1. Introduction
Evaporative fraction (EF; the ratio of latent heat flux to the sum of the latent plus sensible heat fluxes) can be measured in the field to an accuracy of approximately 10% (Sellers and Hall 1992; Hall and Sellers 1995). To what accuracy must soil moisture be known in order to simulate EF within this observational uncertainty? This question was posed at a workshop hosted jointly by the Global Energy and Water Cycle Experiment (GEWEX) and the International Geosphere Biosphere Programme (IGBP) on land surface parameterizations and soil–vegetation–atmosphere transfer schemes (International GEWEX Project Office 1998). A discussion group on soil moisture was charged with composing questions or hypotheses that could and should be addressed by the community with the current generation of land surface schemes (LSSs). It is not well known just how well LSSs need to perform the simulation of soil wetness. The question above was crafted to address this issue.
A second question, closely related to the first, also is relevant to land surface modeling and observation: Is there a firm relationship between the variability of soil moisture and surface fluxes? In other words, does the sensitivity of surface fluxes to variations in soil moisture fluctuate in a systematic way? Better knowledge of this key element would aid model development and the design and implementation of field campaigns and soil moisture monitoring networks.
If the necessary accuracy of soil moisture simulation can be quantified, it could provide a standard for evaluating LSSs and a guideline for their future development. It also would provide perspicacity toward the utility of melding in situ and model-produced land surface data, and understanding of the compatibility of the observations and models used in land surface data assimilation. Computational economy may be gained by focusing resources on the most critical components of LSSs.
If the necessary accuracy of soil moisture measurements can be quantified, the information could be used in instrument design. Soil moisture notoriously is difficult to measure, requiring the careful monitoring of expensive instrumentation or the laborious collection, preparation, and measurement of soil samples at a multitude of points in a region of interest. A relationship between soil moisture and surface flux sensitivity could lead to more economical measuring systems.
If spatial or temporal relationships between surface fluxes and soil moisture can be found, they may offer a means to simplify the methods of extrapolation into data-sparse areas. It may be found that accuracy is critical only in certain regions, providing guidance for deployment of resources in the field and the design of soil moisture observing networks.
We look for relationships in three different land surface schemes: the simplified Simple Biosphere model (SSiB; Xue et al. 1991), Mosaic (Koster and Suarez 1992, 1996), and the Biosphere–Atmosphere Transfer Scheme (BATS; Dickinson et al. 1993) by using results from the Global Soil Wetness Project (GSWP; Dirmeyer et al. 1999) and ancillary offline integrations. These models are among the current state-of-the-art LSSs but incorporate different approaches in many of their parameterizations. Thus, a measure of model uncertainty can be gained by contrasting the performance of the schemes.
The models themselves are compared in section 2. The experiments performed with the three schemes are described in section 3. Section 4 presents results, and conclusions are provided in section 5.
2. Models
LSSs calculate the fluxes of heat, moisture, momentum, and, in some cases, carbon and trace gases between the atmosphere and components of the land surface such as soil, vegetation, snow, and ice. LSSs also calculate the storage and distribution of heat, moisture, and sometimes carbon within those land surface components. LSSs in use today vary in sophistication from very simple one-layer models of water storage and evaporation to complex schemes with multiple layers, subgrid distributions of land surface components that may vary in time, explicit representation of the biophysical processes of plants, multiple phases of water and chemical storage and transport in the soil, and even the ability to simulate the growth, competition, and death of plant species or biomes over time.
The three LSSs examined here are all designed to be coupled to dynamical models of the atmosphere used in climate research. One of the main applications for LSSs in general is to provide the lower boundary conditions over land for global or regional general circulation models. These schemes all account for the effects of seasonally varying vegetation on the surface energy and water balances. All three LSSs generally reproduce observed fluxes, as shown by various tests in the Project for the Intercomparison of Land-Surface Parameterization Schemes (Chen et al. 1997; Wood et al. 1998). Table 1 provides a comparison of the main physical components of the three models.
a. BATS
BATS was introduced in the mid-1980s (Dickinson et al. 1986) for the purpose of coupling to the National Center for Atmospheric Research Community Climate Model, in order to represent land surface contributions to the large-scale energy and water budgets better. It also has been used offline and coupled with other atmospheric models (Seth et al. 1994; Timball and Henderson-Sellers 1998) and has been modified by many different research groups for a wide range of studies. Many of the changes incorporated into the currently distributed version of the model (BATS version 1e) are described in Dickinson et al. (1993). Some of the modifications currently in progress involve a new 10-layer soil model (Dickinson and Yang 1998), a new multilayer snow model (Jin et al. 1999), and a new canopy/carbon model (Dickinson et al. 1998). Some recent publications with more descriptions, validation studies, and sensitivity studies of BATS (both offline and coupled) include Gao et al. (1996), Yang and Dickinson (1996), Yang et al. (1997, 1998) and Hahmann and Dickinson (1997).
