Zhou et al. 10.1073/pnas.0700290104. |
Fig. 6. Spatial patterns of observed annual mean trends in DTR (°C/10 yr), Palmer Drought Severity Index (PDSI) (1/10 yr), and precipitation (mm/day /10 yr) for each 5°by 5° grid box during the period of 1950-2004. The PDSI dataset is from Dai et al. (1) and the negative trends represent drought. The precipitation dataset is from Chen et al. (2). The trends marked with "x" within the grid boxes are statistically significant at the 5% level. A two-tailed student's t test was used to test whether the trend at each grid box differs significantly from zero. For the DTR data (3), only grid boxes with at least 7 months data for each year and at least 31 years of data during the period of 1950-2004 were analyzed.
1. Dai A, Trenberth KE, Qian T (2004) J Hydrometeorol 5:1117-1130.
2.Chen M, Xie P, Janowiak JE, Arkin PA (2001) J Hydrometeorol 3:249-266.
3. Vose RS, Easterling DR, Gleason B (2005) Geophys Res Lett 32:L23822, doi:10.1029/2005GL024379.
Fig. 7. Model grid cells at about 2.8° by 2.8° (dashed line) and observational grid boxes at 5° by 5° (solid line) (1) over our study region. The observations (10°N to 20°N, 15°W to 20°E) are mainly located within four rows of model grid cells, with edges at 8.4° to 19.5°N and 18.3°W to 21.1°E. The vegetation amount is very sparse beyond 16.7°N (the fourth row) and increases southward between 8.4° and 16.7°N (the first three rows). Regional averaging of model outputs was calculated over the inner two rows (11.2° to 16.7°N and 15.5°W to 19.7°E), the center of our study region.
1. Vose RS, Easterling DR, Gleason B (2005) Geophys Res Lett 32:L23822, doi:10.1029/2005GL024379.
Fig. 8. (Upper) Seasonal cycle of leaf area index (LAI), stem area index (SAI), and leaf and stem area index (LSAI) averaged over the Sahel. (Lower) Spatial patterns of annual mean LSAI over the Sahel in the CLM3.1.
Fig. 9. Diurnal cycle of simulated annual mean differences (°C) in ground temperature (Left), surface air temperature (Center), and lowest atmospheric level temperature (Right) between the four experiments (NVLE, NV, HVLE, and HV) and the control run (CTL) under clear-sky (Upper) and all-sky (Lower) conditions averaged over the Sahel. Daily averaged low cloud fraction <20% is defined as clear-sky conditions. The model outputs are 3-hr averages from each time step (20 min). The values marked with triangles (diamonds) are statistically significant at the 5% (10%) level. A two-tailed Student's t test was used to test whether the difference differs significantly from zero.
Fig. 10. Spatial patterns of simulated annual mean ground temperature differences (°C) between the four experiments (NVLE, NV, HVLE, and HV) and the control run (CTL) during nighttime (4-6 a.m.) and daytime (4-6 p.m.) under clear-sky conditions in the Sahel. Stippling shows grid cells where the differences are statistically significant at the 5% level. The significance test was done as for SI Fig. 9.
Fig. 11. Same as SI Fig. 10 but for lowest atmospheric level temperature.
Fig. 12. Same as SI Fig. 9 but for surface radiation and energy budget (W/m2). The variables include net longwave radiation, net solar radiation, soil flux, sensible heat, and latent heat. The soil flux is defined positive toward the ground and the other variables are defined positive toward the atmosphere. The values marked with triangles (squares) are statistically significant at the 5% (10%) level.
Fig. 13. Same as SI Fig. 10 but for net longwave radiation (W/m2). The net longwave radiation is defined as positive toward the atmosphere.
Fig. 14. Same as SI Fig. 10 but for outgoing longwave radiation (W/m2). The outgoing longwave radiation is defined as positive toward the atmosphere.
Fig. 15. Same as SI Fig. 10 but for latent heat flux (W/m2). The latent heat flux is defined as positive toward the atmosphere.
Fig. 16. Same as SI Fig. 10 but for sensible heat flux (W/m2). The sensible heat flux is defined as positive toward the atmosphere.
