EVALUATION OF AMIP SOIL MOISTURE SIMULATIONS

EVALUATION OF AMIP SOIL MOISTURE SIMULATIONS

Alan Robock^1,2, C. Adam Schlosser^1,3, Konstantin Ya. Vinnikov¹,

Nina A. Speranskaya⁴ and Jared K. Entin¹

¹Department of Meteorology, University of Maryland, College Park, Maryland 20742 USA

²Now at: Department of Environmental Sciences
Rutgers, The State University of New Jersey
New Brunswick, New Jersey 08901-8551

³Now at: Center for Ocean-Land-Atmosphere Interactions, Calverton, Maryland 20705 USA

⁴State Hydrological Institute, St. Petersburg 199062 Russia

Global and Planetary Change

1998, Vol. 19, pp. 181-208.

Corresponding Author:

Alan Robock

Department of Environmental Sciences

Rutgers, The State University of New Jersey

New Brunswick, New Jersey 08901-8551

Phone: 732-932-9478

Fax: 732-932-8644

E-mail: robock@envsci.rutgers.edu

ABSTRACT

The Atmospheric Model Intercomparison Project (AMIP) conducted simulations by 30 different atmospheric general circulation models forced by observed sea surface temperatures for the 10-year period, 1979-1988. These models include a variety of different soil moisture parameterizations which influence their simulations of the entire land surface hydrology, including evaporation, soil moisture, and runoff, and their simulations of the energy balance at the surface. Here we compare these parameterizations, and evaluate their simulations of soil moisture by comparing them with actual observations of soil moisture, literally ground truth. We compared model-generated "data sets" and simulations of soil moisture with observations from 150 stations in the former Soviet Union for 1979-1985 and Illinois for 1981-present. The spatial patterns, mean annual cycles, and interannual variations were compared to plant-available soil moisture in the upper 1 m of soil.

The model-generated "data sets" of Mintz and Serafini and Schemm et al. are quite different from the observations. The model-generated "data sets" are quite different from each other in many regions, even though they use the same bucket model calculation method. The AMIP model simulations are also quite different from each other, especially in the tropics. Models with 15-cm field capacities do not capture the large high latitude values of soil moisture. In addition, none of the models properly simulate winter soil moisture variations in high latitudes, keeping soil moisture constant, while observations show that soil moisture varies in the winter as much as in other seasons. The observed interannual variations of soil moisture were not captured by any of the AMIP models. Several models have large soil moisture trends during the first year or two of the AMIP simulations, with potentially large impacts on global hydrological cycle trends and on other climate elements. This is because the simulations were begun without spinning up the soil moisture to the model climatology. The length of time it took for each to reach equilibrium depended on the particular parameterization. Although observed temporal autocorrelation time scales are a few months, some models had much longer time scales than that. In particular, the 3 parameterizations based on the Simple Biosphere model (SiB) had trends in some regions for virtually the entire AMIP simulation period.

Keywords: SOIL, MOISTURE, FIELD CAPACITY, GENERAL CIRCULATION MODELS, CLIMATE, SIMULATION

1. Introduction

The Atmospheric Model Intercomparison Project (AMIP; Gates 1992) conducted simulations by 30 different atmospheric general circulation models (GCMs) forced by observed sea surface temperatures (SSTs) for the 10-year period, 1979-1988. These models include a variety of different soil moisture parameterizations, which influence their simulations of the entire land surface hydrology, including evaporation, soil moisture, and runoff, and their simulations of the energy balance at the surface. The parameterizations each incorporate a variety of different approaches to representing vegetation, evaporation, runoff, vertical structure of the soil, and field capacity. In order to evaluate the effects of these choices on the model simulations, as part of AMIP Diagnostic Subproject 11, we compare the models to each other, and compare their simulations to observations of soil moisture.

As part of the AMIP Standard Output, all modeling groups were asked to provide monthly-average gridded fields of soil moisture. Most groups provided plant-available soil moisture in the top 1 m of soil, the quantity calculated by bucket models, but some provided other quantities from their more complicated schemes, making direct comparisons more difficult. Furthermore, correct simulation of soil moisture depends not only on an accurate soil moisture parameterization, but also on accurate cloudiness and precipitation simulations, which are being addressed in detail by other Diagnostic Subprojects (Weare et al. 1995; Lau et al. 1996). A complete diagnosis of the AMIP hydrological cycles is not possible at this time, as runoff was not included in the Standard Output. Also the monthly averages of soil moisture, snow depth, and precipitable water, which are included in the Standard Output, do not allow evaluation of the water balance as these quantities are reservoirs, and end-of-month values are required. The second AMIP experiment, now in the final planning stages (Gleckler, 1997), will correct these deficiencies and allow a complete evaluation of the water budget.

The land surface parameterizations evaluated in this paper represent the state of the art as of 1994. Many of the GCMs discussed here have modified their land surface schemes in current versions, and we will evaluate them in AMIP II (Gleckler, 1997). Nevertheless, this study serves as an important baseline evaluation and points the direction for future improvements.

The Project for Intercomparison of Land-surface Parameterization Schemes (PILPS - Henderson-Sellers et al., 1993) has completed related intercomparisons of land surface schemes forced either by idealized climate model output (Phase 1; Pitman et al., 1993, 1998; Koster and Milly, 1997) or by actual observations (Phase 2; Henserson-Sellers, 1996a; Chen et al., 1997; Qu et al., 1998). More results from these experiments are discussed in other papers in this special issue. Two related differences between the AMIP and PILPS Phase 2 studies make the AMIP analysis more complex. In the first place, the land surface schemes in AMIP are fully coupled with the GCM atmospheres, so that they are free to seek and feed back on their own model climates. The second related difference is that the surface hydrology is driven by model-generated precipitation and radiation, and GCM precipitation (Lau et al., 1996; Sengupta and Boyle, 1997) and clouds (Weare et al., 1995) were not very accurately simulated. The differences we find between AMIP model output and observed soil moisture may therefore not be due to the diversity of soil moisture schemes alone. Nevertheless, this study is valuable for two reasons. In the first place, some of the errors are so large that it is possible to determine factors in the land surface schemes that are responsible. In the second place, this study serves to document the current state of the art of coupled simulations of soil moisture, against which future results can be measured.

