1. Introduction
Land surface models (LSMs) simulate the exchange of energy, moisture, and momentum between the land surface and the atmosphere. Many LSMs have been developed for use in climate and weather prediction models. While there have been efforts to explore the relationship between the way the land surface is parameterized and climate model performance (Sato et al. 1989; Crossley et al. 2000; Desborough et al. 2001), the coupling of the land surface to the atmosphere leads to difficulties in evaluating the contribution of the LSMs separated from the response of the atmospheric model. This has led to a major effort to explore the strengths and weaknesses of LSMs uncoupled (in stand-alone mode) from the atmosphere.
The land surface is responsible for partitioning available water (from rainfall, snowfall, and snowmelt) between runoff and evaporation. Studies by, for example, Wetzel et al. (1996), Koster and Milly (1997), and Gedney et al. (2000), have explored the significance of the hydrological components of LSMs and concluded that they play an important role in the contribution by the land surface to current uncertainty in how to represent the surface. The land surface is also responsible for partitioning available energy between sensible and latent heat fluxes. This balance directly affects the local climate since larger sensible heat fluxes tend to warm the atmosphere and increase the depth of the planetary boundary layer, while more latent heat flux tends to increase the specific humidity and the likelihood of precipitation (Betts et al. 1996). The recognition that the surface energy balance (SEB) is central to simulating land–atmosphere exchanges has led to considerable development in this component (see discussion by Sellers et al. 1997). Similarly, considerable development of the hydrological component has occurred (e.g., Wood et al. 1992; Liang et al. 1994; Koster et al. 2000; Ducharne et al. 2000).
To evaluate, compare, and help improve LSMs, a series of model intercomparisons have been coordinated over the last decade. The Project for the Intercomparison of Land-surface Parameterization Schemes (PILPS; Henderson-Sellers et al. 1995) has identified and attempted to explain the differences between the partitioning of available water and available energy between various LSMs. PILPS carefully designed a series of offline experiments incorporating a large number of models and using common atmospheric forcing and land surface parameters. A wide range of results were produced (Pitman et al. 1999; Chen et al. 1997; Shao and Henderson-Sellers 1996; Schlosser et al. 2000), but with key exceptions (e.g., Koster and Milly 1997; Liang et al. 1998; Slater et al. 2001; Nijssen et al. 2003) relatively little progress was made on identifying the precise causes of differences between the LSMs. Similar difficulties were found in other major intercomparison exercises including the Global Soil Wetness Project (GSWP; Dirmeyer et al. 1999). This lack of progress was understandable given the variability in parameterizations and effective parameter values used in LSMs, coupled to inevitable limitations in available observational data. In effect, there were commonly too many degrees of freedom in the experimental design to constrain the results.
In an effort to understand differences between models in PILPS, and to provide a more controlled environment within which specific sources of LSM variability could be isolated, Desborough (1999) designed the Chameleon land surface model (CHASM). CHASM (see section 2 and the appendix) examines variations in the parameterization of the SEB, keeping all other aspects of the land surface, in particular the hydrological parameterizations, constant. CHASM has proved useful in identifying whether the variations in the parameterization of the SEB included in LSMs explain variations in the results in PILPS (Desborough 1999; Pitman et al. 2003) and the Atmospheric Modeling Intercomparison Project (AMIP-II; Pitman et al. 2004). It is important to recognize the limitations of CHASM. It is effectively a hierarchy of LSMs where only the formulation of the pathways, and mechanisms for the evaporation of liquid water, change as complexity increases. These changes in formulation mirror how other LSMs parameterize the evaporation of liquid water. Thus CHASM is, within a single modeling framework, a surrogate for many models used within the community, but only in terms of the parameterization of the complexity of some key components of the SEB. If a suite of models is run within PILPS or GSWP and a range of results is obtained, CHASM can be used to identify how much of the differences can be associated with SEB complexity. CHASM “partitions” the causes of intermodel differences into SEB complexity as distinct from differences caused by effective parameter differences and how hydrology, soil temperature, etc. are parameterized. CHASM is therefore a diagnostic tool that provides insight to direct future analyses toward specific areas within models that contain the causes of differences between models.
This paper reports on the application of CHASM within the Rhône-Aggregation Intercomparison Project (Rhône-AGG), a Global Energy and Water Cycle Experiment (GEWEX) Global Land–Atmosphere System Study initiative (GLASS; http://hydro.iis.u-tokyo.ac.jp/GLASS/). The Rhône is the largest European river discharging into the Mediterranean Sea, draining over 86 000 km2 of southeastern France. Boone et al. (2004) describe the Rhône-AGG experiment in detail and provide results from those LSMs that participated. Rhône-AGG was designed to increase the understanding of LSMs and to explore reasons for differences in model simulations. Rhône-AGG most resembles PILPS Phase 2c (Wood et al. 1998) and PILPS Phase 2e (Bowling et al. 2003; Nijssen et al. 2003) in that the observed river discharge at a basin scale (and snow depth observations at the local scale) were used to evaluate the LSMs and in that many individual grid elements are simulated, rather than a single point (PILPS Phase 1c, e.g., see Chen et al. 1997). However, the Rhône-AGG experiment differs from earlier intercomparison projects in several ways. First, it adopts a much higher spatial resolution for the atmospheric forcing and surface parameters. Second, there is a large within-basin range in vegetation types. Third, the Rhône Basin accommodates a broad range of climatic types (from Mediterranean to alpine) as well as a large grid-box average altitude gradient (3000 m over a horizontal distance of approximately 300 km). Fourth, the higher net radiation available over the Rhône basin, in comparison to the Torne–Kalix Basin used in PILPS 2(e) (located in Scandinavia), provides a different test of the significance of the SEB parameterization in LSMs. Finally, Rhône-AGG explored various methods of spatial aggregation (an issue not explored in this paper). The low net radiation environment of PILPS 2(e) might reasonably be expected to minimize the influence of the SEB on LSM simulations and emphasize the role of hydrology. Indeed, when CHASM was used to explore the influence of the SEB in PILPS 2(e), it was found that the addition of a constant surface resistance into the simplest mode of CHASM (SIMP, see section 2) led to a significant increase in runoff as a result of the reduced latent heat flux. Beyond that, the addition of explicit canopy interception and bare-ground evaporation proved insignificant. In the final most sophisticated mode (SLAM), the addition of a spatially and temporally variable surface resistance increased the spatial variability of the runoff results. We suspect that the higher net radiation available over the Rhône Basin might lead to a stronger influence of the SEB on model performance.
This paper therefore explores the role of the SEB in explaining results from experiment 1 of Rhône-AGG (Boone et al. 2004). Boone et al. (2004) showed that there was significant intermodel scatter in the partitioning of surface energy fluxes and surface hydrology, a result typical of other intercomparison studies. By exploring the results from CHASM and comparing them against models participating in Rhône-AGG, it should be possible to conclude whether or not the intermodel scatter identified by Boone et al. (2004) can be attributed to variations in the complexity of the SEB parameterization, or whether an explanation lies within other components of LSMs.
This paper is divided into four sections. Section 2 presents an overview of CHASM. The Rhône-AGG Project, including the methodology, is provided in section 3. The results are presented in section 4 and the discussion and conclusions are in the final section.
2. CHASM
In PILPS, and in other intercomparison exercises, the choice of effective parameter values affects the results since the appropriate value for a particular parameter is model dependent (Chen et al. 1997; Desborough 1999). Desborough (1999) developed CHASM to provide a more controlled modeling environment within which differences between simulations caused by differences in the pathways and parameterization of liquid water evaporation might be isolated from differences caused through intermodel parameter variations and/or differences in the parameterization of other processes.