Fig. 17. Same as SI Fig. 10 but for soil flux (W/m2). The soil flux is defined as positive toward the ground.
Fig. 18. Same as SI Fig. 10 but for net solar radiation (W/m2). The net solar radiation is defined as positive toward the ground.
Fig. 19. (Upper) Seasonal cycle of observed (1) and simulated rainfall (mm/day) in the control run (CTL) averaged over the Sahel. (Lower) Simulated rainfall differences between the four experiments (NVLE, NV, HVLE, and HV) and the CTL averaged over the Sahel. The differences marked with triangles (squares) are statistically significant at the 5% (10%) level. The significance test was done as for SI Fig. 9.
1. Chen M, Xie P, Janowiak JE, Arkin PA (2001) J Hydrometeorol 3:249-266.
Fig. 20. Seasonal cycle of measured 3-year average Bowen ratio and rainfall (mm/day) (Upper) and sensible heat flux (SH) (W/m2) and latent heat flux (LH) (W/m2) (Lower), at the AmeriFlux site, Vaira Ranch, together with that of the control run (CTL) at the corresponding model grid cell where this site is located.
Fig. 21. Monthly average sensible heat (W/m2), latent heat (W/m2), and Bowen ratio at the HAPEX-Sahel site, Super Site Central West, together with the seasonal cycle of the corresponding model grid cell in the CTL, NVLE, and NV. The sensible and latent heat fluxes were estimated in a 10-min interval using three methods, eddy covariance method (ECM), gradient method (GM), and Bowen ratio method (BRM). We estimated the monthly average measured fluxes from 2 days in August, 5 days in September, and 2 days in October. (Left) Results for all 12 months. (Right) Only the results from August to October.
Table 2. Differences in annual mean ground and surface air temperature, radiation and energy budget, and surface albedo, emissivity, and roughness length between the four experiments (NVLE, NV, HVLE, and HV) and the control run (CTL) averaged over the Sahel
Variable | NVLE - CTL | NV -CTL | HVLE - CTL | HV -CTL | NVLE - NV | HVLE - HV |
Nighttime (4-6 a.m.) | ||||||
Net longwave radiation, W/m2 | -0.76 | 1.57 | -1.17 | 0.82 | -2.33 | -1.99 |
Latent heat, W/m2 | 0.77 | 0.48 | 0.40 | 0.23 | 0.28 | 0.17 |
Sensible heat, W/m2 | 3.80 | 2.30 | 2.00 | 1.86 | 1.50 | 0.15 |
Soil flux, W/m2 | -3.81 | -4.36 | -1.24 | -2.91 | 0.55 | 1.67 |
Net solar radiation, W/m2 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
Ground temperature, °C | 0.70 | 0.00 | 0.49 | 0.12 | 0.69 | 0.36 |
2-m surface air temperature, °C | 1.30 | 0.72 | 0.74 | 0.50 | 0.59 | 0.24 |
Daytime (4-6 p.m.) | ||||||
Net longwave radiation, W/m2 | -5.86 | 0.11 | -6.16 | 0.34 | -5.97 | -6.51 |
Latent heat, W/m2 | -8.27 | -8.67 | -3.72 | -3.32 | 0.39 | -0.40 |
Sensible heat, W/m2 | -6.46 | -13.35 | -2.81 | -8.36 | 6.90 | 5.55 |
Soil flux, W/m2 | 10.66 | 10.90 | 4.53 | 7.24 | -0.24 | -2.71 |
Net solar radiation, W/m2 | -9.96 | -11.04 | -8.18 | -4.11 | 1.09 | -4.07 |
Ground temperature, °C | 1.46 | 0.79 | 0.69 | 0.67 | 0.67 | 0.01 |
2-m surface air temperature, °C | 0.33 | -0.08 | 0.10 | 0.12 | 0.41 | -0.02 |
Surface albedo, emissivity, and roughness length | ||||||
Albedo | 0.014 | 0.014 | 0.006 | 0.008 | 0.000 | -0.002 |
Soil emissivity | 0.070 | 0.000 | 0.070 | 0.000 | 0.070 | 0.070 |
Vegetation emissivitya | 0.357 | 0.357 | 0.178 | 0.178 | 0.000 | 0.000 |
Roughness length, m | 0.046 | 0.046 | 0.023 | 0.023 | 0.000 | 0.000 |
The net longwave radiation, net solar radiation, sensible heat, and latent heat are defined as positive toward the atmosphere, and the soil flux is defined positive toward the ground. The temperatures and radiation and energy budget are shown for clear-sky conditions and were 3-hr averages from each model time step (20 min). Each model grid consists of up to four patches of bare soil and plant functional types. The values shown here represent grid average values. aVegetation emissivity represents the fraction of ground longwave radiation that is shaded locally by vegetation, calculated as 1 - e-LSAI, where LSAI is leaf and stem area index.