First, we explain why it is important to evaluate soil moisture behavior in climate models. We then show that the climatologically-driven scale of soil moisture variations controls the variations of soil moisture in our observational data set, and thus our data set, based on individual stations, is appropriate for comparison to GCM grid-box scale calculations. Next, our unique collection of actual in situ soil moisture observations is described. It only covers parts of the former Soviet Union and the United States, but represents actual gravimetric or neutron probe observations. We then compare the observations to 2 global widely-used so-called soil moisture "data sets," which are actually model simulations driven with observed monthly-average temperature and precipitation. Next, the different soil moisture parameterization schemes are classified and compared. The model simulations are then evaluated by looking at their spatial and temporal fields as compared to observations. Severe spinup problems are found in some of the simulations, and these are investigated. These results have implications for the design of climate model intercomparison projects like AMIP, and suggestions for improved future studies are discussed.

A shorter preliminary version of this paper was published as Robock et al. (1995b).

2. Why study soil moisture

Atmospheric GCMs include land surface parameterizations in order to more accurately calculate vertical fluxes of moisture, energy, and momentum. The energy fluxes include latent and sensible heat fluxes as well as radiative fluxes. Soil moisture affects latent heat flux, and hence the partitioning of outgoing convective fluxes between sensible and latent heat, with a strong effect on the resulting surface temperature. Soil moisture, in some models, also affects surface albedo, with wetter surfaces being darker, which also affects surface temperature.

Many land surface modelers do not consider the soil moisture portion of their models to be physically based, and think of the soil moisture representation to be more of an index used for evapotranspiration and runoff calculations rather than representative of the actual mass of moisture in the soil. Therefore, they consider comparison of their model calculations with actual soil moisture data to be unfair or not relevant. Koster and Milly (1997) even show that it is the relationship between the dependence of evapotranspiration and runoff on soil moisture that is important, not the absolute amount of soil moisture, when using a simple monthly water balance simulation of complex land surface models.

This point of view is focused only on fluxes to the atmosphere, and ignores the impacts of the actual amount of soil moisture on other systems, especially agriculture. Our philosophy is that land surface models should be correct for the right physical reasons, and that soil moisture evaluation is therefore as important as validation of the other hydrological fluxes. Henderson-Sellers (1996b) presents similar arguments. Previous intercomparisons of land surface models with our Russian data, including Robock et al. (1995a), who evaluated the bucket model (Budyko, 1956; Manabe, 1969) and the Simplified Simple Biosphere model (SSiB; Xue et al., 1991), have led to important improvements in these schemes. For example, Robock et al. (1997) and Schlosser et al. (1997) showed how proper specification of field capacity produces much improved bucket model simulations. Douville et al. (1995a) used our data to develop the Météo-France land surface model, Yang et al. (1997) subsequently used the same data with the BATS model (Dickinson et al., 1986) to evaluate and improve the snow cover parameterization, Slater et al. (1997a, b) have used the data to evaluate snow parameterization and subsurface winter hydrology with their BASE model (Desborough, 1997), and Mocko and Sud (1997) have used the data to improve their version of SSiB. Here we list a number of reasons why soil moisture is important:

Soil moisture affects not only the vertical fluxes of energy and moisture, but also the horizontal fluxes of moisture, namely runoff. While atmospheric GCMs implemented in the AMIP experiment with specified SSTs can ignore runoff, these same models, with the same land surface parameterizations, are also being used in coupled land-ocean-atmosphere GCMs. One of the crucial connections between these three parts of the climate system is the thermohaline circulation of the ocean, which is driven in part by the fresh-water flux from the continents, the runoff. If the land surface parameterization does not produce the correct partitioning, the ocean circulation cannot be simulated correctly. If a 15-cm bucket model, or any other land surface parameterization, over high midlatitudes is always full because its field capacity is too small, it cannot properly simulate the month-to-month variations of runoff, and therefore it is important to evaluate the soil moisture per se, and not just the vertical fluxes.
One of the most dire predictions of global warming is mid-latitude summer drought in the important agricultural regions of the Northern Hemisphere (Manabe et al., 1981; Houghton et al. 1996; Watson et al. 1996). Drought is a soil moisture deficiency. Plants care about the amount of soil moisture, not the fluxes. Therefore, evaluation of the ability of models to correctly simulate soil moisture is crucial to understand how well these models can calculate the impacts of global warming.
Flooding is a soil moisture excess. Enhanced variability of climate has also been suggested as a result of global warming, with more flooding episodes (Houghton et al., 1996). In order to properly simulate this aspect of climate change, the soil moisture calculation must be correct.
Any parameterization of physical quantities should be correct for the right physical reasons. In early versions of models, simple parameterizations may ignore some of the more complex physical interactions. For example, the original bucket model (Budyko, 1956) included partitioning of precipitation into runoff and infiltration, but ignored stomatal resistance to evapotranspiration. As implemented for the first time in a GCM by Manabe (1969), the partitioning was disabled and stomatal resistance was still missing. Some of the more complex parameterizations now include quite detailed treatment of runoff and evapotranspiration. It remains an open question as to whether this complexity results in improved simulations of soil moisture.
Soil moisture, along with snow cover, is also the most important component of meteorological memory for the climate system over the land. Delworth and Manabe (1993) have shown that varying soil moisture (as opposed to fixed soil moisture) induces a redder spectrum in simulated climate by retaining the hydrological signature of anomalies for time scales longer than those of the atmosphere alone. Soil moisture anomalies should also be important for more accurate seasonal forecasts.
Soil moisture variations from month to month are as large as the other terms in the hydrologic balance, namely precipitation, evaporation, and runoff, and correct simulation of these variations is necessary to correctly model the hydrological cycle.