BATS currently allows for only a single-story canopy with one vegetation type (which could be an averaged combination of types) and one soil type per model point (or cell). Vegetated areas and nonvegetated bare soil areas are assumed to be distributed evenly. Vegetation and soil parameters are usually a function of vegetation and soil type, respectively. Fractional vegetation cover and other vegetation properties such as leaf area index vary seasonally as a function of deep soil temperature. BATS uses three soil layers, which share the common top surface at the soil–air interface but have bottom surfaces at increasing depths. All layers have homogeneous soil characteristics. Roots can occur unequally in both the surface and the rootzone–surface layers. Potential infiltration into the soil results from precipitation onto the soil plus leaf drip plus dew plus snowmelt less evaporation. Any water that cannot infiltrate immediately because of saturated soil conditions becomes surface runoff. Water is removed from the soil by transpiration, evaporation, and gravitational drainage.
b. SSiB
The Simple Biosphere model (SiB) first was described by Sellers et al. (1986). SiB’s treatment of moisture and heat fluxes among soil, vegetation, and atmosphere is analogous to the flow of electricity in a simple circuit of rheostats. Conductances vary in response to both specified and dynamic variations in the state of the soil, vegetation, and atmosphere. SiB was implemented as the land surface scheme in the Center for Ocean–Land–Atmosphere Studies (COLA) general circulation model (GCM), replacing the original bucket scheme (Sato et al. 1989), and in the National Aeronautics and Space Administration (NASA) Goddard Laboratory for Atmospheres (GLA) GCM (Sud et al. 1990).
Xue et al. (1991) developed a simplified version of SiB (SSiB) in response to criticism from the atmospheric modeling community that the original model was computationally expensive and more complex than truly was necessary for GCM applications. This model supplanted the full SiB scheme at COLA and GLA, although a version of the full SiB model is still in use by the Japanese Meteorological Association. A thorough history of the evolution and application of SSiB is given by Dirmeyer and Zeng (1997).
The offline version of SSiB consists of a main program that performs data input/output and balance checks, and subroutines that perform four general tasks:
determining the amount of radiant energy and water that will enter and interact with the canopy–soil system,
calculating the physical and physiological parameters used in the flux calculations,
calculating temperatures and the fluxes of heat, and
updating the moisture state variables.
c. Mosaic
The Mosaic LSS, which currently is used with the Aries GCM at NASA Goddard Space Flight Center, is a derivative of the SiB model of Sellers et al. (1986). Many components of the model (most notably the parameterization of stomatal resistance) mimic the corresponding components of SiB. Several important differences, however, give the Mosaic LSS a unique flavor.
Subgrid heterogeneity in surface characteristics is treated explicitly by subdividing each grid cell into several “tiles,” with each tile consisting of a single vegetation or bare soil type. The number and relative sizes of the tiles in a given grid cell are determined from vegetation maps. Independent energy and water balance calculations are performed over each tile, with moisture and temperature state variables for each tile maintained separately. Because each tile is homogeneous in surface cover (at least relative to the grid cell as a whole), the one-dimensional energy and water balances applied in each tile have added physical relevance.
Rather than compute a separate energy balance for the canopy and the ground beneath the canopy, as in SiB, the canopy–ground system is treated with a single energy balance calculation, and a single temperature represents both the canopy and the ground surface. Again, though, different vegetation and bare soil tiles in a single grid cell maintain different temperatures.
Various aspects of SiB’s hydrological parameterizations have been replaced. For example, canopy interception during storms accounts for temporal correlation in subgrid storm position in a very simple way that nevertheless prevents an excessive accumulation of interception water. Surface runoff, rather than forming through infiltration excess, essentially forms over a saturated fraction of the tile, parameterized in terms of surface water content.