SI Text
Uncertainties in Model Simulations and Observations
Our quantification of vegetation-soil impacts on DTR has various uncertainties. First, the seasonal cycle of simulated land surface temperature and precipitation is weaker than observations due to deficiencies in the simulations of the atmospheric model rather than those of the land surface processes (1). These include an excessive amount of rainfall and clouds, a large negative bias in downward longwave radiation, and a cold bias. The cold bias over the Sahel during the wet season will cancel part of our simulated warming at that time from the reduction in soil emissivity and vegetation cover shown in the main text.
Second, the land model has some deficiencies, and some land surface parameters or processes were not considered. The vegetation was not a dynamic component of the model, but was prescribed based on a new land surface dataset derived from MODIS products during the 2001-2003 periods (2). Interannual variations of sea surface temperature were not considered to exclude their impacts on surface air temperature. No substantial rainfall reduction was simulated and thus no significant drought occurred in any of four experiments relative to the control run (SI Fig. 19).
One major concern regarding these weak aspects of the model is whether the inability to simulate drought in the model will invalidate or weaken our conclusions. Our simulations showed a strong and significant warming in Tmin and a small and insignificant warming in Tmax (and thus the DTR decreased substantially), after the soil emissivity and/or vegetation amount were reduced. The magnitude of such changes was largest during dry seasons (largest Bowen ratio) and smallest during wet seasons (smallest Bowen ratio). The inability to simulate drought in our experiments in fact weakens our simulated warming in wet seasons, i.e., if the model handled drought well, we would see a stronger warming in Tmin and a larger decrease in DTR due to overestimated evapotranspiration (ET) than those shown in the main text. From the point of view of energy budget, our experiments would have overestimated latent heat due to overestimated ET (thus underestimated Bowen ratio) relative to those in the presence of drought. As a result, more energy was dissipated into the form of latent heat rather than sensible heat that otherwise would be used to warm the temperature. Although there is no drought simulated in our experiments, the vegetation amount and soil emissivity were reduced to reflect drought-related changes. Therefore, the weak aspect of the model should not invalidate or weaken, but strengthen, our conclusions.
Another concern is whether the drought-related changes in soil albedo will change our conclusions. It is expected that the surface albedo should increase with drought as well. In our experiments, the soil albedo was set as a constant value (i.e., no modification). This is not true when drought occurs, but more incoming radiation due to less cloud cover largely cancels such effect over drought regions as shown in many previous studies. Since our simulations show no significant changes in cloud cover and incoming solar radiation, the inability to simulate drought in the model should have a small impact on our results contributing from changes in soil albedo.
The modeling of the canopy radiation environment as implemented in current land climate models such as CLM3.1 is deficient in its partitioning of radiation between vegetation canopies and underlying surfaces due to its unrealistic assumption of plane parallel canopy geometry. As a consequence, the canopy shading effect on its underlying bright soil may be mostly underestimated over sparsely vegetated regions such as the Sahel.
Third, some missing data in the observations and observational uncertainties, especially for cloud cover, may bias the quantification of the relations between DTR and precipitation/clouds and the estimate of the observed linear trends.
Changes in Modeled Surface Radiation and Energy Budget
Changes in the model-simulated surface radiation and energy budget are critical for interpreting the changes in the model-simulated DTR, as discussed in the main text and shown in SI Figs. 9-18. To provide additional information about such changes, here we first validate the model-simulated sensible and latent heat fluxes in the control run (CTL), and then discuss changes in surface albedo, emissivity, roughness, and radiation and energy budget in our four experiments (NVLE, NV, HVLE, and HV) relative to the CTL.