3. Data sets

3.1. Scales of soil moisture variations

As shown in detail by Vinnikov et al. (1996), there are two distinct spatial scales of soil moisture variations. The small scale, familiar to hydrologists, is of order tens of meters. Soil moisture can vary on this scale due to small scale variations of topography, soil type and texture, and vegetation. Superimposed on this small-scale variability is a much larger scale of hundreds of kilometers that is due to the meteorological processes that influence soil moisture and that have this same scale on a monthly average, precipitation and evapotranspiration. It is well known that the complex topography of natural landscapes, with spatially variable vegetation and soil types, and gravitational drainage and infiltration of water after heavy rains, is responsible for very small-scale spatial (tens of meters) and temporal (up to few days) variability in the soil moisture field (Vachaud et al., 1985; Rodriguez-Iturbe et al., 1995). This hydrological component of soil moisture field variability looks like random (white) noise in comparison with the long-term (about 1-4 months) and large-scale (about 400-800 km) signal related to atmospheric forcing. This meteorological component of variability of soil moisture field has been found in observations (Meshcherskaya et al., 1982) and later received theoretical explanation (Delworth and Manabe, 1988, 1993; Vinnikov and Yeserkepova, 1991; Vinnikov et al., 1996). This soil moisture component, driven by atmospheric forcing, may be modeled using routine meteorological observations at regular meteorological stations (Robock et al., 1995a; Yang et al., 1997; Schlosser et al., 1997). Small scale variability of the soil moisture field is unpredictable and appears as a stochastic process in this context.

Figure 1 illustrates this separation of scales schematically. While some of the spatial autocorrelation rapidly decreases with distance, beyond the 10 or 20 m catchment scale the meteorological scale dominates and provides a relationship between neighboring soil moisture stations that allows a large scale soil moisture distribution to be adequately sampled with our existing network of stations (see next section). The soil moisture station networks were set up with this scale in mind. Our recent analysis of the Illinois soil moisture data (Hollinger and Isard, 1994) gives the same scales as found for the Russian data by Vinnikov et al. (1996), and will be reported in a forthcoming paper.

3.2. Observational data sets

There are two data sets of actual measurements of soil moisture available for comparison to GCM simulations, one from the former Soviet Union [hereafter referred to as Russia] (Vinnikov and Yeserkepova, 1991; Vinnikov et al., 1996), and one from Illinois in the United States (Hollinger and Isard, 1994). Several other measurement programs have been conducted at various places in the world, and some of these data sets, including new ones from Russia and ones from China, Mongolia, and India, are now being assembled by the authors, but are not yet complete. Our Global Soil Moisture Data Bank (http://climate.envsci.rutgers.edu/soil_moisture) is dedicated to collection, dissemination, and analysis of soil moisture data from around the globe, and the data used in this paper are available there.

The Russian data are gravimetric measurements taken in 10 cm layers from the top 1 or 1.5 m of soil every 10 days in the warm part of the year and monthly in the winter from 130 stations at grassland sites for 1978-1985. In Illinois, neutron probes, calibrated with gravimetric observations, were used with the same sampling frequency, but with 20 cm vertical resolution, from 17 stations at grassland sites for 1981 to the present. For each data set, we used end-of-month plant-available soil moisture in the top 1 m, for comparison with the AMIP models. The observations are made at the end of each month, so for some comparisons, they are 15 days out of phase with the monthly average output requested from the models. The annual-average data for the Russian region are shown in Fig. 2, and for Illinois are shown in Fig. 3.

For all the Russian stations, the wilting level is measured in the laboratory by measuring the value of soil moisture at which a standard oat crop wilts. Wilting level was not measured for any of the Illinois stations (Hollinger and Isard, 1994) due to its expense. Rather, in Illinois the wilting level was estimated as equal to the average of the 3 lowest values of soil moisture in each layer during the entire measuring period. Since the soil got extremely dry during the summer drought of 1988, it was thought that this would be a good measure.

3.3. Model-generated "data sets"

There are several so-called "data sets" of soil moisture (Mintz and Serafini 1981, 1989, 1992; Schemm et al. 1992; Liston et al., 1993) that have been used for verification, initialization, and surface and subsurface boundary conditions for general circulation models (GCMs). They have the advantage over actual observations that they are global, but they are actually model simulations, forced with monthly-average temperature and precipitation observations. Mintz and Serafini (1981, 1989, 1992) used a 15-cm bucket (Manabe, 1969) with Thornthwaite (1948) evaporation and all precipitation treated as rain (no storage as snow and no snow melt), forced by monthly precipitation climatology (Jaeger, 1976) and monthly temperature climatology (Spangler and Jenne 1984) on a global 4 x 5° latitude-longitude grid. Schemm et al. (1992) used the Mintz and Serafini method forced by monthly average observations from NCAR data for individual months for 1979-1992 on a global 2 x 2.5° grid, so that they produced different "soil moisture" for different years. Liston et al. (1993) used a modified version of SSiB (Xue et al. 1991).

We compared the two bucket-based "data sets" to our observations and to each other. We regridded the Mintz and Serafini output to a 2 x 2.5° grid, and compared it to a 10-year average of Schemm et al. for the 1979-1988 AMIP period (Fig. 4). They are quite different from each other, even though they use the same method. In most of the tropics, western North America, and South and Southeast Asia, Mintz and Serafini is wetter than Schemm et al., by as much as 10 cm in China and northern India. In eastern North America and northern Asia, Schemm et al. is wetter by a few cm. Both are much drier than the observations for most of the area covered in Fig. 2. In northwestern Russia, observations are 15-20 cm higher, a result of the 15-cm field capacity used in the models.

