheric CO2 was set to 365 ppmv over all simulated years.2
2.3. Observational data
This paper focuses on the vegetation distribution, carbon cycle and leaf area index (LAI). The observed vegetation distribution and LAI came from CLM4 surface data, which were themselves derived from Moderate Resolution Imaging Spectroradiometer (MODIS) measurements (Lawrence and Chase, 2007). The benchmarks for gross primary productivity (GPP), net primary production (NPP) and fire carbon emissions were from (Beer et al., 2010), MODIS (Zhao and Running, 2010) and version 4 of the Global Fire Emissions Database (GFEDv4; Randerson et al., 2015), respectively. The average of the last five years (2000-2004) of the IAP simulation was compared with that of CTL and these benchmarks. Furthermore, to reduce the impacts of crops, the vegetation cover in each grid cell was weighted by a factor of (100%-FCcrop), where FCcrop is the fraction of crop coverage in the CLM4 surface dataset (Zeng et al., 2014).3.1. Vegetation distribution
In general, IAP simulated more realistic distributions for the four aggregated vegetation types (trees, shrubs, grasses and bare soil) than CTL. Over most latitudes, trees simulated by IAP were in better agreement with the observation, relative to CTL, which produced more trees (Fig. 1a). In CTL, trees were overestimated over northern high latitudes, southeastern South America and Africa, with magnitudes of 50% (Fig. 2a). Moreover, there was a band over central Eurasia where CTL's trees were underestimated by more than 50%. In contrast, IAP simulated fewer trees over the tropics and more over central Eurasia than CTL, which resulted in a reduction in IAP's biases (Fig. 2b). Further investigation indicated that the new establishment scheme contributed most to the reduced biases of tropical trees, especially over transition zones (Song et al., 2016).
Figure3. Global weighted average fractional coverage (%) of each PFT for CTL (blue), IAP (red) and observation (black). The abbreviations correspond to the information in Table 1.In terms of shrubs, both CTL and IAP underestimated them over arctic regions, such as northeastern Canada and the northern coastline of Eurasia (Figs. 1b, 2c and d). Further sensitivity experiments and analysis suggested that these underestimated shrubs were limited by the minimum threshold of growing degree days over 5°C (GDD5) set by the model, which is 350. Shrubs were unable to establish because the annual GDD5 was smaller than 350. Over northern high latitudes, shrubs could not grow and were severely underestimated in the CTL simulation, while IAP simulated a more similar shrub pattern than CTL with that observed (Fig. 1b). This improvement can be attributed to the boreal shrub sub-model of IAP-DGVM, which can distinguish shrubs from grasses effectively (Zeng et al., 2008; Zeng, 2010). However, IAP simulated more shrubs than CTL over northern middle latitudes, such as western North America and central Eurasia, which further increases the biases of CAS-ESM. Moreover, both CTL and IAP underestimated the shrub coverage in the Southern Hemisphere, such as in Australia.
IAP's grasses also agreed better with the observation than those of CTL (Fig. 1c). The grasses in CTL were largely overestimated over middle and high latitudes, such as northeastern Canada, central North America, middle and eastern Russia and central Eurasia, where the biases exceeded 50%. However, IAP reduced these deficiencies to around 10% over these regions (Fig. 2f). Over the tropics, both CTL and IAP underestimated grasses, although IAP's biases were a little smaller than those of CTL. The main underestimation was in southeastern South America and most regions of Africa.
Figure 1d shows that the bare soil simulated by CTL was underestimated over middle and high latitudes, and overestimated over the tropics, while IAP simulated more bare soil over almost all latitudes except southern middle latitudes. Over high latitudes, such as northeastern Canada, the underestimated bare soil of CTL mainly resulted from its overestimated grasses (Fig. 2g). IAP significantly reduced the fractional coverage of grasses over northeastern Canada (Fig. 2f); however, other vegetation, such as boreal shrubs, did not grow in this region (Fig. 2d). Consequently, the bare soil of IAP was overestimated over northern high latitudes (Fig. 2h). In the tropics, both CTL and IAP simulated more bare soil than observations (Fig. 1d). These overestimations were mainly in northeastern Africa, southern Africa and Australia for CTL, and most regions of Africa for IAP (Fig. 2h). The underestimated grasses were mostly responsible for these biases of tropical bare soil.