Validation of Sensible and Latent Heat Fluxes in the CTL.
Realistic characterization of the partitioning between sensible and latent heat is needed for an accurate simulation of land surface air temperature. Validating large-scale model results with site field measurements is difficult, especially for semiarid ecosystems due to the land heterogeneity and the spatiotemporal variability in rainfall. Nevertheless, we compared our modeled results in the CTL with field measurements at two flux sites over semiarid systems, indicating that the model adequately simulates rainfall and surface fluxes over the Sahel.More than 400 tower sites are currently operating on a long-term and continuous basis in FLUXNET, a global network of micrometeorological tower sites that measure the exchanges of carbon dioxide, water vapor, and energy between terrestrial ecosystem and atmosphere (www.daac.ornl.gov/FLUXNET/). Although some sites with several years of measurements are available for savanna ecosystems in FLUXNET, none is located in the Western African Sahel where our study region is located. There are several savanna sites in AmeriFlux (part of FLUXNET), but almost all sites have gaps and only one site, Vaira Ranch, is gap-filled (ftp://cdiac.ornl.gov/pub/ameriflux/data/Level4/). Vaira Ranch is located at 38.41°N, 120.95°W in Ione, California. It is a grassland opening in a region of oak/grass savanna. Continuous measurements are available from 2001 to 2003. We compared the seasonal cycle of measured 3-year average rainfall, sensible heat, latent heat, and Bowen ratio (defined as sensible heat divided by latent heat) at Vaira Ranch, to that of modeled 18-year average values in the CTL from the corresponding model grid cell where Vaira Ranch is located (SI Fig. 20). Evidently, the model simulates the seasonal cycle of rainfall, surface fluxes, and Bowen ratio very well over this savanna ecosystem.
There are limited field measurements over our study region from HAPEX-Sahel (Hydrological and Atmospheric Pilot Experiment in the Sahel). The HAPEX-Sahel was an international land-surface-atmosphere observation program undertaken in western Niger, in the Western African Sahel (www.ird.fr/hapex/htdocs/whatis.htm). It obtained measurements of atmospheric, surface, and certain subsurface processes in a 1° ´ 1° grid cell that contains most major land surface types found throughout the Sahel. An intensive operations period was undertaken for 8 weeks from mid to late growing season of 1992. The 1° ´ 1° grid cell consisted of three super sites (southern, central, and east), each with three or more subsites. The southern and central west sites were intended primarily for surface flux and energy balance studies. Although there are several sites available, only one site, Super Site Central West, GRID 5154, has a complete data description document. This site is located at 13.55°N, 2.5°E and has measurements for only 9 days during the wet season, 2 days in August, 5 days in September, and 2 days in October. The sensible and latent heat fluxes were estimated in a 10-min interval using three methods, eddy covariance method (ECM), gradient method (GM), and Bowen ratio method (BRM). We estimated the monthly average from these 9 days.
SI Fig. 21 shows the monthly average sensible heat, latent heat, and Bowen ratio at Super Site Central West, together with the seasonal cycle of modeled values from the corresponding model grid cell where the site is located in the CTL. Since the model results are grid averages for 18 years forced by the climatology of sea surface temperature whereas the site measurements were only from several days in 1992, some differences should be expected between the measured and modeled. Nevertheless, the model generally simulates the observed seasonal trend of surface fluxes and Bowen ratio, especially those from the GM and BRM, from August to October. For the same site, three methods gave different estimates of surface fluxes. The seasonal cycle of modeled Bowen ratio is consistent with that of rainfall (SI Fig. 19), with the maximum value during the dry season and the minimum values during the wet season, as shown for Vaira Ranch in SI Fig. 20. This is expected because latent heat is determined largely by the availability of water. During the early wet season, rainfall events are widely separated in time and the soil surface dries quickly, but later the frequency of rain increases and the soil remains wetter for longer periods. Consequently, soil evaporation increases significantly and so does vegetation evapotranspiration.