Any GCMs that use these model-based "data sets" as fixed surface or subsurface conditions will therefore be biased toward a particular soil moisture distribution that is different from what is observed. And using these "data sets" to evaluate or design GCMs will similarly be biased. For model initialization, these fields are valuable as they provide a climatology with the same properties as the model, but they do not agree well with observations. In an attempt to improve this situation, as part of the Global Energy and Water Cycle Experiment (GEWEX), the International Satellite Land-Surface Climatology Project (ISLSCP) has implemented a Global Soil Wetness Project (GSWP). For the years 1987 and 1988 using a 1 x 1° resolution for the global land areas, a large number of land-surface modeling groups are calculating soil moisture forced by daily surface observations. By comparing preliminary output with Russian data sets for agricultural regions (Vinnikov et al., 1997) we have found similar large discrepancies between model output and ground truth. GSWP is now being expanded to a 10-year period. As PILPS, GSWP, and AMIP studies continue, and new soil moisture data sets become available, model simulations of soil moisture are bound to improve.

4. Soil moisture schemes

To compare the models and their results, it is useful to try to classify them. Most of the soil moisture schemes in the AMIP GCMs are based on the Manabe (1969) bucket model with 15 cm field capacity, implicit vegetation, and runoff only with bucket overflow. This model is discussed in much more detail and evaluated by Robock et al. (1995a). Three of the schemes are based on the Simple Biosphere (SiB) model of Sellers et al. (1986). In the past, it was convenient to refer to bucket-type models with implicit vegetation as simplistic, and models with explicit consideration of vegetation as sophisticated, complex, and presumably more accurate. Sellers et al. (1997) classify these models as belonging to the first and second of 3 generations of land surface schemes.

The choices made by model developers, however, include many more aspects than vegetation, and the soil moisture schemes used in the AMIP GCMs can be classified in 7 different ways (Table 1) all of which might affect the results of the simulations. 1) calculated (25 models) or prescribed (5 models) soil moisture; 2) initialization from climatology or previous model solution; 3) available or total soil moisture as prognostic variable; 4) field capacity ranging from 12 to 212.5 cm; 5) number of vertical soil levels from 0 to 6, and depth ranging from 0.5 to 10 m; 6) vegetation either implicit, with evaporation calculated as a soil moisture-dependent parameter (b) times the potential evaporation, or explicit, with canopy interception and stomatal resistance (Dickinson et al. 1991); and 7) runoff just from bucket overflow, or also immediate, as fraction of precipitation, or also as subsurface drainage. Shao and Henderson-Sellers (1996) describe these different runoff schemes in more detail for some of the models here. We collected this information for all the models from Phillips (1994) and requests to individual modeling groups, and used it to evaluate the model performances. The variety of soil moisture schemes and the information provided as standard output made the comparison of some results difficult or impossible, without ancillary information, such as the horizontal variation of field capacity.

4.1. Calculated or prescribed soil moisture

Table 2 lists the 30 AMIP models, with their abbreviations. Twenty-five of them explicitly calculated soil moisture, but 5 of them (GSFC, NCAR, NRL, RPN, UCLA) prescribed soil moisture from 5 different soil moisture "data sets," 2 of which are evaluated above. Because none of these data sets is in equilibrium with the energy or water balance of the model atmosphere, these soil moisture fields provide forcing for the atmosphere, just like the prescribed SST patterns. Figure 5 shows the annual average of the prescribed fields for 3 of the models. RPN and UCLA did not provide their soil moisture to AMIP.

4.2. Initialization

Table 3 organizes the models according to their soil moisture initialization choices. Fifteen of the models were initialized with output from previous model simulations (Phillips 1994), so presumably there would not be any soil moisture trends at the beginning of their AMIP simulations. This would only be the case, however, if the runs used to produce the initial soil moisture fields had forcing consistent with the AMIP experiments. Ten of the models were initialized with the climatological soil moisture fields, and so spinup problems might be expected with these. Five models had specified soil moisture, so there was no soil moisture spinup. The model may have had a spinup anyway, to adjust to these fields, but this was not investigated here; the time scale of this adjustment was probably less than the AMIP monthly Standard Output sampling time, anyway.

4.3. Prognostic variable

Table 3 also classifies the soil moisture parameterizations as "bucket," "SiBling," or prescribed. This is shorthand for describing how the parameterization carries soil moisture as a prognostic variable, and is in line with previous use of the word "bucket," but really all soil moisture parameterizations explicitly or implicitly have a field capacity, so all are really buckets (Robock et al. 1995a). By "bucket" here, we mean that the parameterization explicitly considers the depth of plant-available soil moisture (W) in each layer of the soil, as was done with the original 1-layer Manabe (1969) bucket. This is also the quantity that is in the observed data sets. By "SiBling" we mean the 3 models based directly on SiB (GLA, COLA, JMA) and the GISS model, all of which consider soil wetness fraction, which can range from 0 to 1. This is important, because it requires several assumptions to convert wetness to W. Robock et al. (1995a) converted wetness for SSiB (Xue et al. 1991) to W for one particular vegetation type and rooting depth, but the modelers did not, in general, provide W in the top 1 m of soil for each grid point in the AMIP Standard Output. In particular, the wilting level has not been removed. The one exception is the COLA model, for which the wilting level was provided directly to the authors (David Straus and Yongkang Xue, personal communication). For this model, the effect of removing the wilting level on the results is evaluated below.

4.4. Field capacity

The field capacity (Wf), the maximum water-holding capacity of the soil after gravitational drainage has occurred, is an important determinant of runoff and evapotranspiration. Table 3 also includes the field capacity for all the models. For 10 of them, Wf = 15 cm in the top 1 m, the same as the standard Manabe (1969) bucket. For one, Wf = 12 cm, for 6, Wf > 15 cm, for 8 Wf varies from grid point to grid point, and for 5, W is prescribed, and Wf is not given (Phillips, 1994). The 2 models with Wf > 200 cm explicitly consider soil layers with a total depth much greater than 1 m.

For many high latitude regions (Fig. 6), W > 15 cm and Wf > 15 cm in the top 1 m, so the 10 models that restrict Wf to 15 cm cannot realistically simulate the soil moisture variations at these latitudes. Furthermore, for many of these stations, W often exceeds Wf, indicating that the water table depth is less than 1 m. Schemes which fail to explicitly consider this phenomenon may not be able to produce realistic simulations.