CHASM attempts to capture the behavior of various LSMs involved in intercomparison experiments. CHASM focuses on SEB complexity rather than addressing differences between all components of an LSM. CHASM accounts for cover fractions of vegetation, snow, and ground through the use of a “grouped mosaic approach” (Koster and Suarez 1992). A four-layer soil temperature model with a zero-flux boundary condition at the base of the profile is coupled with a Manabe (1969) hydrology model. The hydrological model allows geographical variations in water-holding capacity, but subgrid runoff parameterizations are not included. Snow is represented as a single composite layer (see Slater et al. 2001) with a reasonably sophisticated snow albedo parameterizaton (see appendix). Each tile, depending on the mode, can have up to four evaporation sources: canopy evaporation, transpiration, bare-ground evaporation, and snow sublimation. Again depending on the mode, resistances may be applied to reduce evaporation and transpiration rates. A general description of CHASM is provided in the appendix. The key characteristic of CHASM, the ability to build various levels of SEB complexity on top of the basic soil temperature, soil moisture, and snow (plus model parameters) modules, is discussed next.
a. CHASM's modes
Table 1 lists the different modes of CHASM used in this study in increasing order of complexity. The features that differentiate the modes are described in detail below.
SIMP is the simplest mode of CHASM (Fig. 1a). The aerodynamic resistance to turbulent transport for heat and moisture (ra) is calculated without an atmospheric stability constant (and is thus denoted r*a) to parallel the model used by Koster and Milly (1997). Moisture available for evaporation is only stored in the root zone (Wr) and on the surface as snow (Wn) so there are only two evaporation sources: evaporation from the snow (En) and evaporation from the root zone (Etr) to which ra is applied (Fig. 1a). Here an represents the fraction of the surface covered by snow, and Etr is reduced below the potential rate by an additional moisture availability resistance (βtr). The calculation of Etr and En also involves the density of air (ρa), the surface saturated specific humidity (q*a) (calculated as a function of skin temperature T0), and specific humidity of the air (qa).
Mode RS is the same as SIMP but with a temporally invariant surface resistance (rs) added to the resistance pathway of snow-free evaporation. Also, ra is calculated with an atmospheric stability correction (see Fig. 1b).
Mode RSI builds onto the RS mode by adding an explicit parameterization for canopy interception of precipitation (Fig. 1c). There are therefore three evaporation sources with the inclusion of evaporation of intercepted water (Ec) following the addition of storage of water within the canopy (Wc). The canopy is divided into fractions of wet (awet) and dry (adry) areas that are dependent on the amount of precipitation and evaporation rates.
Mode RSGI adds bare-ground evaporation to the RSI mode. Moisture can be stored at the surface for evaporation up to a maximum of 40 kg m−2. Bare-ground evaporation is affected by moisture availability where a resistance, βg, is included in the evaporation pathway; βg reduces bare-ground evaporation linearly to zero as moisture availability reduces to zero. The RSGI mode is illustrated in Fig. 1d and includes four evaporation fluxes (Etr, Ec, En, and Eg) and their corresponding moisture storage terms (Wr, Wc, Wn, and Wg).
Mode SLAM-1T is the same as RSGI but the surface resistance is temporally and spatially variable based on vegetation type and the Jarvis (1976) model of stomatal conductance. The most complex mode (SLAM) builds on to SLAM-1T by dividing the surface into two tiles with one representing a combination of bare ground and exposed snow and the other reserved for vegetation. The rationale for tiling the land surface is discussed by Koster and Suarez (1992), but basically it is an attempt to include surface heterogeneity in the LSM. The tiles are not necessarily the same size as they are area weighted depending on the individual fractions of the land surface type. A separate SEB is calculated for each tile, which allows for temperature variations across the land atmosphere interface, a feature not present in the less complex modes.
3. The Rhône-Aggregation Intercomparison Project
a. Methodology
In experiment 1 of Rhône-AGG (Boone et al. 2004) the Rhône Basin was divided into 1471 (8 km × 8 km) grid boxes. The atmospheric forcing was provided using the Système d'Analyze Fournissant des Renseignements Atmosphérique à la Neige (SAFRAN) analysis system (Durand et al. 1993). These data consisted of standard screen-level observations at approximately 60 Météo-France weather network sites within the domain, European Centre for Medium-Range Weather Forecasts (ECMWF) analyses, and climatological data for 249 homogeneous climatic zones and total daily precipitation data from over 1500 gauges.
Atmospheric forcing was provided for 4 yr (1985–89) for air temperature at 2 m, wind speed at 10 m, specific humidity at 2 m, downwelling solar radiation, downwelling longwave radiation, liquid and solid precipitation rates, and surface pressure (see Etchevers et al. 2001 for further details). The soil parameters were defined using soil textural properties. The vegetation parameters were defined using a vegetation map (Giordano 1992) and satellite data (Champeaux et al. 2000).
CHASM was run for 3 yr using the atmospheric forcing and surface parameter data prescribed by the Rhône-AGG experiment (Boone et al. 2004). This required interpolation of the atmospheric forcing data from 3- to 1-hourly intervals. The Rhône Project provided data for one extra year (August 1986–July 1987) to be used as a spinup year, and each of the modes of CHASM was run through 10 cycles of spinup on this year.
In CHASM, the surface resistance in the two most complex modes (SLAM-1T and SLAM) is simulated explicitly following Jarvis (1976) but in the intermediate modes (RS, RSI, and RSGI) it must be prescribed (Desborough 1999; Desborough et al. 2001). The values for the surface resistance for modes RS, RSI, and RSGI were obtained through iterative calibration. SLAM was first run for August 1986 to July 1987 and total evaporation for each grid element calculated. Then each intermediate mode that includes a surface resistance was run for the same period, adjusting the prescribed value of the surface resistance until the total evaporation for a given mode matched SLAM (at each grid element) to a tolerance of 5 mm per year. The final values of the surface resistance were then aggregated to a single value, which was then used in subsequent simulations for a given mode. For mode RS, the calibrated surface resistance was 61 s m−1. This increased slightly to 69 s m−1 in mode RSI, but more significantly to 141 s m−1 in RSGI, indicating that LSMs that prescribe a surface resistance need to use a model-specific value that reflects how evaporation components are parameterized. The value of the calibrated surface resistance increases as canopy interception and then bare-soil evaporation are stripped out from the bulk (implicit) parameterization and represented explicitly. This is because when interception is added explicitly (RSI), evaporative fluxes from the interception store are not reduced by a surface resistance (Fig. 1). This therefore requires the surface resistance applied to other sources of evaporation to be increased in compensation. Similarly, when bare-ground evaporation is added (RSGI), this flux is not resisted by the surface resistance (see Fig. 1d), again requiring an increase in the value of the surface resistance that is applied to the remaining sources of evaporation. These changes in the surface resistance are almost certainly model dependent.
The calibration of the surface resistance for the intermediate modes only used the period August 1986 to July 1987. The calibration was performed point-by-point but resulting values were then lumped to a catchment average, reflecting how models of the complexity of RS, RSI, and RSGI would represent this fixed parameter. The calibration of the surface resistance was performed to ensure that the total evaporation for the 12 months (August 1986 to July 1987) was the same as SLAM. Had we not done this, differences shown in the results could simply have reflected uncertainty in the choice of the surface resistance parameter. Because of the calibration however, we omit the August 1986 to July 1987 period from subsequent annually averaged analyses since we have forced SLAM and the intermediate modes to be similar. We note that this similarity is only constrained in the 12-month total evaporation (August 1986 to July 1987). Since the surface resistance can vary between years based on air temperature, incoming solar radiation, moisture availability, etc., the calibration does not preclude large differences between the modes in subsequent years, or at time scales substantially below those of the annual period. Further, since we lump the surface resistance to a single catchment value, the calibration does not preclude large spatial differences between the modes across the catchment. Since we used August 1985 to July 1986 for model spinup, and August 1986 to July 1987 for calibrating the surface resistances, our analysis is largely restricted to August 1987 to July 1988. This is the same period Habets et al. (1999) used in their simulations of the Rhône Basin.
b. Statistical tests
To identify the significance of any differences between CHASM's modes, we use the statistical tests developed by Wigley and Santer (1990) and Santer and Wigley (1990). Significance levels are assessed using the pool-permutation procedure of Preisendorfer and Barnett (1983) to overcome problems arising from nonideal behavior of the data (particularly spatial autocorrelation and uncertain sampling distributions). The selected tests are shown in Table 2 and details are provided by Wigley and Santer (1990).
