The land surface temperature is a key feedback in the boundary layer of an atmosphere model. Figure 2 shows the performance of models in simulating the climatological annual mean land surface temperature. Spatial patterns of the land surface temperature obtained from the NCEP-2 dataset as well as ECHAM5 and ECHAM5-CoLM simulations are similar (Figs. 2a–c). Both ECHAM5-CoLM and ECHAM5 simulations show biases in many regions (Figs. 2d, e) and some of the bias patterns are similar, for example, a warm bias (about 2–6°C) appears in the mid- to low-latitude regions of both the Northern and Southern Hemispheres.
Figure 2. Climatological annual mean land surface temperature (°C). (a) Reanalysis dataset (NCEP-2), (b) ECHAM5, (c) ECHAM5-CoLM, (d) ECHAM5 minus NCEP-2, and (e) ECHAM5-CoLM minus NCEP-2.
By contrast, the significant cold bias (about –2 to –6°C) found in ECHAM5 over the high latitudes of Eurasia does not appear in the ECHAM5-CoLM simulation, which implies that coupling to CoLM improves the simulation skill of ECHAM5 in the boreal region. Table 1 shows the land surface temperature RMSE between the model and NCEP-2 reanalysis data calculated over the whole globe and six regions. The RMSE of ECHAM5-CoLM is significantly smaller than that of ECHAM5, both globally and regionally, except over North America. ECHAM5-CoLM therefore significantly improves the simulation of land surface temperature compared with ECHAM5.
Region ECHAM5 ECHAM5-CoLM Globe 1.88 1.70 Eurasia
2.05 1.83 North America
1.20 1.42 Africa and Austria
1.13 0.87 South America
0.84 0.80 Arctic
1.23 1.28 Antarctic
Table 1. RMSE (°C) of the climatic mean land surface temperature between the model simulation and NCEP reanalysis data (bold font indicates that the ECHAM5-CoLM simulation is better than ECHAM5)
To evaluate the simulation skill for seasonal cycles, Fig. 3 shows the seasonal variation of zonal mean soil temperature in the sub-surface layer (~1 cm). ECHAM5 and ECHAM5-CoLM simulations both show a decrease in soil temperature with latitude in both the Northern and Southern Hemispheres. Peaks in soil temperature are mostly simulated to occur during the summer months (June–July–August; JJA). The highest soil temperatures (> 30°C) are simulated in the subtropical regions in the two models.
Figure 3. Climatological annual cycles of zonal mean land surface temperature (°C) from (a) NCEP dataset, (b) ECHAM5, and (c) ECHAM5-CoLM simulations.
The area over which the mean land surface temperature is > 30°C is clearly larger in the ECHAM5 simulation than that in the ECHAM5-CoLM simulation and NCEP-2 dataset. The soil temperature in the equatorial region (from about 10°S to 10°N) decreases by about 5°C at the beginning of June and then recovers from the end of July (Fig. 3a). The ECHAM5-CoLM simulation almost captures this seasonal variation (Fig. 3c), but this feature is not completely simulated by ECHAM5, which shows no significant downward trend in soil temperature of the equatorial region over one year (Fig. 3b).
More robust comparisons with in situ observations are used to validate simulations. ECHAM5 and ECHAM5-CoLM are compared with observations of the monthly mean soil temperature at 20- (Fig. 4a) and 80-cm (Fig. 4b) depths at Valdai Station. ECHAM5 and ECHAM5-CoLM both capture the seasonal cycle of soil temperature. ECHAM5-CoLM is a better match to the observations, whereas the phase of ECHAM5 is ahead of the observations. RMSEs between the simulations and observations in Table 2 suggest that the ECHAM5-CoLM simulation is superior to ECHAM5 simulation.
Figure 4. Time series of the monthly mean soil temperature (°C) from observational data as well as ECHAM5 and ECHAM5-CoLM simulations at Valdai Station (57.6°N, 33.1°E) from 1980 to 1985 at depths of (a) 20 and (b) 80 cm.