4.5. Number of soil layers

Table 4 gives the number of soil layers explicitly considered by the soil models. A value of 0 is assigned to the 5 models with prescribed soil moisture. Thirteen models have 1 layer, but the other 12 models have 2-6 layers, in an attempt to more accurately model the actual soil physics, explicitly accounting for evaporation from a thin surface soil layer, or for different root distributions or water transfer at different depths. The results, however, show no obvious dependence on number of soil layers.

4.6. Vegetation

As shown in Table 5, for 21 of the schemes, vegetation is included implicitly. Evapotranspiration from the top 1 m of soil could not take place without roots to extract moisture from levels beneath a thin layer at the top of the soil. In these schemes, evaporation is calculated as a soil moisture-dependent parameter (

) times the potential evaporation. In 9 of the schemes, vegetation is considered explicitly, including the effects of canopy interception and re-evaporation of precipitation, and stomatal resistance to evapotranspiration. As pointed out by Dickinson et al. (1991), in the absence of accurate atmospheric simulations of precipitation and cloudiness, the inclusion of detailed canopy submodels may not improve the results, and we find no evidence of such an improvement for soil moisture simulations here.

4.7. Runoff

Table 6 shows the choices made for runoff parameterization in the soil moisture schemes. For the 5 schemes with specified W, this choice is irrelevant in its effect on soil moisture. (Since runoff was not one of the variables saved in standard AMIP output, it is not possible to evaluate the effects of these choices on the runoff simulations. Wetzel et al. (1996), in a PILPS Phase 2a experiment, found that all but 1 of 14 models studied significantly underestimated runoff when forced with the HAPEX data from southern France.) For 11 schemes, runoff only occurred when the field capacity of the bucket was exceeded (Manabe, 1969). In 7 of the schemes, a fraction of the precipitation was immediately partitioned into surface runoff, in accordance with the original bucket scheme by Budyko (1956) upon which Manabe (1969) based his model. In the 7 remaining schemes, runoff also occurs from subsurface drainage. Again, we find no evidence here of an improvement in the soil moisture simulations by inclusion of more sophisticated runoff parameterizations.

5. Comparison of spatial fields

We compared the observed spatial fields of soil moisture (Figs. 2-3) with those generated by the AMIP models. All of the Russian and Illinois stations have a grass vegetation cover, while the SiBlings specify different vegetation cover at each grid point according to observations, which may influence the soil moisture. At one location in Russia (Valdai), however, Vinnikov et al. (1996) found that soil moisture was virtually the same at grass-covered and forest-covered catchments, even though evapotranspiration and runoff were different. We would expect vegetation to have a larger impact on soil moisture in the tropics, where evapotranspiration is larger, but without any long-term data sets from the tropics, this is not possible to evaluate in the AMIP context.

Two of the AMIP groups that used prescribed soil moisture (RPN and UCLA) did not provide soil moisture fields to AMIP. The 3 other models with prescribed soil moisture (GSFC, NCAR, NRL) all had different units and different patterns from each other and from the observations (Fig. 5). For the 10 models with standard 15-cm buckets, and CNR (12 cm) and CSIRO (16.2 cm), none captured the observed high values of soil moisture in northern Russia (Figs. 7-8). YONU, IAP, CNR, and DERF are particularly dry in the tropics, and DNM and CSU are saturated over most of Asia. Even when examining the wetness patterns (by dividing soil moisture by field capacity, not shown), only LMD comes close to reproducing the observed spatial pattern over Asia. Three models with multilevel schemes only provided output from surface layers with a small capacity (UGAMP - 2 cm, MRI - 2 cm, and SUNYA/NCAR - 7.5 cm). The wetness patterns showed MRI very wet and SUNYA/NCAR very dry over Asia (Fig. 9). UGAMP (wetness of top layer), MPI (Wf = 20 cm), and ECMWF (Wf = 26 cm) all had similar wetness patterns over Asia and all looked correct for Russia (Fig. 9). They also resembled each other over the rest of the world, except UGAMP was much drier over South America. Thus it seems that, by just increasing the field capacity to 20 or 25 cm, the high latitude patterns can be captured, as previously shown in stand-alone experiments by Schlosser (1995) and Robock et al. (1997).

The 3 models that use variations of SiB (GLA - SSiB (Xue et al., 1992), COLA - SSiB, and JMA - SiB (Sellers et al., 1986)), we designate as SiBlings. GISS is similar in its complexity, and in considering total soil moisture. Since soil moisture values exceed 100 cm for these models in the high latitudes and tropics, representing moisture from various soil depths at different grid points, it is difficult to compare to observations of available soil moisture from the top 1 m. For the COLA model, the wilting level was provided directly to the authors (David Straus and Yongkang Xue, personal communication). For this model, we subtracted the wilting level from the values provided to AMIP, but the values are still much higher than the observed values, possibly because the depth of soil considered is more than 1 m in many locations. Fig. 10 shows these global patterns. They look reasonable, except the SiBlings seem to be excessively wet in northeast Asia. GISS, on the other hand, is quite dry in northeast Asia. JMA (SiB) is much drier in the tropics than COLA and GLA (SSiB).

None of the 4 other bucket models with unknown variable field capacity produced patterns resembling the observations (Fig. 11). CCC had values exceeding 100 cm over all high latitudes and in China, but had very low values in western tropical Africa. UIUC, UKMO, and MGO all were very dry globally.

6. Comparison of seasonal cycles

The seasonal cycles of the models do not agree with observations in any of the regions examined. In general, the phase of the seasonal cycle is correct, but not the amplitude, with both the winter and summer values different from observations. In Fig. 12, we illustrate these points with the mean seasonal cycles for 1980-1988 (eliminating the first year of the simulations due to spinup problems in some of the models, which are discussed in the next section) for regions from each location for which we have observations, Russia and Illinois (see Figs. 2-3). Because the observations represent only plant-available soil moisture, models that provided total soil moisture have higher values. But because they did not provide wilting levels, it is not possible to make a direct quantitative comparison. For the Russian box (Fig. 12a), models with a 15-cm bucket all have values that are too low. The interesting seasonal cycle of the SUNYA/NCAR model, with zero values in the winter, is because they only provided liquid soil moisture, and the soil is frozen in the winter. The models in general lag the observations by 1 or 2 months, and CCC and GISS have lags of 3 months. For the Illinois box (Fig. 12b), almost all the models are too dry in the summer.