The specific statistics used were [using Wigley and Santer's (1990) terminology] NT5, a gridpoint-by-gridpoint comparison of time-mean fields at a 5% significance level; T1, a comparison of overall means; NF5, a gridpoint-by-gridpoint comparison of temporal variances at a 5% significance level; SPRET1, a comparison of overall temporal variances; and SPREX1, a comparison of overall spatial variances.
The pool-permutation procedure (PPP) eliminates spatial bias in these statistics by creating its own reference distribution. It compares two space–time datasets in which the respective dimensions are identical and records significant differences at the local level. After the initial control run, it shuffles the dataset grid points and reruns the comparison tests. The shuffling of grid points and retesting is repeated at least 500 times and the results recorded each time to complete the reference distribution. The p value is then produced by progressively comparing each value in the reference distribution (the random result) with the control value (the actual result). Each time the value in the reference distribution is greater than that in the control run the p value (converted to a proportion) is increased.
The statistics T1, SPRET1, and SPREX1 deal with overall means and variances and so have a two-tailed interpretation of the PPP p value for a significant difference. The result can be either close to zero or close to one depending on which of the two datasets being compared has the greater overall mean or variance. However, NT5 and NF5 are comparisons of the results of local gridpoint-by-gridpoint t and F tests and their interpretation is one tailed: the higher the p value the greater the similarity. A low p value means that the proportion of significantly different gridpoint comparisons achieved in the control run was significantly greater than would be expected to occur by chance. Therefore for NT5 and NF5 a p value less than or equal to 0.05 represents a statistically significant difference. A p value greater than or equal to 0.95 represents a statistically significant similarity.
4. Results
a. Domain averaged, annual time scale
Figure 2 shows the annually averaged evaporation, plotted against the annually averaged runoff (both averaged over the Rhône Basin) for each mode of CHASM. Results from each of the LSMs that participated in Rhône-AGG are also plotted (but not identified). The simplest mode (SIMP) simulates the highest evaporation and lowest runoff of the modes of CHASM, while the most complex version (SLAM) simulates the highest runoff and lowest evaporation. Given that the hydrological and snow parameterizations are common to all modes of CHASM, the range of values of runoff and evaporation can only be explained by variations in the SEB complexity. Figure 2 also shows that the range of results from CHASM largely captures the total range of results from other LSMs participating in Rhône-AGG. SIMP lies at one extreme and SLAM at the other, while results from RS, RSI, and RSGI are very similar. SLAM-1T lies between RSGI and SLAM, indicating that half of the difference between RSGI and SLAM is due to an explicit representation of surface resistance (RSGI to SLAM-1T) and half is due to tiling (SLAM-1T to SLAM). The statistical significance of the differences in the global mean averaged over the basin can be tested using statistic T1 (Table 2). In the case of runoff, adding the constant surface resistance (RS) to SIMP generates a statistically significant difference in runoff (Table 3) because it significantly reduces evaporation. While all modes are statistically significantly different from SIMP, once the constant surface resistance is included, further additions in SEB complexity do not cause statistically significant differences in the mean runoff (T1; Table 3). However, in latent heat, SLAM is statistically significantly different from all modes except SLAM-1T. This can be interpreted as demonstrating that tiling does not produce statistically significant differences in the basin-scale mean latent heat flux.
Figure 3 shows a similar result for the sensible and latent heat fluxes. The range in the latent heat flux is fully captured by CHASM with SLAM simulating the least and SIMP simulating the most latent heat of all the Rhône-AGG models. Some of the differences in the latent heat flux shown in Fig. 3 are statistically significant in the mean. Adding a constant surface resistance into SIMP generates mean differences that are statistically significant (T1; Table 3). However, the addition of interception, and the subsequent addition of bare-soil evaporation do not generate statistically significant differences (T1; Table 3). The increase in complexity from RSGI to SLAM (tiling and a time- and spatially variable surface resistance) does cause statistically significant differences in the latent heat flux (T1; Table 3). However, the intermediate step (SLAM-1T) is not statistically significantly different from SLAM, which suggests that neither tiling nor the spatially variable surface resistance alone is sufficient to produce a statistically significant difference, but combined the impact is significant.
Overall, the variations in the partitioning of available energy and water at the annual time scale and at the spatial scale of the Rhône Basin seen in the Rhône-AGG experiments can be explained by variations in the SEB complexity. However, only the addition of a constant surface resistance to mode SIMP (i.e., mode RS), and the combination of tiling and a temporally and spatially variable surface resistance to RSGI (i.e., mode SLAM), generated a statistically significant difference in the mean latent heat flux, and only the change from SIMP to RS led to statistically significant differences in runoff (Table 3).
The simulation of basin-scale snow cover, averaged annually, varies considerably between SLAM and all other modes. SLAM is statistically significantly different from SLAM-1T due to the addition of tiling (T1; Table 3) but other changes in SEB complexity do not lead to statistically significant differences in the mean (Table 3). The importance of tiling in the simulation of snow is not surprising (e.g., Boone et al. 2004).
These findings help confirm earlier results. The need for a surface resistance to limit evaporation has been shown many times before (Milly 1992; Chen et al. 1997) and clearly explains the anomalous results from one of the Rhône-AGG models. A major impact occurs with the addition of this resistance, and incremental increases in the SEB complexity (RSI and RSGI) do not add substantial value. It is therefore unlikely that interception or bare-soil evaporation explains much of the scatter in the basin-scale results reported by Boone et al. (2004). Amongst the Rhône-AGG models, the two areas that are likely to explain the scatter at an annual time scale and averaged over the whole basin are a spatially and temporally variable surface resistance and tiling.
b. Domain averaged, seasonal time scale
Figure 4 shows the seasonal variation in the partitioning of water between evaporation and runoff, and Fig. 5 shows the seasonal variation in the partitioning of energy between sensible and latent heat fluxes. The differences shown between seasons are largely driven by changes in net radiation. In spring, CHASM captures about half of the variation in both evaporation and runoff (Fig. 4) and most of the variation seen among the Rhône-AGG models in the latent and sensible heat fluxes (Fig. 5). There is negligible difference between RS, RSI, and RSGI while SLAM-1T lies equidistant between RSGI and SLAM. Mode SIMP simulates excessive latent heat while the additional resistances in SLAM reduce this flux below that of the intermediate modes. In summer, CHASM captures almost all the variation in evaporation and the latent heat flux seen in the Rhône-AGG results, and with one exception CHASM captures the variability in the sensible heat flux. However, the variability in simulated runoff by the Rhône-AGG models is poorly captured in CHASM. The range of runoff totals for summer in Rhône-AGG is from 32 to 73 mm. SIMP simulates 21 mm, and SLAM 38 mm. Where runoff variability in the Rhône-AGG models is strongly related to water availability and evaporation, CHASM does capture many of the Rhône-AGG model differences. However, CHASM includes a single representation of hydrology, and where the differences between the Rhône-AGG models is strongly controlled by hydrological processes, CHASM should fail to capture these differences. In autumn, SIMP is again anomalous with low runoff, consistent with the excess evaporation in spring and summer. There is little variation in the sensible heat flux and, while SIMP simulates high latent heat, the other modes are generally indistinguishable. Finally, in winter, the modes of CHASM do not capture the variety of simulations of runoff or evaporation from the Rhône-AGG models. About half of the latent heat variability is captured, suggesting that the SEB complexity does contribute to the Rhône-AGG model variability but that other processes, specifically hydrology and snow parameterizations, are central to explaining differences.
