The main purpose of LS3MIP is to provide a comprehensive assessment of land surface processes and associated feedbacks on climate variability and climate change and to diagnose systematic biases in the land component of current CSMs (van den Hurk et al., 2016). Two sets of experiments are designed for LS3MIP: the first addresses systematic biases of land surfaces in the offline mode and the second addresses land feedbacks attributed to snow cover and soil moisture in an integrated framework. LS3MIP has proposed several sets of atmospheric forcing data to drive LSMs in offline mode. The results of historical experiments with BCC_AVIM1.0 and BCC_AVIM2.0 driven by the Princeton global forcing dataset (Sheffield et al., 2006) are analyzed in this section under the framework of ILAMB (Mu et al., 2016), which focuses on land surface energy budgets and terrestrial carbon cycles at the seasonal timescale, especially the summer and winter seasons.
According to the CERES data (Kato et al., 2013), the seasonal variation of net surface radiation (NSR; including both shortwave and longwave) strongly depends on the seasonal evolution of solar radiation, which peaks in summer and reaches its minimum in winter in both hemispheres. The tropics sees relatively strong year-round net radiation (Figs. 7a1, b1). Taking the summertime NH as an example (Fig. 7a1), the magnitude of NSR is above 90 W m−2 almost everywhere except in the Sahara desert in northern Africa, the Arabian Peninsula, and the Taklimakan desert in northwestern China. Most parts of mid- and high-latitude areas in the NH see more than 120 W m−2 of NSR; the southeastern part of the US even experiences more than 150 W m−2 of NSR in JJA. Generally, BCC_AVIM2.0 has reasonably captured the geographi-cal distribution and seasonal evolution of NSR (figure omitted), although the simulation biases are obvious in both summer and winter, especially in the SH (Figs. 7a2, b2). During the boreal JJA, negative biases of NSR in southern US and in the Sahel region are mainly associated with negative net longwave radiation biases (Figs. 7a2, a4). Patchy positive biases in the northern Eurasian continent, the Sahara desert of North Africa, and the Arabian Peninsula are associated with overestimation of net solar radiation in the above areas (Figs. 7a2, a3). The aforementioned underestimation (overestimation) of net shortwave radiation results from larger (smaller) SA simulation in relevant regions (figure omitted). In contrast to the above shortwave radiation-dominant cases, overestimation of NSR in the northeast part of the Eurasian continent in DJF is due to the positive net longwave radiation bias in the BCC_AVIM2.0 simulation in that area (Figs. 7b2, b4). Underestimations of DJF net shortwave radiation in BCC_AVIM2.0 simulations in the SH are responsible for the negative NSR biases, and negative biases of net longwave radiation in the southern American and southern African continents may have also made some contributions (Figs. 7b2–b4). The aforementioned systematic biases indicate that there is much room to improve BCC_AVIM2.0 to reasonably simulate the surface energy budget.
Figure 7. Distributions of multi-year (2000–09) seasonal mean net surface radiation (NSR; W m−2) in (a1–a4) boreal summer and (b1–b4) boreal winter. (a1, b1) CERES (net radiation); (a2, b2) BCC_AVIM2.0 minus CERES (net radiation); (a3, b3) BCC_AVIM2.0 minus CERES (shortwave); (a4, b4) BCC_AVIM2.0 minus CERES (longwave).
Land surface evaporation includes three parts: soil surface evaporation, evaporation from canopy intercepted precipitation, and vegetation transpiration through leaf pores (Lawrence et al., 2007). In addition to wind speed, the availability of net surface energy, soil water, and canopy-intercepted precipitation are among those factors that influence land evaporation. The LHFX magnitudes in the tropical areas remain more than 80 W m−2 year-round, and the NH experiences maximum LHFX in JJA, whereas LHFX peaks in DJF in the SH (Figs. 8a1, b1). There is an obvious northeastward gradient of LHFX in the Eurasian continent in JJA and DJF. In JJA, the magnitude of LHFX decreases from more than 80 W m−2 in western Europe to approximately 60–80 W m−2 in the central part and less than 60 W m−2 in the northeastern part of the Eurasian continent. LHFX can be more than 80 W m−2 in eastern Asia and as high as 100 W m−2 in southern Asia (Fig. 8a1). In DJF, most parts of the NH to the north of 50°N experience surface freezing, and the LHFX values in these regions are thus below 5 W m−2. In contrast, LHFX in the subtropical areas in South America and the African continent reach peaks as high as 100 W m−2 (Fig. 8b1). BCC_AVIM2.0 has captured the overall geographical distribution and seasonal evolution of LHFX, except for some magnitude difference from the GBAF observations (Figs. 8a2, b2). The systematic biases of the BCC_AVIM2.0 simulation in JJA are obvious. There are approximately 20-W m−2 underestimations of LHFX in eastern US, tropical South America, tropical Africa, and southeastern China. In contrast, approximately 20-W m−2 overestimations of LHFX exist in semiarid regions in the NH such as northern Canada, northern Sahel, the Tibetan Plateau, and the Mongolian plateau (Fig. 8a3). In DJF, the LHFX underestimations in tropical South America, tropical Africa, and South Asia are approximately 10–20 W m−2 (Fig. 8b3). The aforementioned underestimations in LHFX simulation in both JJA and DJF can be attributed to the negative biases in NSR shown in Figs. 7a2, b2, which indicates the dominance of available energy on land surface evaporation.