The energy balance calculations, including the treatment of the unstable boundary layer, are formulated carefully to avoid iteration. Various other components of the original SiB code (e.g., the albedo and the temperature stress function) have been replaced by simplified parameterizations.
3. The experiment
As part of GSWP, 24-month stand-alone integrations were performed with each LSS covering 1987 and 1988. The LSSs were not coupled to an atmospheric model but instead were driven by gridded analyses of meteorological conditions. The integrations were performed at a 1° × 1° cell size covering all land points free of permanent ice, as determined from the International Satellite Land Surface Climatology Project (ISLSCP) Initiative I dataset (Meeson et al. 1995). Each model used the same spatial distributions of vegetation and soil types and properties and the same monthly evolving vegetation parameters. Figure 1 shows the distribution of vegetation used in the GSWP integrations. In addition, each model used the same meteorological forcings (near-surface air conditions, downward radiation, and precipitation). All of these data came from the ISLSCP Initiative I dataset and are described by Sellers et al. (1996). Soil wetness in each land surface scheme was spun up from an initial state equal to 75% of saturation by looping through the 1987 data until the soil state variables equilibrated. A complete description of the GSWP experiment setup and execution can be found in International GEWEX Project Office (1995).
These control integrations formed the basis of the examination of the sensitivity of surface fluxes to the vegetation and soil wetness properties. We have chosen to focus on July since it is the heart of the warm season (when vegetation is most active) in the Northern Hemisphere, and all land cover types are represented in the Northern Hemisphere.
Figure 2 shows the spatial distribution of surface soil wetness index, latent heat flux, and evaporative fraction for each model, averaged for July 1987 and 1988. BATS exhibits greater latent heat flux and less sensible heat flux than the other schemes. Mosaic and SSiB behave very similarly over sparsely vegetated areas (e.g., North Africa, Arabia, and central Asia). They show differences over vegetated regions, however, where SSiB has less latent heat flux in the warm regions but somewhat more at high latitudes. This discrepancy is reversed for sensible heat flux and leads SSiB to have a lower evaporative fraction at lower latitudes and a higher evaporative fraction over northern Asia in particular. All models show a strong tendency toward high evaporative fraction over areas of ample rainfall or dense vegetation and at high latitudes. Evaporative fraction is low over deserts and generally is high over forested areas.
Next, we assume that the “correct” soil moisture for the given LSS is that which was produced in the control integration for GSWP. It should be pointed out that by correct we do not imply that the model can duplicate observed soil moisture. In fact, it is debatable whether the gridded model variable called soil moisture is representative of any observable parameter (Koster and Milly 1997). That is the main reason why a soil wetness index is defined; for purposes of validation it is best to compare model indices with observed indices. Our assumption is that the soil moisture simulated by an LSS driven by observed meteorological forcings will be appropriate for that model and lead to the best possible simulation of surface fluxes by that model. Even this may not be the case if model parameters or parameterizations are flawed. Model errors are generally systematic in offline experiments such as this, however, so that in a sensitivity study the response to a change in forcing is presumed to be more reliable than the mean value itself.
4. Spatial distribution of sensitivity
Figure 3 shows the spatial distribution of the sensitivity of EF to variations in the root zone soil wetness index for each of the three schemes. This sensitivity has been calculated at each point using the control integration and pairs of perturbation cases in (4). For both positive and negative perturbations of SWI, the change in EF from the control case is calculated at each grid box for progressively larger values of d until a change in EF of at least 10% is found. The change in SWI that corresponds to a change in EF of 10% then is estimated by linear interpolation between the pair of bounding integrations. The values of d estimated for the positive and negative perturbation cases then are averaged together to arrive at a final value. If a perturbation of 10% cannot be induced by either the positive or negative case, then the value of d from the opposite case is used alone. If a 10% fluctuation of EF cannot be induced at all, the grid box is marked as insensitive.
The Mosaic scheme (middle panel) has large regions where surface fluxes are very sensitive to variations in soil wetness index. Virtually all arid and semiarid regions and many areas of the polar latitudes show values of d of 0.10 or less. This high sensitivity probably is due to the tiling approach in Mosaic; sparsely vegetated regions include a bare soil tile of significant size. Evaporation from this bare soil tile is strongly sensitive to soil moisture. There is little area with moderate sensitivity. Many forested areas, particularly in the Tropics, Europe, east Asia, and eastern North America show little sensitivity, presumably because the bare soil fraction there is small.