It is noteworthy that, in Figs. 4 and 5, the modes of CHASM are ordered sequentially in terms of complexity. SIMP and SLAM are at the extremes of the results from CHASM and the intermediate modes are either indistinguishable, or are ordered in terms of complexity. Thus, varying the SEB complexity in CHASM contributes significantly to the variations in the partitioning of available energy and water, and at least at the Rhône Basin scale appears to explain a large fraction of the total scatter among the Rhône-AGG models annually and in spring, summer, and autumn. Where the relationship breaks down is where non-SEB-related processes dominate the causes of the Rhône-AGG model differences. CHASM does not attempt to incorporate variations in the hydrological complexity that exists within the Rhône-AGG models. Thus, where CHASM captures the differences between the Rhône-AGG models, SEB complexity is the likely explanation for the intermodel variations. Where CHASM fails to capture these differences, processes that do not vary in CHASM (hydrology and snow processes) are the source of the differences.
c. Domain averaged, monthly
Results in this section cover the period September 1986 to August 1988. The first 12 months of this period were used for calibration (see section 3). However, since calibration was on annual evaporation, exploring results substantially below the annual time scale is useful in exploring the role of SEB complexity in contributing to the differences among the Rhône-AGG models.
Figure 6 shows the monthly simulated total runoff from September 1986 through to August 1988. The range of results from the Rhône-AGG models (excluding SLAM, which is shown explicitly) is shown as an envelope to aid clarity. The variation in the monthly runoff shown by the modes of CHASM is approximately 50% of the total range of runoff simulated by the Rhône-AGG models, but most of this is generated by SIMP, which, due to high evaporation throughout the year (Fig. 4), simulates very low runoff in comparison to the other modes (but not anomalous runoff in comparison to Rhône-AGG since SIMP's runoff is within the envelope of the Rhône-AGG models more than 60% of the time). Other than SIMP, it is difficult to differentiate between the other modes of CHASM in terms of monthly runoff.
Figure 7 shows the high evaporation simulated by SIMP. SLAM and SLAM-1T are separated from the intermediate modes in spring where the spatially and temporally varying canopy resistance and tiling reduces the latent heat flux. About half of the differences between RSGI and SLAM can be attributed to each process. There are no systematic differences between the intermediate modes, but a close look at the monthly results for summer shows an ordering of results consistent with Fig. 5. A similar result was obtained for the sensible heat flux (not shown). CHASM's simulated surface temperature and change in soil moisture is within the range of the Rhône-AGG models and the various CHASM modes produce very similar results (not shown).
Figure 8 shows that CHASM's simulation of snow is within the range of Rhône-AGG models. However, there is a major difference between SLAM and all other modes with up to 50% more snow simulated by SLAM. This is not related to accumulation or end-of-season melt since the length of time snow lies on the surface is similar between modes (Fig. 8). The differences in snow between SLAM and the other modes of CHASM are caused by tiling since SLAM-1T is indistinguishable from RSGI. When snow cover is potentially high, the explicit snow tile used in SLAM (Fig. 1) allows snow to accumulate independently of vegetation. The snow has parameters and characteristics explicitly defined. Where tiling is not present the surface parameters across the grid element are defined effectively, and are therefore a blend of snow, soil, and vegetation. This reduces albedo and increases roughness and leads to a tendency to reduce snow accumulation and enhance midwinter sublimation and melt.
Overall, therefore, variations in the SEB complexity in CHASM can explain about half the variability amongst the Rhône-AGG models in the simulation of runoff and the latent heat flux. A large fraction of this explanation is due to the increment in SEB complexity from SIMP to RS. In the case of snow, Fig. 8 indicates that SEB complexity explains little of the variability in Rhône-AGG results, except the addition of tiling. This result needs to be placed within the context of how snow processes are modeled in CHASM. The parameterization of all snow processes is the same across all modes of CHASM. The lack of any simulated differences between the modes in terms of SEB complexity should not be interpreted as suggesting that the parameterization of the SEB of snow is not important. It is simply that the types of SEB complexity variations incorporated into CHASM are not important to the simulation of snow in the Rhône Basin.
d. Domain results, annually averaged
The effect of SEB complexity on CHASM's results can also be explored geographically over the Rhône Basin. The variety of vegetation and soil types, coupled with the variation in orography, suggests that the role of the SEB might vary across the basin. However, it is impractical to compare these basin results to all other Rhône-AGG models; thus here we explore the differences between the intermediate CHASM models in comparison to SLAM.
Figures 9 –11 show the geographical distribution of various quantities across the Rhône Basin as annual averages for the period September 1987 to August 1988. Each figure has the same format, with the first panel [(a)] showing the difference of RS from SIMP. The other panels show differences in the results between a mode of CHASM and the mode immediately preceding it (in terms of complexity). Thus the impact of introducing a constant surface resistance to SIMP mode is seen in (a) (RS–SIMP), of adding explicit canopy interception in (b) (RSI–RS), of bare-ground evaporation in (c) (RSGI–RSI), of introducing the variable surface resistance in (d) (SLAM-1T–RSGI), and of tiling in (e) (SLAM−SLAM-1T). The actual simulation by SLAM is shown in (f).
Figure 9 shows the simulation of the latent heat flux by CHASM. The impact of adding a geographically and temporally constant surface resistance onto SIMP is shown in Fig. 9a. The latent heat flux is reduced across the entire basin by more than 5 W m−2. The smallest differences are in the regions of highest topography (see Boone et al. 2004, their Fig. 1) in the eastern-central region of the basin. Over most of the basin, the latent heat flux is reduced by more than 10 W m−2 and locally by more than 20 W m−2. The addition of explicit interception (mode RSI; Fig. 9b) and bare-soil evaporation (mode RSGI; Fig. 9c) has a negligible effect with changes almost always being less than 5 W m−2, and these changes are not statistically significant in the overall mean (T1; Table 3). The addition of a time- and spatially varying surface resistance has negligible impact (SLAM-1T; Fig. 9d). However, the addition of tiling reduces the latent heat flux by 5–10 W m−2 over regions of higher orography in the eastern half of the domain (SLAM; Fig. 9e). This change is statistically significant in the overall mean (T1; Table 3). Thus the major impact is clearly the initial step of adding a constant surface resistance to mode SIMP. However, adding the tiling and a time- and spatially varying surface resistance also has a statistically significant effect but neither of these additions is statistically significant individually in the global mean (Table 3). While the changes in Fig. 9e do not look that large, the overall impact of a reduction in the annually averaged latent heat flux across the basin represents about 50% of the variation among all but one of the Rhône-AGG models (Fig. 3).
Statistic NT5 tests the point-by-point differences in the time means, and NF5 tests the point-by-point differences in the variance. There are statistically significant differences in the latent heat flux as the SEB complexity is changed from SIMP to RS in both NT5 (Table 4) and NF5 (Table 5). The results from mode RS are statistically similar to those from RSI in both the point-by-point means (NT5; Table 4) and variances (NF5; Table 5). An increase in complexity from RSI to RSGI affects the variance significantly but not the mean. The increment to SLAM-1T leads to a statistically significant difference in the variance (though not in the mean) and the final increment to SLAM does not lead to statistically significant differences in either NF5 or NT5 (Tables 4 and 5). Most changes are statistically insignificant because the number of individual grid elements that show changes are, with the exception of Fig. 9a, too few in number for an overall significant result at the basin scale.
