CLM4.5 is the latest version of the CLM model family and is the land component of version 1.2 of the Community Earth System Model (Oleson et al., 2013). It succeeds CLM4 with updates to photosynthesis, soil biogeochemistry, fire dynamics, cold region hydrology, the lake model, and the biogenic volatile organic compounds model (Li et al., 2014). The resulting model is fully prognostic with respect to all water, energy, carbon, and nitrogen variables in the terrestrial ecosystem (Oleson et al., 2013). A detailed description of its biogeophysical and biogeochemical parameterizations and numerical implementation is given in Oleson et al. (2013).

CLM4 and CLM4.5 can both be run with or without an active CN. When CN is active, i.e., nitrogen limitation is prognostic, potential GPP is calculated from the leaf photosynthetic rate without nitrogen constraint. The nitrogen required to achieve this potential GPP is then diagnosed, and the actual GPP is decreased if there is insufficient nitrogen to maintain the potential biomass increment. The nitrogen inducing the reduction of GPP is applied after the photosynthesis calculation and is independent of V_cmax, but does implicitly decrease V_cmax(Bonan et al., 2011, 2012). When CN is inactive, i.e., nitrogen limitation is prescribed, the potential is directly reduced by multiplying a prescribed nitrogen limitation factor f(N):

, (1)

where f(N) is scaled between zero and one so that GPP is similarly decreased for nitrogen availability.

In CLM4.5, the original potential used in CLM4, specified for each plant functional type (PFT) in the absence of nitrogen limitation (Thornton and Zimmermann, 2007), is replaced with (Bonan et al., 2012); and to make this meaningful, PFT-specific values forf(N) in CLM4 (denoted as f₂(N)) for runs without CN, were all changed to 1 in CLM4.5 (denoted as f₁(N)). However, when CN is active, CLM4.5 still uses as the potential values for the calculation of GPP. Unfortunately, this may cause double nitrogen down- regulation. Therefore, to address this problem, we still use f₂(N) as the explicit nitrogen limitation factor in CLM4.5 and divide the values, which already reflect nitrogen via f₂(N) (like Eq. (2)), to obtain the potential values of that match with . Those values for , f₁(N), f₂(N), and are all listed in Table 1. The revision makes no difference to the calculation of GPP when CN is inactive, but does when CN is active.

. (2)

An accurate land cover map can significantly reduce the uncertainty of land surface modeling. Based on Dempster-Shafer evidence theory, Ran et al. (2012) developed a high-accuracy (1 km) multi-source integrated Chinese land cover dataset (MICLCover), which shows great improvement compared to other popular land cover maps. It was generated using the International Geosphere Biosphere Programme classification system by combining five sets of relevant land cover data: the Vegetation Atlas of China (1:1β 000β 000) (Hou, 2001); a land use map of China (1:100β 000) for the year 2000 (Liu et al., 2002); a glacier distribution map of China (1:100β 000) (Wu and Li, 2004); a swamp-wetland map of China (1:1β 000β 000) (Zhang, 2002); and the Moderate Resolution Imaging Spectroradiometer (MODIS) product for 2001 (MODIS2001) (Friedl et al., 2002). The MICLCover map (http://westdc. westgis.ac.cn/dat) is expressed by the percentages of glaciers, lakes, wetland, urban land, and 16 PFTs in each grid, consistent with the classification in CLM. Yu et al. (2014) and Wang et al. (2015) found that the MICLCover map has the potential to improve land surface modeling with CLM. Here, we upscale the MICLCover map from 1 km to 0.5° by means of an area-weighted average and use it as the land cover data in CLM4.5 (Wang et al., 2015).

Table 1

Table 1

Table 1 PFT (plant functional type)-specific photosynthetic parameters. is the realized Vcmax25 values obtained from Kattge et al. (2009); f1(N) is the prescribed nitrogen limitation factor used in CLM4.5, while f2(N) is from CLM4.0; and , which represents the potential values for Vcmax25, is given by . Plant functional type f1(N) f2(N) Needleleaf evergreen tree, temperate 62.5 1 0.72 86.8 Needleleaf evergreen tree, boreal 62.6 1 0.78 80.2 Needleleaf deciduous tree, boreal 39.1 1 0.79 49.5 Broadleaf evergreen tree, tropical 55.0 1 0.83 66.3 Broadleaf evergreen tree, temperate 61.5 1 0.71 86.8 Broadleaf deciduous tree, tropical 41.0 1 0.66 62.1 Broadleaf deciduous tree, temperate 57.7 1 0.64 90.1 Broadleaf deciduous tree, boreal 57.7 1 0.70 82.4 Broadleaf evergreen shrub, temperate 61.7 1 0.62 99.5 Broadleaf deciduous shrub, temperate 54.0 1 0.60 90.0 Broadleaf deciduous shrub, boreal 54.0 1 0.76 71.0 C3 grass, arctic 78.2 1 0.68 115.0 C3 grass 78.2 1 0.61 128.1 C4 grass 51.6 1 0.64 80.6 Crop 100.7 1 0.61 165.1

Table 1 PFT (plant functional type)-specific photosynthetic parameters. is the realized Vcmax25 values obtained from Kattge et al. (2009); f1(N) is the prescribed nitrogen limitation factor used in CLM4.5, while f2(N) is from CLM4.0; and , which represents the potential values for Vcmax25, is given by .