Depth (cm) ECHAM5 ECHAM5-CoLM 20 6.51 2.75 80 5.44 2.74
Table 2. RMSE (°C) of the monthly mean soil temperature between simulations and observations at Valdai Station (57.6°N, 33.1°E) from 1980 to 1984 (bold font indicates that the ECHAM5-CoLM simulation is better than ECHAM5)
CoLM defines 10 soil layers (0–1.8, 1.8–4.5, 4.5–9.1, 9.1–16.6, 16.6–28.9, 28.9–49.3, 49.3–82.9, 82.9–138.3, 138.3–229.6, and 229.6–343.3 cm), which means that there are 10 different soil moisture outputs from ECHAM5–CoLM. However, due to its simple physical parameterization of the mean transfer of heat and water between the layers of a soil, ECHAM5 can only produce a mean soil moisture content within one soil layer. As the CPC monthly soil moisture reanalysis dataset gives only one-layer soil moisture content at 1.6-m depth, we integrate the modeled soil moisture content from the surface to 1.6 m and then compare the results with the CPC reanalysis dataset. The total soil water content is integrated from the surface to the eighth (1.38 m) plus part of the ninth (0.22 m) layer in ECHAM5-CoLM, whereas in ECHAM5 the total soil water is calculated by multiplying the soil moisture content at 1.6-m depth.
Figure 5 shows the spatial distribution of climatological annual mean soil moisture content. The soil moisture from ECHAM5 simulation is much lower than that of CPC simulation, whereas ECHAM5-CoLM simulation is closer in magnitude to ECHAM5 simulation. Global mean values of the ECHAM5, ECHAM5-CoLM, and CPC simulations are about 30.9, 232.6, and 196.3 mm, respectively. The spatial pattern of ECHAM5-CoLM matches that of the CPC simulation better than that of ECHAM5 simulation. CPC (Fig. 5a) and ECHAM5-CoLM (Fig. 5c) simulations show a high soil moisture content in southern East Asia, Indonesia, central Africa, southeast North America, and the Amazon basin, consistent with previously reported results (Balsamo et al., 2011; Albergel et al., 2012). By contrast, ECHAM5 predicts a high soil moisture content in different regions: northern East Asia, southern Africa, northwest North America, and southern Amazon basin (Fig. 5b). ECHAM5 also simulates a wetter Australia and a drier Indonesia, opposite of the CPC simulation.
Figure 5. Climatological annual mean total soil moisture content (mm) in a 160-cm column from (a) CPC, (b) ECHAM5, and (c) ECHAM5-CoLM simulations.
The seasonal variation in soil moisture content is mostly captured by ECHAM5-CoLM, but not ECHAM5 in the six regions of interest (Fig. 6). The observed soil moisture content from ERA-Interim dataset peaks during the summer months (JJA) on the Tibetan Plateau, in South America, and on the Indochina and Indian peninsulas. ECHAM5-CoLM simulates these peaks well and is close to ERA-Interim data. ECHAM5 systematically underestimates the soil moisture content in the six regions and is unable to reproduce the seasonal variation. ECHAM5-CoLM produces a large amplitude seasonal variation in Siberia and North America, which is probably due to that there is too much soil ice in soil layers of the model in these regions.
Figure 6. Climatological annual cycles of simulated and reanalysis total soil water content (mm) for (a) Tibetan Plateau (30°–50°N, 80°–100°E), (b) Siberia (50°–65°N, 60°–130°E), (c) North America (7°–72°N, 168°–56°W), (d) South America (53°S–12.5°N, 81°–35°W), (e) Indochina Peninsula (10°–20°N, 90°–110°E), and (f) Indian Peninsula (8°–37°N, 61°–97°E).
Figure 7 shows the spatial distribution of the climatological annual mean sensible heat flux of the MTE product (Fig. 7a), ERA-Interim dataset (Fig. 7b), ECHAM5 simulation (Fig. 7c), and ECHAM5-CoLM simulation (Fig. 7d). Spatial patterns of the two models (ECHAM5 and ECHAM5-CoLM) are similar to those of the observational and reanalysis data (the MTE product and ERA-Interim dataset), especially in mid to low latitudes. Global mean (60°S–80°N) values for the MTE product, ERA-Interim dataset, and ECHAM5 and ECHAM5-CoLM simulations are 34.4, 24.0, 19.0, and 35.3 W m–2, respectively. Note that ECHAM5-CoLM simulates the closest sensible heat flux to observations (MTE product).