BATS shows a similar pattern, with a general shift toward less sensitivity. The area with the highest sensitivity is much smaller and is confined to the more arid regions, although there are still large areas of Siberia, Alaska, and northwestern Canada that have high sensitivities. The region of insensitivity is somewhat larger than for Mosaic, and there are also more grid boxes showing moderate sensitivity.
The SSiB scheme is by far the least sensitive of the three. Virtually all forested areas show no sensitivity of surface fluxes to soil wetness perturbations about the GSWP climatological mean. There are large areas of moderate sensitivity, including much of North Africa and central Asia, that are highly sensitive in the other two models. The largest coherent regions of high sensitivity for SSiB are in the Southern Hemisphere, in contrast to the other schemes.
Figure 4 shows the frequency distribution of July root zone soil wetness index for the control integrations of the three schemes as a percentage of all ice-free land grid boxes. Bins of width 0.1 are used except at the extremes, where terminal bins of width 0.01 are applied. Both BATS and Mosaic have more grid points in the dry bin (0.01–0.1) than in any other. BATS shows a fairly uniform distribution throughout the other bins from 0.1 through 0.99, although there is a weak secondary maximum near 0.7. Mosaic has a smoother decrease from the dry maximum to the middle range of soil wetness index. Thus, there is a distinct dry tendency in Mosaic. SSiB shows the opposite tendency, with a maximum at the wet end of the range near 0.8. The vertical lines on each distribution give the approximate median SWI for each model. All schemes also show at least a hint of a bimodal distribution, with a local minimum in the middle range of soil wetness index.
Figure 4 suggests therefore that, under identical meteorological forcings, SSiB is the stingiest at releasing soil moisture as evaporation. Both BATS and Mosaic are much more capable of attaining a condition of very dry soil over a large number of grid boxes. Interestingly, although SSiB shows a wet bias, it also has the greatest number of points near or below the wilting point (less than 0.01). This result most likely is due to an enforced baseflow in the model that slowly drains water from the lowest layer at all times (Xue et al. 1996).
5. Conceptual model of sensitivity
Figures 5 through 8 show the scatter of July EF versus SWI for all Northern Hemisphere grid points of the vegetation categories: forest, grass and shrub, tundra, and bare soil, respectively. Rather than attempt to sample the range of SWI at a specific point as in the previous section, the simulated values for all points in the control simulation of a particular vegetation class have been combined to attempt to populate the full range of SWI. The Northern Hemisphere alone is examined, because July is the active season for mid- and high-latitude vegetation and to avoid inclusion of dormant deciduous vegetation that does not transpire. Plotted on the scatter is the piecewise best fit calculated by binning the values over SWI intervals of 0.05 and connecting the mean values for each bin.
Figure 5 shows the distribution of EF versus SWI for all forested grid boxes in the Northern Hemisphere averaged for July. Included are all broadleaf and coniferous forest points, both deciduous and evergreen (vegetation types 1–5 in the ISLSCP Initiative I dataset; black areas in Fig 1). Also shown are the mean of each SWI bin of width 0.05, and the best-fit function of (8) using the coefficients from Table 2. BATS shows a rapid increase of EF over the range of low SWI, with EF leveling off at a value above 0.8. SSiB has a similar rapid increase of EF but reaches a much lower plateau at about 0.56. Mosaic levels off at a value of EF near 0.66, between the other two schemes. For Mosaic, the increase of EF with SWI at the low end of the range is less steep than for the other two schemes. At high values of SWI, all schemes show a similar lower threshold for EF, with few points below EF equal to 0.3. SSiB has few points where EF exceeds 0.9, Mosaic has a few dozen, and BATS has hundreds of points with very high values of EF. It also is evident that most of the forest points in SSiB maintain high soil wetness. This fact may be a consequence of the low limits on transpiration in this scheme as compared with the other schemes, which also is suggested by Fig 4. Mosaic shows the least scatter about the curve of best fit. In fact, for each vegetation type, Mosaic shows the highest correlation of any scheme between the best-fit curve and the individual points (see Table 2).