Changes in Surface Albedo, Emissivity, Roughness, and Radiation and Energy Budget in Our Simulations.
Understanding concurrent changes in the model-simulated surface albedo, emissivity, roughness, and radiation and energy budget helps us interpret the simulated changes in temperature. Below is a simple description how the land surface albedo, emissivity, and roughness were calculated over the Sahel in the CLM3.1 (details in ref. 3).The land surface albedo at each model grid is calculated from soil and vegetation albedos. The soil albedo varies with soil color and decreases with increasing soil moisture. The CLM3.1 uses eight soil color types globally from dark to light, and each color has prescribed seasonally constant albedo values. The vegetation albedo is calculated by a two-stream scheme. In our simulations, the prescribed soil albedos were unchanged and so the change in albedo at each model grid is mainly from the change in vegetation, and to a lesser extent from the change in soil moisture.
The land surface emissivity at each model grid is calculated separately for vegetation and underlying soil. The soil emissivity is set as a global and seasonal constant value of 0.96 for bare soil in the model. The leaves and stems of vegetation are taken to have an emissivity of 1.0. For a vegetated patch in a model grid cell, the fraction of ground longwave radiation that is shaded locally by vegetation is modeled by 1 - e-LSAI (referred to as vegetation emissivity), where LSAI represents leaf and stem area index. Hence, reduced vegetation cover exposes more soil directly to the atmosphere, increasing the importance of its lower emissivity on both the absorption and emission of longwave radiation. The net longwave radiation flux over the land surface in the model is calculated as a sum of fluxes for vegetation and its underlying ground. For nonvegetated surfaces (LSAI = 0), the flux is only from the ground.
The land surface roughness length at each model grid consists of nonvegetated and vegetated components. The vegetated component is calculated as a function of canopy top height, which is a prescribed constant value by vegetation type. The nonvegetated soil roughness is calculated as a function of the roughness Reynolds number (4). In our experiments relative to the CTL, changes in vegetation should be the major contributor to changes in surface roughness length over vegetated areas, whereas changes in soil roughness length are expected to be small over nonvegetated areas.
SI Table 2 lists the differences in annual mean albedo, emissivity, roughness length, surface radiation and energy budget, and ground and surface air temperature between the four experiments (NVLE, NV, HVLE, and HV) and the CTL averaged over the Sahel. Such differences represent grid average values rather than the averages over the vegetated patches. The removal of vegetation increased the grid average values of albedo by 0.006-0.014 and roughness length by 0.023-0.046 m in the four experiments. It decreased the grid average values of vegetation emissivity by 0.178-0.357, which represents the fraction of longwave radiation that is shaded locally by vegetation (e.g., 1 - e-LSAI). Hence, reduced vegetation cover exposes more soil directly to the atmosphere. The soil emissivity decreased by 0.07 in NVLE and HVLE, and it remained unchanged in NV and HV relative to CTL. These changes modified the surface radiation and energy budget and thus ground and surface air temperature. Below we discuss such modifications with emphasis on the results of NVLE - CTL, NV - CTL, and NVLE - NV, and those of HVLE - CTL, HV - CTL, and HVLE - HV can be explained similarly.
The reduction of vegetation emissivity alone increased the net longwave radiation (i.e., more outgoing longwave radiation) by 1.6 W/m2 during daytime and 0.1 W/m2 during nighttime (NV - CTL). The reduction of soil emissivity alone reduced the net longwave radiation (i.e., less outgoing longwave radiation) by 6.0 W/m2 during daytime and 2.3 W/m2 during nighttime (NVLE - NV). Evidently, the reduction in soil versus vegetation emissivity had an opposite effect on the net longwave radiation (NVLE - CTL) because vegetation with zero heat capacity reaches colder temperatures than the soil at night (SI Figs.13 and 14). The removal of vegetation increased the soil heat storage by 10.7-10.9 W/m2 during daytime and soil heating by 3.8-4.4 W/m2 during nighttime (NVLE - CTL and NV - CTL). It also decreased the latent heat by 8.3-8.7 W/m2 during daytime and had little effect during nighttime. The sensible heat decreased by 6.5-13.4 W/m2 during daytime and increased by 2.3-3.8 W/m2 during nighttime (NVLE - CTL and NV-CTL). It was increased by 6.9 W/m2 during daytime and by 1.5 W/m2 during nighttime after the reduction in soil emissivity (NVLE - NV). Consequently, the ground (surface air) temperature increased by 0.8-1.5°C (-0.1 to -0.33°C) during daytime and by 0.0-0.7°C (0.7-1.3°C) during nighttime.