Figure 7. Climatological annual mean sensible heat flux (W m–2) from (a) MTE product, (b) ERA-Interim reanalysis dataset, (c) ECHAM5, and (d) ECHAM5-CoLM.
The spatial distribution of climatological annual mean latent heat flux is shown in Fig. 8. It is clear that spatial patterns of the modeled latent heat flux are close to those for the observations and reanalysis data. Values of the global mean (60°S–80°N) latent heat flux for the MTE product, ERA-Interim dataset, and ECHAM5 and ECHAM5-CoLM simulations are 35.7, 49.8, 27.7, and 42.7 W m–2, respectively. ECHAM5-CoLM simulates a latent heat flux that is closer to the MTE product and ERA-Interim dataset than that by ECHAM5.
Figure 8. As in Fig. 7, but for climatological annual mean latent heat flux (W m–2).
Seasonal cycles of the zonal mean sensible heat flux and latent heat flux are shown in Figs. 9, 10, respectively. The two centers of sensible heat flux > 40 W m–2 in subtropical regions are reproduced by both EHCAM5 and ECHAM5-CoLM simulations. ECHAM5-CoLM simulates two stronger centers than ECHAM as well as observational and reanalysis data (Fig. 9). The pattern of latent heat flux simulated by ECHAM5-CoLM matches the observations better than that by EHCAM5 (Fig. 10). The latent heat flux in tropical regions simulated by ECHAM5 is too weak, perhaps because physical processes of evapotranspiration by vegetation are too simple in ECHAM5.
Figure 9. Climatological annual cycles of zonal mean sensible heat flux (W m–2) for (a) MTE product, (b) ERA-Interim dataset, (c) ECHAM5, and (d) ECHAM5-CoLM.
Figure 10. As in Fig. 9, but for zonal mean latent heat flux (W m–2).
Figure 11 compares the spatial distribution of the climatological annual mean precipitation from the models with reanalysis data. The general pattern simulated by the models resembles that of GPCP reanalysis data. However, the simulated precipitation is slightly heavier than the observed precipitation in many of the major rainfall centers, for example, the intertropical convergence zone, south Pacific convergence zone, and tropi-cal Indian Ocean. The simulated south Pacific convergence zone extends further east than GPCP rainfall. However, the precipitation averaged over the low-latitude region (30°S to 30°N) (or the global land surface) by the GPCP, ECHAM5, and ECHAM5-CoLM simulations is 2.97 (1.82), 3.39 (1.87), and 3.26 (1.82) mm day–1, respectively. Precipitation over the land surface simulated by ECHAM5-CoLM is closer to the observed data than that of ECHAM5. The positive bias of summer (JJA) precipitation over China simulated by ECHAM5 is efficiently reduced by ECHAM5-CoLM, especially over the southern Tibetan Plateau as well as middle and lower reaches of the Yangtze River (Fig. 12). The overestimated rainfall around the southern slope of Himalaya decreases by > 6 mm day–1. Note that the two-step shape-preserving advection scheme (Yu, 1994) is not used in ECHAM5 in this study. Precipitation over the Yangtze River region also decreases by about 2–4 mm day–1. ECHAM5-CoLM therefore improves the simulation skill of summer monsoon precipitation in China.
Figure 11. Climatological annual mean precipitation (mm day–1) for (a) GPCP, (b) ECHAM5, (c) ECHAM5-CoLM, (d) ECHAM5 minus GPCP, and (e) ECHAM5-CoLM minus GPCP.
A Taylor diagram is used to compare the performance of ECHAM5-CoLM and ECHAM5 simulations (Fig. 13). ECHAM5-CoLM shows a better pattern correlation in the simulation of the land surface temperature, sensible heat flux, latent heat flux, and precipitation from 60°S to 90°N. RMSEs of the land surface temperature and sensible heat flux in ECHAM5-CoLM are better than those in ECHAM5, whereas RMSEs of the latent heat flux and precipitation are close to those in ECHAM5. Precipitation in ECHAM5-CoLM clearly has a smaller bias, whereas there is no improvement in sensible and latent heat fluxes. The Taylor diagram therefore indicates the better performance of ECHAM5-CoLM in modeling the spatial distribution and variability of meteorological variables.
|North America |
|Africa and Austria |
|South America |