Figure 6 shows the same distribution as does Fig. 5, but for shrub and grassland vegetation types (types 6–9, 14, and 15 in ISLSCP Initiative I). All three schemes show a much more gradual increase of EF with increasing SWI [represented by lower values of α in (6), see Table 2]. The best-fit curve for BATS lies below the majority of points near SWI = 1; this result likely is due to the effect of the cluster of points with low EF and high SWI on the least squares calculation. Otherwise, it appears that BATS would again have the highest ceiling of EF of the three schemes and SSiB would have the lowest. There is a suggestion of a bimodal distribution in all of the schemes, with points clustering at either tail. SSiB has the most gradual increase of EF with SWI. Correlations are higher than were found for the forest grid boxes.
Tundra grid boxes (ISLSCP type 10) are plotted in Fig. 7. Few of these high-latitude points exhibit very low values of SWI. Mosaic shows a uniform distribution, and SSiB and BATS show points clustering toward high values of SWI. Again, SSiB has the most gradual slope of the arctangent curve. BATS again shows a tendency toward the highest values of EF, as compared with the other schemes. Correlations generally are poor yet still significant at the 99% confidence level.
Bare soil grid points (ISLSCP type 11) are displayed in Fig. 8. For each model, the maximum values of EF suggested are among the lowest of all vegetation types. These bare soil points usually exhibit low values of SWI because they are mostly warm desert grid boxes. Mosaic shows the strongest clustering toward the dry end of the soil wetness scale. The scatter of points for BATS suggests that there may in fact be two different relationships between EF and SWI at play: one where EF is highly restricted throughout nearly the entire range of soil wetness, and another with a more familiar distribution, reaching a plateau at about EF equal to 0.4. The best fit for the SSiB scheme is actually a linear trace with a slope of 0.369.
Again, the binned values and the fitted curves are only approximations to the intrinsic EF–SWI relationships in the models; the wider the scatter of points around the curves, the less reliable are these approximations. The shapes of the curves nevertheless should reflect the true relationships, at least qualitatively. They thus can be used to make some generalizations about how the sensitivity of EF to soil moisture (i.e., the value of δ) varies with SWI, with vegetation type, and with land surface scheme. Note that, from (7), δ varies with EF and with the inverse of the slope of the EF–SWI relationship. Because EF tends to be small when the slope is large and vice versa, a higher slope implies a lower value of δ and thus a higher sensitivity.
The curves generally show a high slope for the lowest values of SWI and a leveling off of this slope (sometimes to near 0) at high values of SWI. The implication is that δ is much smaller for dry conditions than for wet conditions; when conditions are drier, soil moisture needs to be determined more accurately to attain a given precision in evaporative fraction. In fact, for those curves that show a complete flattening out at high SWI (e.g., the forest curves), δ at high SWI essentially is high enough to negate the importance of measuring soil moisture accurately.
The nature of the “leveling off” at high SWI clearly differs between vegetation types. Whereas the forest curves suggest a complete lack of sensitivity at high SWI, the grassland curves for SSiB and perhaps BATS suggest a large sensitivity (relatively low value of δ) at all SWI values. For bare soil evaporation, all of the models show considerable sensitivity for all soil moistures.
Intermodel differences in the curves for a single vegetation type are more difficult to characterize, given the proximity of the curves to each other and the scatter in each plot. Still, for forest, the curve for Mosaic suggests that it maintains large sensitivity over a larger portion of the SWI range. SSiB maintains the highest sensitivity (lowest δ values) at high SWI for tundra; and, for bare soil, SSiB has a higher sensitivity at high SWI than does Mosaic.
The curves in these figures also can be considered in relation to the spatial distribution of sensitivity found in Fig. 3. Although the SSiB scheme has inherently more sensitivity than the other schemes outside of forested areas, its tendency toward wetter soils (Fig. 4) offsets the increased sensitivity. The net result is that SSiB shows fewer areas of high sensitivity in the Northern Hemisphere in Fig. 3. Mosaic’s tendency toward dry soil conditions results in expansive areas of high sensitivity. This result is particularly true because with Mosaic, even “vegetated” grid cells can include a large bare soil component, and Fig. 8 suggests that Mosaic’s δ for bare soil is low over much of the SWI range. BATS generally falls between the other two schemes in terms of soil wetness and thus in the spatial extent of highly sensitive grid cells.
6. Conclusions
Three land surface schemes (BATS, Mosaic, and SSiB) were integrated on a global 1° × 1° grid, driven with meteorological forcing derived from observations during July 1987 and 1988. The relationship between surface evaporative fraction and soil moisture was examined to determine the accuracy to which soil moisture index (and by extension, soil moisture itself) must be known in order to simulate evaporative fraction to within observational uncertainty (about 10%).