The impact of each step in SEB complexity on the overall temporal variance is not statistically significant (SPRET1; Table 6) in the latent heat flux, except in the step from SIMP to RS. SLAM is statistically significantly different from RSI, RS, and SIMP but not from RSGI or SLAM-1T, indicating that the changes are too small from each individual addition of complexity to produce significant results. However, the increment from RSI to RSGI, and the increment from SLAM-1T to SLAM both produce statistically significant differences in the latent heat flux in terms of the spatial variance (SPREX1; Table 7).
While the parameterization of runoff processes are common to all modes of CHASM, the variations in the latent heat flux affect the amount and timing of runoff through the surface water balance. Figure 10 shows the simulation of runoff by CHASM. The addition of the surface resistance to SIMP increases runoff by 2–10 mm month−1 at higher altitudes and 10–50 mm month−1 over the remainder of the basin (Fig. 10a). The addition of interception (mode RSI; Fig. 10b) and bare-soil evaporation (RSGI; Fig. 10c) affects runoff in a small percentage of grid boxes and is not statistically significant in the overall mean (T1; Table 3). The addition of a time- and spatially explicit canopy resistance has a range of effects across the basin, but these are mostly changes of less than 5 mm month−1 and are not statistically significant in the overall mean (T1; Table 3). Overall, the difference between mode SIMP and SLAM (Fig. 10f) is an increase in runoff that averages 20 mm month−1 (Fig. 2). This is a reduction of over 30% in annually averaged runoff from SIMP and is statistically significant in the overall mean (T1; Table 3).
Tables 4 and 5 show that there are statistically significant similarities between the modes in the simulation of runoff (i.e., the results are more similar than would be expected by chance) in both NT5 and NF5. This is because the point nature of runoff generation is similar across all modes, even if the exact amount changes. Neither the overall time- or spatial-variance statistics (SPRET1 and SPREX1) indicate a statistically significant difference in the results (Tables 6 and 7). This is due to the relatively small fraction of the overall basin that generates significant runoff, coupled with the relatively few months of the year that generate runoff.
Snow, in the Rhône-AGG experiment, is largely restricted to the areas of high orography in the eastern-central region of the basin (Fig. 11f). The impact of increasing the SEB complexity in CHASM from SIMP to RS (Fig. 11a), to RSI (Fig. 11b), to RSGI (Fig. 11c), and to SLAM-1T (Fig. 11d) has negligible (and not statistically significant) impacts on the annually averaged snow water equivalent (T1; Table 3). However, the addition of tiling leads to a large and statistically significant increase (more than 50 kg m−2) in snow (Fig. 11e; T1 in Table 3) in regions of high orography. In SLAM, the land–atmosphere interface is divided into two tiles with one representing a combination of bare ground and exposed snow and the other reserved for vegetation. Thus, instead of snow being spread across a grid element that includes vegetation, it is contained within a separate tile with snow-specific characteristics.
The impact of SEB complexity on snow is statistically insignificant in each additional complexity addition from SIMP to SLAM-1T in all statistics (Tables 3 –7). However, the addition of tiling has a significant impact on both the overall spatial and temporal variances (SPRET1 and SPREX1; Tables 6 and 7) and on the point-by-point time-mean and variance (NT5 and NF5; Tables 4 and 5). The lack of any statistically significant changes in snow as the SEB is made more complex, at least until the final step to SLAM, is clear in Fig. 11 where the actual changes in snow amount are shown to be negligible. Our results therefore implicitly agree with the analysis of Boone et al. (2004) who found that the simulation of snow was most accurate when the snowpack was parameterized explicitly.
5. Discussion
The Rhône-AGG experiment (Boone et al. 2004) explored the ability of a suite of LSMs to simulate energy and water fluxes over the Rhône Basin, a heterogeneous catchment that drains over 86 000 km2 of southeastern France and discharges into the Mediterranean Sea. We used atmospheric forcing, parameter data, and the experimental design of Rhône-AGG (experiment 1) to explore whether varying the complexity of the SEB parameterization could explain the differences between models that took part in the intercomparison. We used CHASM (Desborough 1999), an LSM that was designed to permit an exploration of intermodel differences by removing the contribution to those differences in parameters and differences in the parameterization of processes other than SEB complexity. In effect, CHASM provides a means of identifying the fraction of total intermodel differences that may be explained by differences in the complexity of the SEB parameterization.
At the annual time scale, and at the scale of the Rhône Basin, CHASM captures all the variation displayed by models that were included in Rhône-AGG in the simulation of evaporation, runoff, and sensible heat fluxes. CHASM captures these differences between the Rhône-AGG models by varying the complexity of the key components of the SEB (evaporation pathways and mechanisms). The simulation of runoff, evaporation, and snow mass change as a function of the complexity of the SEB formulation, but in different ways depending on the quantity used as a diagnostic. In the case of the latent heat flux, adding a constant surface resistance into the simplest of CHASM's modes (SIMP) has a statistically significant impact on all statistics listed in Table 2 and shown in Tables 3 –7. The sensitivity of CHASM's results to the inclusion of a constant surface resistance, the similarity in behavior of SIMP and one other Rhône-AGG model (see the single Rhône-AGG model that is positioned similarly to SIMP in Figs. 2 and 3), and the routinely anomalous behavior of SIMP in comparison to all other modes and all but one Rhône-AGG model confirms earlier findings that this mode is too simple to be used as an LSM. Note that SIMP is an implementation of the Manabe (1969) LSM extended to allow variations in water-holding capacity. This is a parameterization that is still (albeit increasingly rarely) used in climate modeling to represent the land surface (see Table 1 of McAvaney et al. 2001).
At the seasonal scale, averaged over the whole basin, the impact of the introduction of a constant surface resistance remains quite clear with SIMP being anomalous, but RS being consistently within the Rhône-AGG scatter. However, in contrast to the annually averaged results, the modes of CHASM do not explain the full range of the Rhône-AGG models at all times (Figs. 4 and 5). As the time scale is reduced, specific mechanisms begin to dominate the simulations produced by each Rhône-AGG model and introduce parameterization-specific behavior that depends on the time evolution of processes operating on longer time scales. CHASM's use of a single parameterization of hydrology means that as non-SEB factors increasingly affect the partitioning of available energy between sensible and latent heat, and changes in this partitioning affect the surface hydrology via the surface water balance, CHASM captures a smaller fraction of the differences between the Rhône-AGG models.
An analysis of the impact of incremental increases in SEB complexity over the Rhône Basin showed different geographical patterns of impact from the SEB. In general, the impacts were restricted to the initial addition of the constant surface resistance and the final increment from SLAM-1T to SLAM (addition of tiling). The simulation of the latent heat flux was strongly sensitive to both these changes in the overall mean (T1; Table 3) and in the overall spatial variance (SPREX1; Table 7). The other statistics were insensitive to the incremental change from SLAM-1T to SLAM, but SLAM was statistically significantly different from modes RS and RSI in the point-by-point time-mean and spatial variances (NT5 and NF5 in Tables 4 and 5, respectively). In general, Tables 3 –7 show that at the basin scale, the runoff simulations were only affected by the SEB complexity at statistically significant levels in the mean. However, this should not be overinterpreted since Fig. 10f shows that significant runoff generation occurs in CHASM over a relatively small fraction of the total basin, and thus measures that sum the point-by-point impact of a change due to SEB complexity may be biased by a large fraction of grid elements where limited runoff is generated. We also emphasize that the use of a simple hydrological model in CHASM may bias our results. While the Manabe (1969) hydrological model works quite well at large spatial scales (Robock et al. 1995) it does not explicitly model the full range of runoff-generating processes. Adding this range of processes into CHASM, to permit an exploration of a fuller range of land surface model complexity, would allow us to assess the magnitude of any bias.