7. Comparison of interannual variations and spinup problems

To investigate the ability of the models to simulate interannual variability, anomalies for each simulation were calculated by subtraction of the monthly-mean soil moisture for the last 9 years of the simulation from the soil moisture for each month. The first year was not used, as several of the models had large trends in that year. These anomalies were compared to observations calculated in the same way for various homogeneous regions. Figure 13 shows these comparisons for the same regions shown in Fig. 12. None of the models simulate the observed interannual variations for any of the regions. For the Russian box (Fig. 13a), the dry conditions in 1981 and wet conditions in 1985 are not captured by any of the models, and the monthly and interannual variability of many of the models is less than observed, because they are already at field capacity. In Illinois, the 1988 drought is evident in the observations, but is not captured by the models (Fig. 13b).

Several models, identified in the left margin, have large trends in the first year, due to their initialization from fields not in balance with the model soil moisture climatologies. For the Russian box (Fig. 13a), the bucket models all reach equilibrium quickly; the 2 that start out too wet (DERF and MGO) take 1½ years, while the 3 that start too dry (CSU, SUNYA, and CCC) take only ½ year.

Three of the SiBling models (JMA, COLA, and GLA) have trends for most of the AMIP period, taking 8 years to reach equilibrium. Observations show that the characteristic autocorrelation time scale for soil moisture in this region is about 3 months (Vinnikov et al. 1996). Schlosser (1995) and Xue et al. (1996) have shown that the long SiBling time scale is due to the slow exchange of soil moisture between the deep third layer and the upper 2 layers. The SiBling output is total soil moisture, not plant-available, and extends to more than 1 m at some grid points, so the error is exaggerated in amplitude by a factor of 2 or 3, but not in time scale, in this figure. Some of the SiBlings also show large initial errors for the Illinois box (Fig. 13b), but they are not as long-lasting as in the Russian box, probably because the frozen soil in Russia slows down the drainage from the lowest SiB layer. Liston et al. (1993) recognize this problem in their simulations of soil wetness for the AMIP period, conducted in the same manner as Schemm et al. (1992) but using the SiB model, by explicitly adding a linear reservoir drainage term to the third soil layer of SiB. This reduces the spinup in their data-driven simulations to the order of months for the Russian box.

Another feature of the simulations evident in Fig. 13a, is that soil moisture is constant during the winter months for every model, but not for the observations. The characteristic horizontal line from December through March or April is evident in each curve. This is because the models assume that when the soil is frozen and snow-covered, liquid water does not enter the soil, but in nature, observations show month-to-month changes.

Phillips (1994) identified which models began their simulations with a previous model solution, and which began with climatological initialization (Table 3). For the Russian box (Figs. 2, 13a), we examined whether the effects could be seen in the results. Table 7 shows that 7 models were initialized with climatology, but appeared to be in balance at the start of the simulations. Conversely, 4 models were initialized with a previous model solution, but were not in balance. In addition, 5 models which were initialized with climatology were also not in balance. For the 6 initially unbalanced models with a bucket parameterization, the spinup time was 0.5-1.5 years. For the 3 SiBlings however, the spinup time was approximately 8 years.

8. Discussion and conclusions

The results presented here evaluate only the soil moisture simulations by the AMIP models. Ideally we would like to evaluate the latent and sensible heat fluxes generated by the models, as this is the main way in which they influence the atmosphere, but such global data sets do not exist. Sophisticated land surface schemes may indeed improve the simulation of these fluxes and the simulation of the diurnal cycle of surface temperature, but such an analysis is beyond the scope of the present paper. Mao and Robock (1997) found that even the simulations of monthly mean surface air temperature by the AMIP models had large errors, so perhaps such a study would be premature. From the viewpoint of soil moisture simulations only, we find that:

The "data sets" of Mintz and Serafini (1981, 1989, 1992) and Schemm et al. (1992) are quite different from the observations. They also differ from each other by as much as 10 cm (out of a total of 15 cm) in the annual average over Asia, and by several cm in general over the rest of the world, even though they use the same calculation method.
There is no evidence that more complicated models produce better simulations. Models which had multiple levels, or explicitly considered vegetation, or explicitly considered immediate or subsurface runoff did not produce better simulations of soil moisture than the bucket models. On the contrary, the SiBlings (JMA, COLA, and GLA) have time scales that are too long, especially in the regions with freezing temperatures.
A field capacity of 15 cm is too low to capture high latitude soil moisture variations, in the AMIP models and the "data sets." The models with higher field capacities (ECMWF, MPI) do reproduce the observed patterns in these regions.
There are large differences in the tropical model simulations of soil moisture. We do not have observations for this region to determine which are correct.
All the winter soil moisture simulations in high latitudes produce constant soil moisture when the temperature is below freezing and the ground is snow covered, while observations show that soil moisture varies in the winter as much as in other seasons.
Models begun without spinning up soil moisture produce long-term trends which can have serious effects on the first year or more of the simulations. These spinup problems probably contribute to the spinup errors in the surface air temperature fields shown by Mao and Robock (1997).
The observed interannual variations of soil moisture in Illinois and Russia were not captured by any of the AMIP models. Because the month-to-month and interannual soil moisture anomalies are strongly controlled by precipitation and cloud cover anomalies, this implies that interannual variations of precipitation and cloudiness in these regions are controlled by factors in addition to the global sea surface temperature variations used to force the AMIP models.

9. Future progress

The results presented above are for AMIP simulations conducted several years ago. Several GCM groups have improved their land surface parameterization schemes from the ones described here, and their current operational models produce different results. For example, the CSU model now uses SiB2 (Sellers et al. 1996a,b; Randall et al., 1996), the ECMWF model now uses the Viterbo and Beljaars (1995) scheme, and the CNRM model now uses the ISBA scheme (Noilhan and Planton 1989; Mahfouf et al. 1995) together with a new snow parameterization (Douville et al. 1995a,b).