The surface SHFX indicates the land surface thermal status and directly affects the overlying atmospheric circulation. Generally, lower surface pressure and near surface cyclonic circulation usually occurs over warmer surfaces with relatively high SHFX, and vice versa. The seasonal variation of SHFX is remarkable in high latitudes to the north of 50°N. Taking the latitude belt of approximately 60°N as an example, SHFX reaches more than 40 W m−2 in JJA and decreases to less than 10 W m−2 and even below zero in DJF, which means that the land surface in high-latitude NH is a heat sink in boreal winter. This heat sink in DJF can extend southward to 50°N in North America and the Eurasian continent to the west of 90°E (Figs. 9a1, b1). BCC_AVIM2.0 can simulate the overall geographical distribution of SHFX, except for the more southward expansion (approximately 40°N) of low SHFX coverage in North America and the northern Eurasian continent to the west of 110°E during DJF (Fig. 9b2). In JJA, there are approximately 20-W m−2 overestimations of SHFX in the BCC_AVIM2.0 simulation in dry lands, such as the central part of North America, the Brazilian Plateau, the Arabian Peninsula, and the midlatitude belt from west Asia to Mongolian Plateau (Fig. 9a3). The positive biases of SHFX in the central part of North America and the Brazilian Plateau coincide with the underestimation of LHFX in these regions, which is a common deficiency in the simulation of arid and semi-arid land surface processes (Wang et al., 2017; Zhang et al., 2017). In DJF, there are approximately 10–20-W m−2 positive biases to the north of 60°N and negative biases in the midlatitudes of the NH between 30°N and 60°N (Fig. 9b3). The overestimation of SHFX by BCC_AVIM2.0 in the northeastern part of the Eurasian continent in DJF is probably due to the underestimation of SCF there in winter (figure omitted).
Figure 9. As in Fig. 8, but for SHFX (W m−2).
The simulations of BCC_AVIM1.0 and BCC_AVIM2.0 are compared in this section to check the overall impacts of the updates in parameterizations described in Section 3. The climatological annual cycles of regional averaged solar and surface fluxes (NSR, SHFX, and LHFX) and LAI are examined and analyzed, over the main body of the Eurasian continent (40°–70°N, 0°–140°E) and the South African continent (0°–35°S, 10°–40°E). These two domains are chosen as examples to represent the NH and SH, respectively.
Figure 10 shows the annual cycle for the Eurasian continent. NSR peaks in June in CERES observations, and both BCC_AVIM1.0 and BCC_AVIM2.0 simulations captured the seasonal cycle, except that the simulated NSRs are less than the observations by approximately 5–15 W m−2 from February to November. BCC_AVIM2.0 values are larger than BCC_AVIM1.0 and closer to CERES from June to October but BCC_AVIM1.0 performs better from February to May, although the difference between the two simulations is within 5 W m−2 (Fig. 10a). The underestimation of NSR in both simulations is due to more reflected shortwave radiation and more emitted longwave radiation than those in the observations (figure omitted); the contribution of longwave radiation is larger than that of shortwave radiation from March to October, especially in boreal spring. SHFX in BCC_AVIM1.0 is approximately 5 W m−2 smaller than GBAF observations from January to April, and SHFX in BCC_AVIM2.0 is smaller than GBAF by approximately 5–10 W m−2 from January to May. SHFXs in BCC_AVIM1.0 and BCC_AVIM2.0 are close to each other from June to December; they are nearly 7 W m−2 larger than GBAF from June to August but approximately 5 W m−2 smaller than GBAF from October to December (Fig. 10b). LHFX in BCC_AVIM2.0 is approximately 5 W m−2 larger than BCC_AVIM1.0 and is closer to GBAF from May to August (Fig. 10c). Further investigation indicates that this improvement in LHFX in BCC_AVIM2.0 is due to more transpiration through vegetation (figure omitted). LAI is larger in BCC_AVIM1.0 than in the AVHRR data by approximately 1 all year round, and the bias of LAI in BCC_AVIM1.0 can be as large as 2 in August when LAI peaks in BCC_AVIM1.0 (Fig. 5b). BCC_AVIM2.0 is better than BCC_AVIM1.0 in both the magnitude and phase of the seasonal evolution of LAI (Fig. 10d). In short, the seasonal cycle is better in BCC_AVIM2.0 due to the better representation of vegetation phenology.
Figure 10. Annual cycle of regional averaged (a) NSR, (b) SHFX, (c) LHFX, and (d) LAI over the main part of the Eurasian continent (40°–70°N, 0–140°E). Unit is W m−2 for energy fluxes in (a–c).