To run CLM in the offline mode, atmospheric forcing data are required. The standard forcing provided with the model is a half-degree 110-year (1901-2010) dataset (CRUNCEP; Viovy, 2011) that includes precipitation, incident solar radiation, temperature, pressure, winds, humidity, and incident longwave radiation. It is a combination of two existing datasets: version 3.2 of the Climate Research Unit Time Series (CRU TS3.2) 0.5° × 0.5° monthly data from 1901 to 2002, and the National Centers for Environmental Prediction (NCEP) reanalysis 2.5° × 2.5° six-hourly data from 1948 to 2010 (Oleson et al., 2013).

Three products are used for benchmarking our comparisons: the global GPP (monthly, 0.5° × 0.5° ), upscaled from FLUXNET observations using the machine learning technique, model tree ensembles (MTE) from 1982 to 2011 (Jung et al., 2009, 2011, denoted as MTE_GPP); the global MODIS NPP product (MODIS17A2, monthly, 0.5° × 0.5° , denoted as MODIS_NPP) from 2000 to 2013; and the MODIS LAI product (MODIS15A2, monthly, 0.5° × 0.5° , denoted as MODIS_LAI) from 2000 to 2013. The MODIS LAI and NPP data are obtained from NASA Earth Observations (http://neo.sci.gsfc.nasa.gov/).

To investigate the effect of the revised V_cmax scheme on terrestrial carbon cycle modeling, two experiments using CLM4.5-CN were conducted with 0.5° × 0.5° resolution using the 20-year (1991-2010) CRUNCEP datasets as atmospheric forcing data and MICLCover map as surface data: (1) a control run with the default V_cmax scheme (denoted as CTL); (2) a new run with the revised V_cmax scheme (denoted as NEW). Besides the difference in the V_cmax schemes of the two experiments, all the other model configurations and input data were exactly the same. Both experiments were spun up for 1500 years for establishing the carbon and nitrogen pools and fluxes (Hudiburg et al., 2013; Kendra et al., 2012), forced circularly by 20 years’ atmospheric forcing. Subsequently, two simulations from 2000 to 2010 are analyzed.

To evaluate the model’ s performance, GPP, NPP, and LAI, which are the dominant components of the terrestrial carbon cycle from the two runs, are analyzed (denoted as CTL_GPP, CTL_NPP, and CTL_LAI, and NEW_GPP, NEW_NPP, and NEW_LAI, respectively). For time consistency, the simulated annual means of GPP, NPP, and LAI from 2000 to 2010 are compared with observations over China for the same time period.

To quantify the performance of the two V_cmax schemes in comparison with observations, we computed the correlation coefficient (R) and root-mean-square error (RMSE), defined as follows:

, (3)

, (4)

where x is the model simulation either from CTL or NEW, obs is the corresponding observation, and are the mean value of x and obs, respectively.

Figure 1 shows the spatial distributions of mean GPP from model simulations (CTL_GPP and NEW_GPP) and observation (MTE_GPP), together with their differences. To examine regional differences in the performance of the V_cmax schemes, four sub-regions over China are defined according to mean annual precipitation, P_mean: arid (P_mean ≤ 200 mm); semi-arid (200 mm < P_mean < 400 mm); semi-humid (400 mm < P_mean < 800 mm); humid (P_mean > 800 mm) (Yu et al., 2014). The spatial patterns of the simulated GPP are very similar (Figs. 1a and 1b), being characterized by a southeast-northwest gradient, consistent with MTE_GPP (Fig. 1c), with the largest annual GPP in humid regions, followed by semi-humid regions, semi-arid regions, and arid regions. The GPP annual means of the two simulations in China are 8.17 and 7.85 Pg C yr^-1 respectively, which are much higher than the 7.05 Pg C yr^-1 of MTE_GPP, albeit with obvious reduction in NEW_GPP. Although both simulations overestimate GPP in humid areas (Figs. 1d and 1e), NEW_GPP shows large improvements in these areas with a mean reduction of -65.4 g C m^-2 yr^-1. Apart from in the arid regions dominated by C3 or C4 grass and bare areas (not shown), the values of GPP are reduced significantly at the 95% confidence level in Qinghai-Tibet and southwest areas, which are dominated by cropland and trees (Fig. 1f). Although the values of new V_cmax are higher than the original values, which may induce higher potential leaf photosynthetic rates, the reduced total soil water availability ( ) and nitrogen availability (f_N) of the NEW run (not shown) decrease the GPP simulation, achieved by multiplying V_cmax by and GPP byf_N, respectively. The scatterplots between the simulated GPP, NPP, LAI and observations and the MTE_GPP are shown in Fig. 3. The RMSE for the CTL_GPP for the whole of China is 360.8 g C m^-2 yr^-1, whereas the NEW_GPP reduces the RMSE by 13.7 g C m^-2 yr^-1, though the new V_cmax has little effect on the spatial correlation between the simulated GPP and MTE_GPP. Although the new V_cmax further underestimates mean annual GPP in semi-humid and semi-arid regions, there are improvements in the other areas, with lower RMSEs (Fig. 4b), more obvious in humid regions, where the RMSE is reduced from 622.5 to 583.2 g C m^-2 yr^-1.