6. Conclusions
We used the Rhône-AGG experimental framework to explore the ability of variations in the SEB formulation to explain the scatter reported by Boone et al. (2004). At the annual scale, and averaged over the Rhône Basin, variations in the complexity of the SEB in CHASM can explain a substantial amount of the intermodel scatter in evaporation, runoff, and the sensible and latent heat fluxes found in Rhône-AGG. As the scale of interest changes from annual to monthly averages over the Rhône Basin, the variations in the SEB complexity explain less of the intermodel differences for two associated reasons. First, as the scale of interest becomes spatially or temporally finer, more mechanisms interact to explain the behavior of a model. Further, the reduction in spatial or temporal smoothing via averaging reveals anomalous model behavior emphasizing the need to evaluate LSMs at high time and spatial resolution where possible.
We conclude that the variations in the SEB complexity can explain a significant fraction of the overall scatter seen in the Rhône-AGG models. CHASM can capture most of the Rhône-AGG scatter by changing only the SEB complexity. For climate modeling applications, all modes of CHASM that include a constant surface resistance perform as well as any of the Rhône-AGG models, or at least, given the quality of observations, it is not possible to demonstrate that these modes perform in an inferior way to the other Rhône-AGG models at coarse time and spatial scales. However, given the results of Desborough et al. (2001) and Milly and Shmakin (2002a, b) we note that a spatially and temporally variable resistance provides a better foundation for simulating the role of the land surface in climate models. Tiling, included in SLAM, is clearly important in the simulation of snow. This would not surprise land surface modelers, but schemes of this level of complexity are not always used in climate models. In the last assessment of climate models for the Intergovernmental Panel on Climate Change, 22% of the climate models used a Manabe (1969) surface scheme and a further 16% used a modified version of the same model. Many of these modified schemes were coupled to multilayer parameterizations of soil temperature (McAvaney et al. 2001).
Our results emphasize the need for LSMs that are balanced in the way they parameterize the surface energy and water balances. To capture the time evolution of the energy and water balance requires the partitioning of available energy at subdiurnal scales. Current LSMs focus on this (Sellers et al. 1997) and likely do it quite well (Chen et al. 1997). However, Boone et al. (2004) showed that many LSMs are relatively weak as the focus increases toward daily and subdiurnal scales. At these scales, the energy and water balances provide an equal contribution, at least in the Rhône-AGG experiments, to intermodel differences. Thus, new LSMs need to focus on coupling these components at similar levels of realism if the uncertainty highlighted in LSM intercomparison experiments such as Rhône-Aggregation and PILPS is to be reduced at all subseasonal and subcurrent climate model spatial scales.
REFERENCES
Betts, A. K., Ball J. H. , Beljaars A. C. M. , Miller M. J. , and Viterbo P. A. , 1996: The land surface–atmosphere interaction: A review based on observational and global modeling perspectives. J. Geophys. Res, 101 , 7209–7225.
Boone, A., and Coauthors, 2004: The Rhône-Aggregation Land Surface Scheme intercomparison project: An overview. J. Climate, 17 , 187–208.
Bowling, L. C., and Coauthors, 2003: Simulation of high-latitude hydrological processes in the Torne-Kalix basin: PILPS Phase 2(e). 1: Experiment description and summary intercomparisons. Global Planet. Change, 38 , 1–30.
Champeaux, J. L., Acros D. , Bazile E. , Giard D. , Gourtorbe J. P. , Habets F. , Noilhan J. , and Roujean J. L. , 2000: AVHRR-derived vegetation mapping over western Europe for use in numerical weather prediction models. Int. J. Remote Sens, 21 , 1183–1189.
Chen, T. H., and Coauthors, 1997: Cabauw experimental results from the project for intercomparison of land-surface parameterization schemes. J. Climate, 10 , 1144–1215.
Cogley, J. G., Pitman A. J. , and Henderson-Sellers A. , 1990: A model of land surface climatology for general circulation models. Trent Tech. Note 90-1, Trent University, Peterborough, ON, Canada, 129 pp.
Crossley, J. F., Polcher J. , Cox P. M. , Gedney N. , and Planton S. , 2000: Uncertainties linked to land surface processes in climate change simulations. Climate Dyn, 16 , 949–961.
Desborough, C. E., 1999: Surface energy balance complexity in GCM land surface models. Climate Dyn, 15 , 389–403.
Desborough, C. E., and Pitman A. J. , 1998: The BASE land surface model. Global Planet. Change, 19 , 3–18.
Desborough, C. E., Pitman A. J. , and McAvaney B. , 2001: Surface energy balance complexity in GCM land surface models. Part II: Coupled simulations. Climate Dyn, 17 , 615–626.
Dirmeyer, P. A., Dolman A. J. , and Sato N. , 1999: The global soil wetness project: A pilot project for global land surface modeling and validation. Bull. Amer. Meteor. Soc, 80 , 851–878.
Ducharne, A., Koster R. D. , Suarez M. J. , Stieglitz M. , and Kumar P. , 2000: A catchment-based approach to modeling land surface processes in a general circulation model. 2. Parameter estimation and model demonstration. J. Geophys. Res, 105 , 24823–24838.
Durand, Y., Brun E. , Mérindol L. , Guyomarc'h G. , Lesaffre B. , and Martin E. , 1993: A meteorological estimation of relevant parameters for snow schemes used with atmospheric models. Ann. Glaciol, 18 , 65–71.
Etchevers, P., Golaz C. , and Habets F. , 2001: Simulation of the water budget and the river flows of the Rhône basin from 1981 to 1994. J. Hydrol, 244 , 60–85.
Gedney, N., Cox P. M. , Douville H. , Polcher J. , and Valdes P. J. , 2000: Characterizing GCM land surface schemes to understand their responses to climate change. J. Climate, 13 , 3066–3079.
Giordano, A., Ed. 1992: CORINE soil erosion risk and important land resources of the European community. EUR Tech. Rep. 13233 EN, 97 pp.
Habets, F., Etchevers R. , Golaz C. , Leblois E. , Ledoux E. , Martin E. , Noilhan J. , and Ottlé C. , 1999: Simulation of the water budget and the river flows of the Rhône basin. J. Geophys. Res, 104 , 31145–31172.
Henderson-Sellers, A., Pitman A. J. , Love P. K. , Irannejad P. , and Chen T. , 1995: The project for Intercomparison of Land Surface Parameterizaton Schemes (PILPS) phases 2 and 3. Bull. Amer. Meteor. Soc, 76 , 489–503.
Jarvis, P. G., 1976: The interpretation of the variations in leaf water potential and stomatal conductance found in canopies in the field. Philos. Trans. Roy. Soc. London, 273B , 593–610.
Koster, R. D., and Suarez M. J. , 1992: Modeling the land surface boundary in climate models as a composite of independent vegetation stands. J. Geophys. Res, 97 , 2697–2715.
Koster, R. D., and Milly P. C. , 1997: The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate, 10 , 1578–1591.
Koster, R. D., Suarez M. J. , Ducharne A. , Stieglitz M. , and Kumar P. , 2000: A catchment-based approach to modeling land surface processes in a general circulation model: 1. Model structure. J. Geophys. Res, 105 , 24809–24822.
Liang, X., Lettenmaier D. P. , Wood E. F. , and Burges S. J. , 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res, 99 , 1415–1428.
Liang, X., and Coauthors, 1998: The Project for Intercomparison of land-surface Parameterization schemes (PILPS) Phase 2(c) Red–Arkansas River basin experiment: 2. Spatial and temporal analysis of energy fluxes. Global Planet. Change, 19 , 137–159.
Manabe, S., 1969: Climate and the ocean circulation. I. The atmospheric circulation and the hydrology of the earth's surface. Mon. Wea. Rev, 97 , 739–805.
McAvaney, B. J., and Coauthors, 2001: Model evaluation. Climate Change 2001: The Scientific Basis, J. T. Houghton et al., Eds., Cambridge University Press, 471–523.