All schemes exhibit similar relationships between evaporative fraction and soil moisture index and similar variations of these patterns as a function of vegetation cover. The sensitivity of surface fluxes to variations in soil moisture generally is concentrated at the dry end of the range of soil moisture index near the wilting point. This result may be because these areas are experiencing soil water stress, which easily is increased or relieved by changes in SWI. In humid regions, stress is small or absent, so the response to fluctuations in SWI also is weak.
Forested regions showed the smallest ranges of soil moisture over which sensitivity was appreciable. The vegetation allows, via root uptake and transpiration of water, much higher values of EF in dry soil conditions than could be expected by soil evaporation alone. Thus, the forest canopy can sustain a very constant EF across a wide range of soil wetness. Where vegetation is sparse, EF varies across a larger range of SWI. Shallow-rooted grasses and short vegetation are not as capable as forests of accessing water in dry conditions and similarly are not as capable of maximizing EF when soils become wet. When vegetation is absent, there is no direct mechanism for moving subsurface soil moisture to the atmosphere. This lack, combined with the strong radiative forcing and low soil moisture usually found in these predominantly desert areas, severely limits the magnitude of EF in bare soil. Arid regions do, however, show the greatest range of sensitivity of evaporative fraction to variations in soil wetness.
There are differences among the schemes, and they are manifest in differences in the threshold between sensitive and insensitive regimes. These lie at different points along the transition from arid to humid climates for the three schemes, suggesting variability among models in the range of soil wetness over which sensitivity of surface fluxes is high. Also, differences are evident in the maximum level of evaporative fraction allowed by the models when soil water stress is absent. These two factors can explain most of the differences in the spatial distribution of surface fluxes and soil moisture index seen in the experiment. Both of these parameters are potentially tunable and could be based on laboratory or field data to improve the simulation of surface fluxes. The differences among LSSs seen in Figs. 5–8 may be related to the different moisture stress functions in the models, since the distributions themselves are analogous to the so-called beta distributions in simple bucket models.
The finding that sensitivity of evaporative fraction to soil moisture variations is confined mainly to regions of sparse vegetation or dry soils has implications, if it is robust, for observational networks. For purposes of simulating surface fluxes, it may not be necessary to have accurate soil wetness data in most forested regions or in areas where soils do not dry out. In particular, the possibility that soil moisture under forests is not important offsets a major shortcoming of soil moisture remote sensing. Methods exist to estimate surface soil wetness from both infrared and microwave brightness temperatures in various channels. All of these methods suffer from an inability to estimate soil moisture conditions under moderate or dense vegetation. This inability is caused by the fact that the vegetation canopy obscures the view of the surface (in the case of infrared methods) or contaminates the signal with the water held in the plants themselves (in the case of microwave-based methods). If it can be verified that surface heat fluxes over canopies are not sensitive to the wetness of the underlying soil, then the locations where remote sensing methods work best actually would correspond to the areas where they are most valuable. Similarly, ground-based observing systems could be concentrated in the drier, less-vegetated regions, conserving resources and manpower and making an operational soil moisture monitoring network more feasible.
Acknowledgments
We thank Pavel Kabat and Han Dolman for motivating this investigation. We also thank the members of GEWEX and IGBP-BAHC who have shown great interest in this work.
REFERENCES
Anderson, E. A., 1976: A Point Energy and Mass Balance Model of Snow Cover. Office of Hydrology/National Weather Service, 150 pp.
Chen, T. H., and Coauthors, 1997: Cabauw experimental results from the Project for Intercomparison of Land-Surface Parameterization Schemes. J. Climate,10, 1172–1193.
Dickenson, R. E., 1988: The force–restore model for surface temperatures and its generalizations. J. Climate,1, 1086–1097.
——, and Z. L. Yang, 1998: Technical note of a 10-layer soil moisture and temperature model. Tech. Note, Department of Atmospheric Sciences, The University of Arizona, Tucson, Arizona, 24 pp.
——, A. Henderson-Sellers, P. J. Kennedy, and M. F. Wilson, 1986:Biosphere–Atmosphere Transfer Scheme (BATS) for the NCAR Community Climate Model. NCAR Tech. Note, TN275+STR, 69 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307.].