The situation in the South African continent is shown in Fig. 11. NSR reaches a maximum of approximately 165 W m−2 in January and a minimum of approximately 75 W m−2 in July in CERES. Both BCC_AVIM1.0 and BCC_AVIM2.0 capture this seasonal evolution with year-round negative biases in both simulations. BCC_AVIM1.0 underestimates NSR by approximately 40 W m−2 in January and approximately 10 W m−2 in July. BCC_AVIM2.0 reduces this negative bias by approximately 5 W m−2 from February to September. The underestimation of NSR for the South African continent is due to more reflected shortwave radiation and more outgoing longwave radiation from the surface in both simulations (figure omitted). SHFX reaches a minimum (about 40 W m−2) in June and a maximum (about 67 W m−2) in October according to GBAF. BCC_AVIM1.0 underestimates SHFX by approximately 7 W m−2 from October to the following April, whereas SHFX in BCC_AVIM2.0 is larger than in BCC_AVIM1.0 year-round and closer to GBAF from October to the following April (Fig. 11b). However, this alleviation of negative SHFX bias in BCC_AVIM1.0 is at the cost of underestimation of LHFX by BCC_AVIM2.0 from January to April (Fig. 11c). The slight improvement in BCC_AVIM2.0 simulation of LHFX from May to October is associated with the better simulation of NSR during that period (Fig. 11a). LAI from AVHRR reaches a peak of approximately 2 in March. BCC_AVIM1.0 overestimates LAI by approximately 1 all year round, with a maximum near 5 in February; whereas BCC_AVIM2.0 captures the seasonal cycle of LAI evolution quite well, except for positive biases of approximately 0.5–1.0 from January to August (Fig. 11d). The maximum of LAI in the BCC_AVIM2.0 simulation from February to April is coincident with the LAI peak time over the South African continent shown in Fig. 5c. This indicates the advantage of the updated phenology scheme in BCC_AVIM2.0.
Figure 11. As in Fig. 10, but over the South African continent (0–35°S, 10°–40°E). Unit is W m−2 for energy fluxes in (a–c).
Statistics of the annual global land average of the aforementioned variables displayed in Table 1 also indicate improvements of BCC_AVIM2.0 simulations after the updates of the parameterizations. The global mean bias of NSR is reduced from −12.0 to −11.7 W m−2, and the RMSE drops from 20.6 to 19.0 W m−2. The bias of LHFX is reduced from 2.3 to −0.1 W m−2 and the RMSE is reduced from 15.4 to 14.3 W m−2. One exception is that the global mean SHFX bias is increased from 2.5 to 5.1 W m−2, whereas the RMSE is reduced from 18.4 to 17.0 W m−2. The bias of LAI is reduced from 0.89 to 0.75, and the RMSE is reduced from 1.46 to 1.27.
BCC_AVIM1.0 BCC_AVIM2.0 NSR −12.0 (20.6) −11.7 (19.0) LHFX 2.3 (15.4) −0.1 (14.3) SHFX 2.5 (18.4) 5.1 (17.0) LAI 0.89 (1.46) 0.75 (1.27)
Table 1. Bias and RMSE values (in brackets) of annual global land averaged simulations of BCC_AVIM1.0 and BCC_AVIM2.0. The observations for NSR are from CERES (W m−2), LHFX and SHFX data are from GBAF (W m−2), and LAI is from AVHRR (unitless)
The overall performance of BCC_AVIM1.0 and BCC_AVIM2.0 in simulating the global annual climatology of surface energy budgets, LAI, and variables associated with terrestrial carbon cycle is displayed in the Taylor diagram in Fig. 12. Both models perform well in simulating 9 variables with higher than 0.6 spatial correlations with the relevant observations [except for net ecosystem exchange (NEE) in the BCC_AVIM1.0 simulation], and 13 out of 18 standardized deviations fall approximately into the 0.5–1.5 range compared with the reference variability. Concerning spatial correlation, BCC_AVIM2.0 performs better than BCC_AVIM1.0 in simulating NEE, LAI, net surface radiation (NSR), SA, and SHFX, whereas BCC_AVIM1.0 performs slightly better in simulating GPP and respiration of ecosystem (RECO). It is noted that the two red circles representing the simulation performances of RECO and LAI in BCC_AVIM2.0 are very close to each other in Fig. 12. With respect to standardized deviation, which indicates the spatial variability of each variable, BCC_AVIM2.0 performs better than BCC_AVIM1.0 and closer to the thick reference circular arc in the simulation of GPP, RECO, tropical biomass, NSR, and SA.
Figure 12. Taylor diagram for the global annual climatology of gross primary productivity (gpp), respiration of ecosystem (reco), net ecosystem exchange of CO2 (nee), biomass, leaf area index (lai), land net surface radiation (nsr), surface albedo (sa), surface sensible heat flux (shfx), and surface latent heat flux (lhfx). The upper case suffix after each variable indicates the source of observation. The radial coordinate shows the standard deviation of the spatial pattern, normalized by the observed standard deviation. The azimuthal variable shows the correlation of the modeled spatial pattern with the observed spatial pattern. Analysis is for the entire globe (except that biomass is for tropi-cal areas). The BCC_AVIM2.0 and BCC_AVIM1.0 simulations are averaged over the same period as that for the relevant reference dataset. Crosses are for BCC_AVIM1.0 and circles for BCC_AVIM2.0.