Milly, P. C. D., 1992: Potential evaporation and soil moisture in general circulation models. J. Climate, 5 , 209–226.
Milly, P. C. D., and Shmakin A. B. , 2002a: Global modeling of land water and energy balances. Part I: The Land Dynamics (LaD) model. J. Hydrometeor, 3 , 283–299.
Milly, P. C. D., and Shmakin A. B. , 2002b: Global modeling of land water and energy balances. Part II: Land-characteristic contributions to spatial variability. J. Hydrometeor, 3 , 301–310.
Nijssen, B., and Coauthors, 2003: Simulation of high latitude hydrological processes in the Torne-Kalix basin: PILPS Phase 2(e). 2: Comparison of model results with observations. Global Planet. Change, 38 , 31–53.
Pitman, A. J., and Coauthors, 1999: Key results and implications from phase 1(c) of the Project for Intercomparison of Land-surface Parameterization Schemes. Climate Dyn, 15 , 673–684.
Pitman, A. J., Xia Y. , Leplastrier M. , and Henderson-Sellers A. , 2003: The CHAmeleon Surface Model: Description and use with the PILPS Phase 2(e) forcing data. Global Planet. Change, 38 , 121–135.
Pitman, A. J., McAvaney B. J. , Bagnoud N. , and Cheminat B. , 2004: Are inter-model differences in AMIP-II near surface air temperature means and extremes explained by land surface energy balance complexity? Geophys. Res. Lett, 31 .L05205, doi:10.1029/2003GL019233.
Preisendorfer, R. W., and Barnett T. P. , 1983: Numerical model–reality intercomparison tests using small-sample statistics. J. Atmos. Sci, 40 , 1884–1896.
Robock, A., Vinnikov K. Ya , Schlosser C. A. , Speranskaya N. A. , and Xue Y. , 1995: Use of midlatitude soil moisture and meteorological observations to validate soil moisture simulations with biosphere and bucket models. J. Climate, 8 , 15–35.
Santer, B., and Wigley T. M. L. , 1990: Regional validation of means, variances and spatial patterns in general circulation model control runs. J. Geophys. Res, 95 , 829–850.
Sato, N., Sellers P. J. , Randall D. A. , Schneider E. K. , Shukla J. , Kinter J. L. , Hou Y-T. , and Albertazzi E. , 1989: Effects of implementing the Simple Biosphere Model in a general circulation model. J. Atmos. Sci, 46 , 27–57.
Schlosser, C. A., and Coauthors, 2000: Simulations of a boreal grassland hydrology at Valdai, Russia: PILPS phase 2(d). Mon. Wea. Rev, 128 , 301–321.
Sellers, P. J., and Coauthors, 1997: Modeling the exchanges of energy, water and carbon between continents and the atmosphere. Science, 275 , 502–509.
Shao, Y., and Henderson-Sellers A. , 1996: Modeling soil moisture: A Project for Intercomparison of Land-Surface Parameterisation Schemes, Phase 2(b). J. Geophys. Res, 101 , 7461–7475.
Slater, A. G., and Coauthors, 2001: The representation of snow in land surface schemes: Results from PILPS 2(d). J. Hydrometeor, 2 , 7–25.
Wetzel, P. J., and Coauthors, 1996: Modeling vadose zone liquid water fluxes: Infiltration, runoff, drainage, interflow. Global Planet. Change, 13 , 57–71.
Wigley, T. M. L., and Santer B. , 1990: Statistical comparison of spatial fields in model validation, perturbation, and predictability experiments. J. Geophys. Res, 95 , 851–865.
Wood, E. F., Lettenmaier D. P. , and Zartarian V. G. , 1992: A land-surface hydrology parameterization with subgrid variability for general circulation models. J. Geophys. Res, 97 , D3. 2717–2728.
Wood, E. F., and Coauthors, 1998: The project for intercomparison of land-surface parameterization schemes (PILPS) Phase 2(c) Red–Arkansas River basin experiment: 1. Experimental description and summary intercomparisons. Global Planet Change, 115–135.
APPENDIX
Model Description of CHASM
This appendix provides basic information about CHASM. The model is described in detail by Desborough (1999).
CHASM combines similar elements throughout a grid square to form tiles (a “grouped mosaic approach”; e.g., Koster and Suarez 1992). Each tile is further divided into aerial cover fractions of vegetation, snow, and ground. Snow cover fractions for ground and foliage surfaces are calculated as functions of the snowpack depth and density and the vegetation roughness length, following methods used in earlier LSMs (Cogley et al. 1990; Desborough and Pitman 1998). The vegetation fraction is further divided into wet and dry fractions if the surface configuration mode allows for canopy interception. Each tile has a prognostic bulk temperature (T1) for the storage of energy and a diagnostic skin temperature (T0) for the calculation of surface energy fluxes. There are individual parameters for shortwave albedo and roughness length for each type of cover (e.g., snow, bare ground, and vegetation). The overall surface albedo and roughness length are calculated from an area-weighted approach across the snow, vegetation, and bare-ground fractions. CHASM also includes seasonality parameters for leaf area index and vegetation fraction.
Soil temperature is simulated to a depth of 4 m with the cross section divided into layers: 10–40 cm, 40–100 cm, 1–2 m, and 2–4 m. Energy transfers within the soil are represented using a finite-difference method with constant values for volumetric heat capacity and thermal conductivity and a zero-flux boundary condition at the base of the profile. CHASM utilizes evaporable moisture rather than volumetric soil moisture content to avoid the inclusion of a thermal conductivity reduction for very dry soils.
CHASM's hydrology follows Manabe (1969) in that the root zone is treated as a bucket with finite water-holding capacity [which varies spatially as specified in Boone et al. (2004)] and beyond this capacity runoff occurs. Runoff also occurs if the fraction of snow cover on the ground exceeds 95% and rainfall occurs. In neither case is runoff redistributed. Apart from moisture in the root zone, water can also be stored as snow or depending on the mode, stored on the canopy following interception of precipitation, or on the surface for bare-ground evaporation. While the use of a simple hydrology model may seem outdated, Robock et al. (1995) have shown it to work well in midlatitude regions and the use of this hydrology model has not been associated with poor performance in earlier PILPS experiments. The soil moisture can be liquid or frozen and energy associated with freezing and thawing is accounted for.
The snow scheme assumes that the precipitation falls as snow if the near-surface air temperature is below 0°C. The snowpack is represented by one composite layer (see Slater et al. 2001) and is isolated to the nonvegetated tile in SLAM. The albedo is modified as a function of snow age, and the fractional cover of snow is calculated as a function of the snow depth and density combined with the roughness length of vegetation (Desborough and Pitman 1998). Snow density in CHASM increases over time as a result of mechanical compaction from overlying snow and decreases when new snow falls. The thermal conductivity is represented as a function of snow density (Desborough and Pitman 1998). The available energy to melt snow is computed as residual of the SEB and/or soil heat energy. Any meltwater is added to the soil moisture store or becomes runoff.
Each tile, depending on the mode, can have up to four evaporation sources: canopy evaporation, transpiration, bare-ground evaporation, and snow sublimation. Again depending on the mode, resistances may be applied to reduce evaporation and transpiration rates.