To investigate the contribution of each PFT to the improvements, Fig. 3 shows the global average fractional coverage for each PFT in the two simulations and the observation. The total coverage of IAP's trees (26.51%) was more consistent with that of the observation (28.51%) than that of CTL (35.60%). The smaller fractional tree coverage in IAP was mainly contributed to by a reduction in "broadleaf evergreen tropical tree" (BET; 3.17%), "broadleaf deciduous tropical tree" (BDT; 3.27%), "broadleaf deciduous temperate tree" (BDM; 2.10%) and "broadleaf deciduous boreal tree" (BDB; 2.10%). However, IAP's BDT was less than half that of the observation, which was the major contributor to the underestimation of IAP's total tree coverage. Further investigation indicated that the new establishment parameterization of IAP-DGVM resulted in this underestimated BDT, mainly over tropical semi-arid regions (Fig. S1 in electronic supplementary material). IAP also simulated 1.32% more "needleleaf evergreen boreal tree" (NEB) than CTL, which contributed to the band of increased tree coverage over central Eurasia apparent in Fig. 2b. In terms of shrubs, the increased "broadleaf deciduous boreal shrub" (BDBsh) coverage (5.44%) in IAP was the main contributor to its better agreement with the observation than CTL. For grasses, although the total fraction in CTL was closer to that observed than IAP's grass fraction, the "C3 arctic grass" (C3Ar) coverage in CTL was 6.33% larger than the observation, which corresponds to the severely overestimated grasses over high latitudes shown in Fig. 1c. The composition of IAP's grasses was generally in good agreement with the observation. However, the "C4 grass" (C4) coverage in IAP was 4.64% less than the observation, resulting in the underestimation of total grasses in IAP. The bare soil of CTL was close to that observed, but IAP simulated 7.67% more bare soil than the observation. The overestimated bare soil in IAP mainly resulted from its underestimated shrubs in arctic regions and underestimated grasses in the tropics (Fig. 2).
Figure4. Global distribution of the dominant vegetation type obtained from (a) CTL, (b) IAP and (c) observation. The abbreviations correspond to the information in Table 1.Figure 4 shows the global distribution of the dominant vegetation type, which is the PFT with the highest fractional coverage. Clearly, the dominant vegetation simulated by IAP was more consistent with that obtained from the observation than that of CTL. In CTL, C3Ar dominated over most northern high-latitude regions. However, IAP showed the northern high latitudes to be dominated by bare soil, BDBsh and NEB, which was a similar situation to that observed. In the tropics, CTL overestimated regions dominated by BET, as compared to observation. For example, BET was the dominant vegetation in CTL over southeastern South America and West Africa, which, according to observation, are actually dominated by C4. With respect to CTL, IAP simulated fewer regions dominated by BET, which agreed well with observation. Nonetheless, IAP's C4 was also not the dominant vegetation in southeastern South America and West Africa because of the underestimated C4 (Fig. 3).
Figure5. Zonal average (a) GPP, (b) NPP and (c) fire carbon emissions (FireC) in CTL (blue), IAP (red) and the benchmarks (OBS; black). All units are gC m-2 yr-1.
Figure6. Differences between CTL and the benchmarks (CTL minus benchmarks) in (a) GPP, (c) NPP and (e) fire carbon emissions (FireC). (b, d, f) As in (a, c, e), respectively, but between IAP and the benchmarks. All units are gC m-2 yr-1.2
3.2. Carbon fluxes
Compared to CTL, IAP simulated an overall more reasonable distribution of key carbon fluxes. Over most latitudes, the GPP in IAP was similar to that of (Beer et al., 2010), while CTL's GPP was overestimated in middle and high latitudes (Fig. 5a). The overestimated GPP in CTL stretched over the whole of central and eastern North America, Europe, South Asia and southeastern South America, while underestimated GPP dominated over the Amazon and Africa (Fig. 6a). Relative to CTL, IAP's GPP was closer to observation, especially over the Amazon (Fig. 6b). Both CTL and IAP underestimated GPP over most regions of Africa.In terms of NPP, IAP showed a better agreement with MODIS than CTL over northern high latitudes (Fig. 5b). Relative to CTL, IAP simulated lower NPP over northeastern Canada and central and eastern Russia, where CTL simulated ～ 200 gC m-2 yr-1 more NPP than the observation (Fig. 6c). Both CTL and IAP underestimated NPP over middle latitudes, mainly over arid and semi-arid regions. In the tropics, the simulated NPP in CTL was higher than the observation, mainly because of the overestimated NPP over the Amazon, West Africa and the Maritime Continent. On the contrary, IAP's NPP was consistent with observations over the Amazon, Africa and the Maritime Continent. Both CTL and IAP underestimated the NPP over most parts of Africa.
Fire carbon emissions were overestimated by IAP over middle latitudes, especially in central and western North America, northeastern China, southern South America and Australia (Fig. 6f). However, the fire carbon emissions simulated by IAP were much more consistent with observation in the tropics, where CTL showed a severe underestimation (Fig. 5c). Fire carbon emissions in CTL were 10 gC\;m-2\;yr-1 higher than the observation in eastern North America, Europe, the Amazon, Southeast Asia and the Maritime Continent, and 50 gC m-2 yr-1 lower in central South America and most regions of Africa (Fig. 6e). IAP reduced these errors to different degrees by increasing or decreasing fire carbon emissions in these regions, respectively. Broadly, IAP captured a better spatial distribution of fire carbon emissions than CTL.