While theoretical advances are important, all new parameterizations should be validated by comparison with observations. All the above groups have used our data in the development of their schemes and we have supplied the data to many other groups. We are continuing our collection, quality control and analysis of soil moisture data from around the world. The data are available to anyone wishing to make use of them.

The second AMIP experiment is now being planned (Gleckler, 1997). To allow a complete evaluation of the hydrological cycle, in contrast to the first AMIP experiment, the plans should make provisions to archive runoff and end-of-month values of precipitable water, snow depth, and plant-available soil moisture in the top 1 m of soil and the root depth for each grid point.

Acknowledgments. We thank Larry Gates for organizing the AMIP project, all the GCM modeling groups who have contributed their simulations, and the staff of the Program for Climate Model Diagnosis and Interpretation for processing the model output into an easy-to-use format. We thank Steve Hollinger for providing the Illinois soil moisture data; Vicky Serafini and Jae Schemm for providing their data sets; David Straus and Yongkang Xue for COLA wilting point data; Greg Walker for the Liston et al. (1993) simulations; Peter Gleckler and Tom Phillips for their help in documenting the soil moisture schemes; Mike Fiorino for help with GrADS, and Tom Delworth, Chris Milly, David Randall, Changan Zhang, Laura Ferranti, Pedro Viterbo, Yogesh Sud, David Straus, and Yongkang Xue for valuable discussions. This work is supported by DOE Office of Energy Research grant DE-FG02-93ER61691.A000, NOAA Climate and Global Change Program grants NA36GPO311 and NA56GPO212, and NASA grant NAGW-5227.

REFERENCES

Budyko, M.I., 1956. Heat Balance of the Earth's Surface, Gidrometeoizdat, Leningrad, 255 pp. (in Russian)

Chen, T.H., Henderson-Sellers, A., Milly, P.C.D., Pitman, A.J., Beljaars, A.C.M., Polcher, J., Abramopolous, F., Boone, A., Chang, S., Chen, F., Dai, Y., Desborough, C.E., Dickinson, R.E., Dümenil, L., Ek, M., Garratt, J. R., Gedney, N., Gusev, Y. M., Kim, J., Koster, R., Kowalczyk, E.A., Laval, K., Lean, J., Lettenmaier, D., Liang, X., Mahfouf, J.-F., Mengelkamp, H.-T., Mitchell, K., Nasonova, O.N., Noilhan, J., Robock, A., Rosenzweig, C., Schaake, J., Schlosser, C.A., Schultz, J.-P., Shao, Y., Shmakin, A.B., Verseghy, D.L., Wetzel, P., Wood, E.F., Xue, Y., Yang, Z.-L., and Zeng, Q., 1997. Cabauw experimental results from the Project for Intercomparison of Land-surface Parameterization Schemes (PILPS). J. Climate, 10: 1194-1215.

Delworth, T., and Manabe, S., 1988. The influence of potential evaporation on the variabilities of simulated soil wetness and climate. J. Climate, 1: 523-547.

Delworth, T., and Manabe, S., 1993. Climate variability and land surface processes. Adv. Water Resources, 16: 3-20.

Desborough, C.E., 1997. The impact of root-weighting on the response of transpiration to moisture stress in a land surface scheme. Mon. Wea. Rev., in press.

Dickinson, R.E., Henderson-Sellers, A., Kennedy, P.J., and Wilson, M.F., 1986. Biosphere Atmosphere Transfer Scheme (BATS) for the NCAR Community Climate Model. NCAR Tech. Note NCAR/TN-387+STR, 72 pp.

Dickinson R.E., Henderson-Sellers, A., Rosenzweig, C., and Sellers, P.J. 1991. Evapotranspiration models with canopy resistance for use in climate models, a review. Agricultural and Forest Meteorol., 54: 373-388.

Douville, H., Royer, J.-F., and Mahfouf, J.F., 1995. A new snow parameterization for the Météo-France climate model, Part I: Validation in stand-alone experiments. Climate Dyn., 12: 21-35.

Douville, H., Royer, J.-F., and Mahfouf, J.F., 1995b. A new snow parameterization for the Météo-France climate model, Part II: Validation in a 3-D GCM experiment. Climate Dyn., 12: 37-52.

Gates, W. L., 1992. AMIP: the Atmospheric Model Intercomparison Project. Bull. Amer. Met. Soc., 73: 1962-1970.

Gleckler, P., Editor, 1997. AMIP II Guidelines, AMIP Newsletter No. 8, Lawrence Livermore National Laboratory, Livermore, California, 20 pp.

Henderson-Sellers, A., Editor, 1996a. Special Issue on Soil Moisture Simulation. Global Plan. Change, 13(1-4).

Henderson-Sellers, A., 1996b. Soil moisture: A critical focus for global change studies. Global Plan. Change, 13: 3-9.

Henderson-Sellers, A., Yang, Z.-L., and Dickinson, R.E., 1993. The Project for Intercomparison of Land-Surface Parameterization Schemes (PILPS). Bull. Amer. Met. Soc., 74: 1335-1349.

Hollinger, S.E., and Isard, S.A., 1994. A soil moisture climatology of Illinois. J. Climate, 7: 822-833.

Houghton, J.T., Meira Filho, L.G., Callander, B.A., Harris, N., Kattenberg, A., and Maskell, K., Editors, 1996. Climate Change 1995-The Science of Climate Change, Cambridge Univ. Press, Cambridge, 584 pp.

Jaeger, L. 1976: Monatskarten des Niederschlags fur die ganze Erde. Berichte Deutschen Wetterdienst 18 Nr 139, Offenbach, Germany.

Koster, R.D. and Milly, P.C.D., 1997. The interplay between transpiration and runoff formulations in land surface schemes used with atmospheric models. J. Climate, 10: 1578-1591.

Lau, K.M., Kim, J.H. and Sud, Y., 1996. Intercomparison of hydrologic processes in AMIP GCMs. Bull. Amer. Met. Soc., 77: 2209-2227.