Illustration of the structure of CHASM for each mode discussed in the text. (a) SIMP includes two evaporation sources, evaporation from the root zone (Etr) and the snowpack (En). Two moisture storage terms represent the root zone (Wr) and the snowpack (Wn). The aerodynamic resistance (ra) is calculated without an atmospheric stability constant (r*a). (b) Mode RS where ra is calculated with an atmospheric stability constant and rs is added to the resistance pathway of Etr. (c) Mode RSI adds canopy interception storage (Wc) and the accompanying flux (Ec). (d) Mode RSGI adds a bare-ground parameterization, an extra moisture storage term (Wg), and an extra evaporative source (Eg). (e) Mode SLAM-1T replaces the temporally invariant surface resistance rs, used in simpler modes by a variable canopy resistance rc, which is applied to the evaporation pathway, Etr. (f) Mode SLAM divides the surface into two tiles. Other terms include ρa (air density), At (the size of tile A), q (specific humidity of the surface, q*, and the air, qa); β is a wetness factor and an, awet, and adry are the fractions of snow, wet canopy, and dry canopy, respectively.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Illustration of the structure of CHASM for each mode discussed in the text. (a) SIMP includes two evaporation sources, evaporation from the root zone (Etr) and the snowpack (En). Two moisture storage terms represent the root zone (Wr) and the snowpack (Wn). The aerodynamic resistance (ra) is calculated without an atmospheric stability constant (r*a). (b) Mode RS where ra is calculated with an atmospheric stability constant and rs is added to the resistance pathway of Etr. (c) Mode RSI adds canopy interception storage (Wc) and the accompanying flux (Ec). (d) Mode RSGI adds a bare-ground parameterization, an extra moisture storage term (Wg), and an extra evaporative source (Eg). (e) Mode SLAM-1T replaces the temporally invariant surface resistance rs, used in simpler modes by a variable canopy resistance rc, which is applied to the evaporation pathway, Etr. (f) Mode SLAM divides the surface into two tiles. Other terms include ρa (air density), At (the size of tile A), q (specific humidity of the surface, q*, and the air, qa); β is a wetness factor and an, awet, and adry are the fractions of snow, wet canopy, and dry canopy, respectively.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Illustration of the structure of CHASM for each mode discussed in the text. (a) SIMP includes two evaporation sources, evaporation from the root zone (Etr) and the snowpack (En). Two moisture storage terms represent the root zone (Wr) and the snowpack (Wn). The aerodynamic resistance (ra) is calculated without an atmospheric stability constant (r*a). (b) Mode RS where ra is calculated with an atmospheric stability constant and rs is added to the resistance pathway of Etr. (c) Mode RSI adds canopy interception storage (Wc) and the accompanying flux (Ec). (d) Mode RSGI adds a bare-ground parameterization, an extra moisture storage term (Wg), and an extra evaporative source (Eg). (e) Mode SLAM-1T replaces the temporally invariant surface resistance rs, used in simpler modes by a variable canopy resistance rc, which is applied to the evaporation pathway, Etr. (f) Mode SLAM divides the surface into two tiles. Other terms include ρa (air density), At (the size of tile A), q (specific humidity of the surface, q*, and the air, qa); β is a wetness factor and an, awet, and adry are the fractions of snow, wet canopy, and dry canopy, respectively.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Annually and basin-averaged runoff and evaporation (mm month−1) for each mode of CHASM and each model included in Rhône-AGG (shown as small dots). CHASM modes are represented as an open triangle (SIMP), an open square (RS), an “X” (RSI), a cross (RSGI), an open circle (SLAM-1T), and an open diamond (SLAM). Averages are over the period Sep 1987 to Aug 1988.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Annually and basin-averaged runoff and evaporation (mm month−1) for each mode of CHASM and each model included in Rhône-AGG (shown as small dots). CHASM modes are represented as an open triangle (SIMP), an open square (RS), an “X” (RSI), a cross (RSGI), an open circle (SLAM-1T), and an open diamond (SLAM). Averages are over the period Sep 1987 to Aug 1988.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Annually and basin-averaged runoff and evaporation (mm month−1) for each mode of CHASM and each model included in Rhône-AGG (shown as small dots). CHASM modes are represented as an open triangle (SIMP), an open square (RS), an “X” (RSI), a cross (RSGI), an open circle (SLAM-1T), and an open diamond (SLAM). Averages are over the period Sep 1987 to Aug 1988.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 2 but for sensible and latent heat fluxes (W m−2).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 2 but for sensible and latent heat fluxes (W m−2).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 2 but for sensible and latent heat fluxes (W m−2).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 2 but for summer, winter, spring, and autumn. The ovals show the season each result belongs to, with the arrows clarifying ambiguous points.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 2 but for summer, winter, spring, and autumn. The ovals show the season each result belongs to, with the arrows clarifying ambiguous points.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 2 but for summer, winter, spring, and autumn. The ovals show the season each result belongs to, with the arrows clarifying ambiguous points.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 3 but for summer, winter, spring, and autumn. The ovals show the season each result belongs to, with the arrows clarifying ambiguous points.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 3 but for summer, winter, spring, and autumn. The ovals show the season each result belongs to, with the arrows clarifying ambiguous points.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 3 but for summer, winter, spring, and autumn. The ovals show the season each result belongs to, with the arrows clarifying ambiguous points.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Monthly runoff (mm month−1) from each mode of CHASM (symbols as Fig. 2). The results from Rhône-AGG models are shown as an envelope calculated using the maximum and minimum value across all Rhône-AGG models in each month. Averages are over Sep 1986 to Aug 1988.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Monthly runoff (mm month−1) from each mode of CHASM (symbols as Fig. 2). The results from Rhône-AGG models are shown as an envelope calculated using the maximum and minimum value across all Rhône-AGG models in each month. Averages are over Sep 1986 to Aug 1988.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Monthly runoff (mm month−1) from each mode of CHASM (symbols as Fig. 2). The results from Rhône-AGG models are shown as an envelope calculated using the maximum and minimum value across all Rhône-AGG models in each month. Averages are over Sep 1986 to Aug 1988.
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 6 but for the latent heat flux (W m−2).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 6 but for the latent heat flux (W m−2).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 6 but for the latent heat flux (W m−2).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 6 but for snow water equivalent (mm).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 6 but for snow water equivalent (mm).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 6 but for snow water equivalent (mm).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Basin-scale variation in the annually averaged latent heat flux (W m−2). The simulated change in the latent heat flux for (a) RS–SIMP, (b) RSI–RS, (c) RSGI–RSI, (d) SLAM-1T–RSGI, (e) SLAM–SLAM-1T, and (f) the actual latent heat flux simulated by SLAM. Averages are from Sep 1987 to Aug 1988 (omitting the calibration period).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Basin-scale variation in the annually averaged latent heat flux (W m−2). The simulated change in the latent heat flux for (a) RS–SIMP, (b) RSI–RS, (c) RSGI–RSI, (d) SLAM-1T–RSGI, (e) SLAM–SLAM-1T, and (f) the actual latent heat flux simulated by SLAM. Averages are from Sep 1987 to Aug 1988 (omitting the calibration period).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Basin-scale variation in the annually averaged latent heat flux (W m−2). The simulated change in the latent heat flux for (a) RS–SIMP, (b) RSI–RS, (c) RSGI–RSI, (d) SLAM-1T–RSGI, (e) SLAM–SLAM-1T, and (f) the actual latent heat flux simulated by SLAM. Averages are from Sep 1987 to Aug 1988 (omitting the calibration period).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 9 but for total runoff (mm month−1).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 9 but for total runoff (mm month−1).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 9 but for total runoff (mm month−1).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 9 but for snow water equivalent (mm).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 9 but for snow water equivalent (mm).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
As in Fig. 9 but for snow water equivalent (mm).
Citation: Journal of Hydrometeorology 7, 1; 10.1175/JHM475.1
Summary of the differences between the CHASM modes used in this paper, indicating which of them include explicit parameterizations for canopy interception, bare-ground evaporation, canopy resistance, and horizontal temperature differentiation.
Summary of the statistical tests used in this paper.
T1 statistical significance of adding surface energy balance complexity into CHASM on the simulation of runoff, snow water equivalent, and latent heat flux. Results are based on daily comparisons between modes of CHASM for the three variables for the period Sep 1987–Aug 1988. The statistics shown in bold indicate statistically significant differences at a 5% level of significance.