——, ——, and ——, 1993: Biosphere–Atmosphere Transfer Scheme (BATS) version 1E as coupled to the NCAR Community Climate Model. Tech. Note NCAR/TN-387+STR, 72 pp. [Available from NCAR, P.O. Box 3000, Boulder, CO 80307.].
——, M. Shaikh, R. Bryant, and L. Graumlich, 1998: Interactive canopies in a climate model. J. Climate,11, 2823–2836.
Dirmeyer, P. A., and F. J. Zeng, 1997: A two-dimensional implementation of the Simple Biosphere (SiB) model. COLA Rep. 48, Center for Ocean–Land–Atmosphere Studies, Calverton, MD, 30 pp.
——, A. J. Dolman, and N. Sato, 1999: The Global Soil Wetness Project: A pilot project for global land surface modeling and validation. Bull. Amer. Meteor. Soc.,80, 851–878.
Gao, X., S. Sorooshian, and H. V. Gupta, 1996: Sensitivity analysis of the Biosphere–Atmosphere Transfer Scheme. J. Geophys. Res.,101, 7279–7289.
Gash, J. H. C., and Coauthors, 1997: The variability of evaporation during the HAPEX—Sahel intensive observing period. J. Hydrol.,188–189, 385–399.
Hahmann, A. N., and R. E. Dickinson, 1997: RCCM2/BATS model over tropical South America: Applications to tropical deforestation. J. Climate,10, 1944–1964.
Hall, F. G., and P. J. Sellers, 1995: First International Satellite Land Surface Climatology Project (ISLSCP) Field Experiment (FIFE). J. Geophys Res.,100, 25 383–25 396.
International GEWEX Project Office, 1995: Global Soil Wetness Project, 47 pp. [Available from IGPO, 1010 Wayne Avenue, Suite 450, Silver Spring, MD 20910.].
——, 1998: Land surface parameterizations/soil vegetation atmosphere transfer schemes workshop: Conclusions and working group reports. IGPO Publication Series 31, 77 pp. [Available from IGPO, 1010 Wayne Avenue, Suite 450, Silver Spring, MD 20910.].
Jin, J. M., X. Gao, Z.-L. Yang, R. C. Bales, S. Sorooshian, R. E. Dickinson, S.-F. Sun, and G.-X. Wu, 1999: Comparative analyses of physically based snowmelt models for climate simulations. J. Climate,12, 2643–2657.
Koster, R. D., and M. J. Suarez, 1992: Modeling the land surface boundary in climate models as a composite of independent vegetation stands. J. Geophys. Res.,97, 2697–2716.
——, and ——, 1996: The influence of land surface moisture retention on precipitation statistics. J. Climate,9, 2551–2567.
——, and P. C. D. Milly, 1997: The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate,10, 1578–1591.
Louis, J., M. Tiedke, and J. Geleyn, 1982: A short history of the PBL parameterization at ECMWF. Proc. ECMWF Workshop on Planetary Boundary Layer Parameterization, Reading, United Kingdom, ECMWF, 59–80.
Meeson, B. W., F. E. Corprew, J. M. P. McManus, D. M. Myers, J. W. Closs, K. J. Sun, D. J. Sunday, and P. J. Sellers, 1995: ISLSCP Initiative I—Global data sets for land-atmosphere models, 1987–1988. Vols. 1–5, NASA, CD-ROM. [Available from Code 902.2, NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
Sato, N., P. J. Sellers, D. A. Randall, E. K. Schneider, J. Shukla, J. L. Kinter III, Y.-T. Hou, and E. Albertazzi, 1989: Effects of implementing the Simple Biosphere model in a general circulation model. J. Atmos. Sci.,46, 2757–2782.
Sellers, P. J., 1985: Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens.,6, 1335–1372.
——, and F. G. Hall, 1992: FIFE in 1992: Results, scientific gains, and future research directions. J. Geophys. Res.,97, 19 091–19 109.
——, Y. Mintz, Y. C. Sud, and A. Dalcher, 1986: A simple biosphere model (SiB) for use within general circulation models. J. Atmos. Sci.,43, 505–531.
——, and Coauthors, 1996: The ISLSCP Initiative I global datasets:Surface boundary conditions and atmospheric forcings for land–atmosphere studies. Bull. Amer. Meteor. Soc.,77, 1987–2005.
Seth, A., F. Giorgi, and R. E. Dickinson, 1994: Simulating fluxes from heteorogenous land surfaces: Explicit subgrid method employing the Biosphere–Atmosphere Transfer Scheme (BATS). J. Geophys. Res.,99, 18 561–18 667.