During nighttime, increased soil heating (mainly over vegetated regions) and reduced outgoing longwave radiation (mainly over nonvegetated regions) are the two dominant factors in increasing the ground temperature and hence the air temperature and sensible heat for all experiments (SI Table 2). During daytime, decreased net solar radiation, reduced roughness, and increased soil heat storage (all mainly over vegetated regions), together with reduced outgoing longwave radiation (mainly over nonvegetated regions), decrease both the sensible and latent heat but with little change of the air temperature. The daytime temperatures of the atmosphere are affected much less by a given change in radiation and energy budget because of the depth of the daytime convective boundary layer. When only the vegetation is removed, the increase of Tmin is controlled by an increase of outgoing longwave radiation balancing the increased release of soil thermal storage. When the reduction of soil emissivity is also considered, the reduced outgoing longwave radiation further increases Tmin.
The reductions in both latent and sensible heat in our simulations are consistent with previous studies (e.g., refs. 5 and 6). There are three possible reasons. First, less net radiation due to increased albedo and soil heat storage during daytime and less surface energy exchanges due to smaller surface roughness reduce sensible and latent heat. Second, the CLM3.1 has a recognized bias in excessive evaporation (1). After examining changes in latent heat for each component of soil and vegetation, we found a comparable increase in soil evaporation relative to the decrease in vegetation evapotranspiration after the removal of vegetation. It suggests that the excessive evaporation in the model may compensate the reduction of evapotranspiration, because vegetation evapotranspiration may be larger than soil evaporation (e.g., ref. 7). The excessive evaporation in our simulations will increase the latent heat and thus weaken our simulated warming during the wet season from the reduction in soil emissivity and vegetation cover shown in the main text. Third, our simulations show no drought as discussed previously and thus no significant changes in cloud cover, downward solar and longwave radiation, but in reality, an increase in incoming solar radiation is expected Because of the drought-associated reductions in cloud cover and rainfall.
Our simulations show that the Bowen ratio increased during the dry season but changed little in the wet season after the removal of vegetation (SI Fig.21). We did not see a substantial increase in the wet season Bowen ratio, possibly because (i) the soil evaporation was overestimated in our simulations as discussed above, and (ii) latent heat is much greater than sensible heat in the wet season and thus a small change in sensible heat little modifies the Bowen ratio. In addition, since the vegetation amount in the Sahel is sparse, removing vegetation may have a smaller effect on Bowen ratio than removing tropical forests as done in many previous studies.
1. Dickinson RE, Oleson KW, Bonan G, Hoffman F, Thornton P, Vertenstein M, Yang ZL, Zeng XB (2006) J Clim, CCSM Special Issue 19(11):2302-2324.
2. Tian Y, Dickinson RE, Zhou L, Shaikh M (2004) J Geophys Res 109:D20115, doi:10.1029/2003JD004499.
3. Oleson KW, Dai Y, Bonan G, Bosilovich M, Dickinson R, Dirmeyer P, Hoffman F, Houser P, Levis S, Niu G-Y, et al. (2004) Technical Description of the Community Land Model (CLM) (Natl Center Atmos Res, Boulder, CO), NCAR Tech Note, NCAR/TN-461+STR.
4. Zeng X, Dickinson RE (1998) J Clim 11:537-550.
5. Dirmeyer PA, Shukla J (1994) J Geophys Res 99(D10):20863-20878.
6. Hoffmann WA, Jackson RB (2000) 13:1593-1602.
7. Dickinson RE (1980) in Plowing in the Wind: Deforestation and Long-Range Implications, Studies in Third World Societies, ed Dickinson RE (College William and Mary, Williamsburg, VA), No. 14, pp 411-441.