Figure 1 also shows the spatial patterns of mean NPP. It can be seen that the CTL and NEW model simulations and observations (MODIS_NPP) show broadly similar patterns, with a northwest-southeast gradient over China (Figs. 1g-i). Compared with MODIS observations, CLM4.5 simulations clearly overestimate NPP in most areas of China, especially humid areas (Figs. 1j and 1k). The NPP means over China for CTL, NEW, and MODIS are 3.81, 3.78, and 2.71 Pg C yr^-1, respectively. The scatterplots between CLM4.5 simulations and MODIS observations show that the new V_cmax scheme reduces the RMSE from 260.9 (CTL) to 260.7 g C m^-2 yr^-1. The mean annual NPP of the NEW run shows improvements in all four sub-regions compared to MODIS_NPP, with an increase of 0.7 g C m^-2 yr^-1 in arid areas and decreases of 8.3, 8.7, and 4.1 g C m^-2 yr^-1 in semi-arid, semi-humid, and humid areas, respectively. Furthermore, the RMSEs of NPP for the four sub-regions from the NEW run are all lower than those from the CTL run, especially for semi-arid and semi-humid areas, which also reflect the success of the revised V_cmax scheme (Fig. 4d).

Figure 1

Figure Option ViewDownloadNew Window

Figure 1 Eleven-year (2000-10) mean gross primary production (GPP) from the (a) control run (CTL_GPP), (b) new Vcmax scheme (NEW_GPP), (c) observations (MTE_GPP), (d) CTL_GPP minus MTE_GPP, (e) NEW_GPP minus MTE_GPP, and (f) NEW_GPP minus CTL_GPP. Panels (g-l) are the same as GPP but for net primary production (NPP) from (g) CTL_NPP, (h) NEW_NPP, (i) MODIS_NPP, (j) CTL_NPP minus MODIS_NPP, (k) NEW_NPP minus MODIS_NPP, and (l) NEW_NPP minus CTL_NPP. The grid shadow regions are significant at the 95% confidence level by student’ s t test, and the isolines refer to the mean annual precipitation (mm yr-1).

Figure 2 shows the spatial distributions of mean annual LAI from MODIS and the two CLM4.5 simulations and their differences. It is found that, as with GPP and NPP, the simulated LAI and MODIS_LAI all show southeast-northwest gradient (Figs. 2a-c). Both LAI simula-tions have positive biases in almost all areas of China, especially humid areas (Figs. 2d and 2e). However, the revised V_cmax scheme causes reductions in LAI in semi-arid, semi-humid, and humid areas (Fig. 2f), resulting in lower RMSE (2.3) in China compared with those of CTL_LAI (RMSE = 2.6) (Figs. 3e and 3f). And just like NPP, the RMSEs of LAI for the four sub-regions from NEW are all notably decreased, especially for humid areas, where RMSE is decreased from 4.8 to 4.2 (Fig. 4f).

Figure 2

Figure Option ViewDownloadNew Window

Figure 2 Eleven-year (2000-10) mean leaf area index (LAI) from the (a) control run (CTL_LAI), (b) new Vcmax scheme (NEW_LAI), (c) observations (MODIS_LAI), (d) CTL_LAI minus MODIS_LAI, (e) NEW_LAI minus MODIS_LAI, and (f) NEW_LAI minus CTL_LAI. The grid shadow regions are significant at the 95% confidence level by student’ s t test, and the isolines refer to the mean annual precipitation (mm yr-1).

Figure 3

	Figure Option ViewDownloadNew Window
	Figure 3 Comparison of annual GPP, NPP, and LAI from CLM4.5 model simulations (CTL and NEW) on 0.5° collocated grid boxes, with the corresponding observational data. RMSE = root-mean-square error; R = correlation coefficient.

Figure 4

	Figure Option ViewDownloadNew Window
	Figure 4 Comparison results of the control run (CTL) and new scheme (NEW) for (a-b) GPP, (c-d) NPP, and (e-f) LAI compared with observations over the four sub-regions: arid, semi-arid, semi-humid, and humid.