Shukla, J., and Y. Mintz, 1982: Influence of land-surface evapotranspiration on the earth’s climate. Science,215, 1498–1501.
Sud, Y. C., P. J. Sellers, Y. Mintz, M. D. Chou, G. K. Walker, and W. E. Smith, 1990: Influence of the biosphere on the global circulation and hydrological cycle—A GCM simulation experiment. Agric. For. Meteor.,52, 133–180.
Timball, B., and A. Henderson-Sellers, 1998: Intercomparisons of land-surface parameterizations coupled to a limited area forecast model. Global Planet. Change,19, 247–260.
Wiscombe, W. J., and S. G. Warren, 1980: A model for the spectral albedo of snow. J. Atmos. Sci.,37, 2712–2733.
Wood, E., and Coauthors, 1998: The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS) Phase2(c) Red-Arkansas River basin experiment: 1. Experiment description and summary intercomparisons. Global Planet. Change,19, 115–135.
Xue, Y., P. J. Sellers, J. L. Kinter, and J. Shukla, 1991: A simplified biosphere model for global climate studies. J. Climate,4, 345–364.
——, F. J. Zeng, and C. A. Schlosser, 1996a: SSiB and its sensitivity to soil properties—A case study using HAPEX–MOBILHY data. Global Planet. Change,13, 183–194.
——, H. G. Bastable, P. A. Dirmeyer, and P. J. Sellers, 1996b: Sensitivity of simulated surface fluxes to changes in land surface parameterizations. J. Appl. Meteor.,35, 386–400.
Yang, Z.-L., and R. E. Dickinson, 1996: Description of the Biosphere–Atmosphere Transfer Scheme (BATS) for the soil moisture workshop and evaluation of its performance. Global Planet. Change,13, 117–134.
——, ——, A. Robock, and K. Y. Vinnikov, 1997: On validation of the snow submodel of the Biosphere–Atmosphere Transfer Scheme with Russian snow cover and meteorological observational data. J. Climate,10, 353–373.
——, ——, W. J. Shuttleworth, and M. Shaikh, 1998: Treatment of soil, vegetation and snow in land-surface models: A test of the Biosphere–Atmosphere Transfer Scheme with the HAPEX–MOBILHY, ABRACOS, and Russian data. J. Hydrol.,212–213, 109–127.
Distribution of the vegetation classes used in GSWP: black = forests (ISLSCP types 1–5), gray = shrubs and grasslands (types 6–9, 14–15), hatched = agricultural (type 12), cross-hatched = tundra (type 10), blank = bare soil (type 11), ice, or ocean.
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
Jul surface soil wetness index (top row), EF (middle row), and latent heat flux (bottom row) for the three LSSs: BATS (left column), Mosaic (center column), and SSiB (right column).
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
Spatial distribution during Jul for each LSS of the sensitivity parameter d that represents the change in SWI necessary to cause a change in EF of 10%. Black areas are the most sensitive (<0.05 perturbation in SWI), gray areas are fairly sensitive (0.1 > d ≥ 0.05), hatched areas are less sensitive (0.5 > d ≥ 0.1), and blank areas are insensitive (d ≥ 0.5).
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
Frequency distribution of Jul SWI for the three LSSs among all ice-free land grid cells globally. The vertical line denotes the median value. The scale is percentage of all grid cells tallied.
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
Scatterplot of EF vs SWI for all forested land grid cells in the Northern Hemisphere during Jul of the control integrations of each LSS. Also plotted is the mean of each SWI bin of width 0.05 (X), and the best-fit function of (8) (thick curve).
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
As in Fig. 5, but for grass and shrub grid cells.
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
As in Fig. 5, but for tundra grid cells.
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
As in Fig. 5, but for bare soil grid cells.
Citation: Journal of Hydrometeorology 1, 2; 10.1175/1525-7541(2000)001<0121:TSOSFT>2.0.CO;2
Description and comparison of elements of the three land surface schemes.
(Continued)
For each scheme and vegetation category, the best–least squares fit values of the coefficients α and M from (6), the value of F at SWI = 1, and the correlation r between the set of points and the best-fit curves as illustrated in Figs. 5–8. All correlations are significant at the 99% confidence level. The * indicates that a linear fit actually was best for this case.