Overall, within the framework of CAS-ESM, IAP simulated carbon fluxes closer to observations than CTL, as summarized in Fig. 7. The GPP simulated by IAP was 150.5 PgC\;yr-1, which is closer to the 123 8 PgC yr-1 reported by (Beer et al., 2010) than CTL's GPP, and comparable to the range from 101 to 150 PgC yr-1 published elsewhere (Farquhar et al., 1993; Ciais et al., 1997). Meanwhile, IAP also overestimated autotrophic respiration, which was almost the same as its counterpart in CTL. Consequently, IAP's NPP, 59.1 PgC yr-1, compared better with the expected value of 60 PgC yr-1 (Castillo et al., 2012) than the result of CTL (75.31 PgC yr-1). Moreover, heterotrophic respiration in IAP was 56.20 PgC yr-1, which is closer to the 55.4 PgC yr-1 from (IPCC, 2013) than that of CTL (74.08 PgC yr-1). Thus, the net ecosystem production in IAP was more reasonable than that in CTL, in comparison to the baseline from (IPCC, 2013). The fire carbon emissions in IAP were slightly higher than those of GFEDv4, because of the overestimated fire carbon emissions in the midlatitudes. Consequently, the net biome production (NBP) of IAP-DGVM was -0.2 PgC yr-1, which is outside the range of 2.63 1.22 PgC yr-1 reported by other process-based terrestrial ecosystem models driven by rising CO2 and by changes in climate (IPCC, 2013). This near-zero value of NBP could be acceptable, however, because the results were based on the equilibrium state, which was cyclically forced by atmospheric datasets and a constant CO2 value (Castillo et al., 2012).
2
3.3. LAI
Generally, both CTL and IAP overestimated LAI, although the bias in IAP was smaller than that in CTL (Fig. 8). The simulated annual mean LAI in CTL and IAP was 1.0 m2 m-2 more than the observation over most of the northern middle and high latitudes, such as central and eastern North America, Europe, central Eurasia and southeastern China. In the tropics, CTL's bias in LAI exceeded 5.0 m2 m-2, while IAP's was ～ 3.0 m2 m-2. In terms of seasonal variability, both CTL and IAP were consistent with observations, the largest being during June-August and the smallest during December-February. However, the simulated LAI amplitudes in CTL and IAP were around twice as large as those observed for each month. Although the LAI values in IAP were closer to observation compared to those of CTL, the improvements were not remarkable. Therefore, it is necessary to further investigate the causes of these systematically overestimated LAI values.
Figure7. Global means of carbon fluxes in CTL and IAP, as well as that of the benchmarks. Units: PgC yr$-1$.
Figure8. Differences in annual mean LAI between (a) CTL and observations, and (b) IAP and observations. (c) Globally averaged LAI in CTL (blue), IAP (red) and observations (black) for each month. All units are m2 m-2.The total tree coverage of IAP-DGVM was found to be in good agreement with observations, because of the reduced tropical trees and increased NEB relative to CoLM-DGVM. The shrub coverage in IAP-DGVM also showed a similar distribution to that observed, which resulted from the significantly increased fractional coverage of BDBsh, with replacement of C3Ar. Meanwhile, the reduced C3Ar was the major contributor to the better consistency between the grass coverage in IAP-DGVM and that observed than between CoLM-DGVM and that observed. Consequently, the global distribution of the dominant vegetation type simulated by IAP-DGVM was similar to that observed, especially over northern high latitudes. Moreover, the biases of IAP-DGVM in terms of GPP and NPP were smaller because of improvements in the GPP over middle and high latitudes, as well as in the tropics. The tropical patterns of fire carbon emissions in IAP-DGVM were much more consistent than CTL with observations. These better performances of IAP-DGVM in simulating the global vegetation distribution and carbon fluxes provide a foundation to use CAS-ESM to study vegetation-climate interactions.
Despite the above positive results, deficiencies in IAP-DGVM were also found. Specifically, BDT was severely underestimated because of the new establishment parameterization; BDBsh could not grow in northeastern Canada because the simulated GDD5 did not exceed the threshold; and the C4 coverage was much smaller than the observation. Therefore, improved parameterization is necessary to simulate a more realistic distribution of vegetation types. Furthermore, both simulations overestimated GPP and autotrophic respiration, which is likely associated to the parameterization in CoLM being insufficiently sensitive to the DGVM. Additionally, the fire emissions simulated by the model in the high and middle latitudes were high. An understanding of the underlying mechanisms of these biases is needed to further improve the model, which will be the subject of future research.
The primary purpose of this study was to evaluate the performance of IAP-DGVM within the framework of CAS-ESM, which makes the evaluations more model specific. However, this work also exerts the following influences on other studies. This study reported the improvements and deficiencies of IAP-DGVM in CAS-ESM, which is valuable information for the application of CAS-ESM, as well as a sample for other model communities in terms of a comparative benchmark. The selection of global observations of carbon fluxes in this study was limited by the spatial and temporal scale of existing datasets, which is a pivotal message for observational scientists to observe carbon fluxes with large spatial scale and continuous time. Overall, we hope that the advantages and disadvantages of the simulations reported in this paper will prove valuable for scientists seeking to investigate climate change.
Electronic supplementary material:Electronic Supplementary material is available in the online version of this article at