Liston, G.E., Sud, Y.C., and Walker, G.K., 1993. Design of a Global Soil Moisture Initialization Procedure for the Simple Biosphere Model, NASA Tech. Mem. 104590, National Aeronautics and Space Administration, Greenbelt, Maryland, 138 pp.

Mahfouf, J.F., Manzi, A.O., Noilhan, J., Giordani, H., and Déqué, M., 1995. The land surface scheme ISBA within the Météo-France climate model ARPEGE. Part I: Implementation and preliminary results. J. Climate, 8: 2039-2057.

Manabe, S., 1969. Climate and the ocean circulation, 1. The atmospheric circulation and the hydrology of the earth's surface. Mon. Weather Rev., 97: 739-774.

Manabe, S., Wetherald, R.T., and Stouffer, R.J., 1981. Summer dryness due to increase of atmospheric CO₂ concentration. Clim. Change, 3: 347-384.

Mao, J. and Robock, A., 1997. Surface air temperature simulations by AMIP general circulation models: Volcanic and ENSO signals and systematic errors. J. Climate, in press.

Meshcherskaya A.V., Boldyreva, N.A., and Shapaeva, N.D., 1982. Districts average plant available soil water storage and depth of snow cover. Statistical analysis and its usage (some examples). Gidrometeoizdat, Leningrad, 243 pp. (in Russian).

Mintz, Y., and Serafini, Y. 1981. Global fields of soil moisture and land-surface evapotranspiration. Research Review - 1980/81, NASA Goddard Space Flight Center Tech. Memo 83907, 178-180.

Mintz, Y., and Serafini, Y.V., 1989. Global monthly climatology of soil moisture and water balance. LMD Internal Report No. 148, Laboratoire de Météorologie Dynamique, Paris France, 102 pp.

Mintz, Y., and Serafini, Y.V., 1992. A global monthly climatology of soil moisture and water balance. Clim. Dyn., 8: 13-27.

Mocko, D.M. and Sud, Y.C., 1997. Comparison of a land-surface model (SSiB) to three parameterizations of evapotranspiration - a study based on ISLSCP Initiative I data. Submitted to J. Geophys. Res.

Noilhan, J. and Planton, S., 1989. A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 113: 1542-1553.

Phillips, T.J., 1994. A summary documentation of the AMIP models. PCMDI Report No. 18, Lawrence Livermore National Laboratory, Livermore, California, 343 pp. http://www-pcmdi.llnl.gov/modeldoc/amip

Pitman, A.J. , Henderson-Sellers, A., Abramopolous, F., Avissar, R., Bonan, G., Boone, A., Cogley, J.G., Dickinson, R.E., Ek, M., Entekhabi, D., Famiglietti, J., Garratt, J.R., Frech, M., Hahmann, A., Koster, R., Kowalczyk, E., Laval, K., Lean, J., Lee, T.J., Lettenmaier, D., Liang, X., Mahfouf, J.-F., Mahrt, L., Milly, P.C.D., Mitchell, K., deNoblet, N., Noilhan, J., Pan, H., Pielke, R., Robock, A., Rosenzweig, C., Running, S.W., Schlosser, C.A., Scott, R., Suarez, M., Thompson, S., Verseghy, D. L., Wetzel, P., Wood, E. F., Xue, Y., Yang, Z.-L. and Zhang, L., 1993. Results from off-line control simulations (Phase 1a), PILPS, GEWEX Report, IGPO Pub. Ser. 7, 47 pp.

Pitman, A.J., Henderson-Sellers, A., Abramopolous, F., Boone, A., Desborough, C.E., Dickinson, R.E., Garratt, J.R., Gedney, N., Koster, R., Kowalczyk, E.A., Lettenmaier, D., Liang, X., Mahfouf, J.-F., Noilhan, J., Polcher, J., Qu, W., Robock, A., Rosenzweig, C., Schlosser, C.A., Shmakin, A.B., Smith, J., Suarez, M., Verseghy, D.L., Wetzel, P., Wood, E.F., Xue, Y., and Yang, Z.-L., 1998. Key results and implications from Phase 1(c) of the Project for Intercomparison of Land-surface Parameterization Schemes. Submitted to Climate Dynamics.

Qu, W., Henderson-Sellers, A., Pitman, A.J., Chen, T.H., Abramopolous, F., Beljaars, A.C.M., Boone, A., Chang, S., Chen, F., Dai, Y., Desborough, C.E., Dickinson, R.E., Dümenil, L., Ek, M., Garratt, J.R., Gedney, N., Gusev, Y.M., Kim, J., Koster, R., Kowalczyk, E.A., Laval, K., Lean, J., Lettenmaier, D., Liang, X., Mahfouf, J.-F., Mengelkamp, H.-T., Milly, P.C.D., Mitchell, K., Nasonova, O.N., Noilhan, J., Polcher, J., Robock, A., Rosenzweig, C., Schaake, J., Schlosser, C.A., Schultz, J.-P., Shmakin, A.B., Verseghy, D.L., Wetzel, P., Wood, E.F., Xue, Y., Yang, Z.-L. andZeng, Q., 1998. Sensitivity of latent heat fluxes from PILPS land-surface schemes to perturbations of surface air temperature. J. Atmos. Sci., in press.

Randall, D.A., Sellers, P.J., Dazlich, D.A., Zhang, C., Collatz, G.J., Denning, A.S., Los, S.O., Field, C.B. and Fung, I., 1996. A revised land-surface parameterization (SiB2) for GCMs. Part III: The greening of the Colorado State University general circulation model, J. Climate, 9: 738-763.

Robock, A., Vinnikov, K.Y., Schlosser, C.A., Speranskaya, N.A. and Xue, Y., 1995a. Use of midlatitude soil moisture and meteorological observations to validate soil moisture simulations with biosphere and bucket models. J. Climate, 8: 15-35.

Robock, A., Schlosser, C.A., Vinnikov, K.Y., Liu, S. and Speranskaya, N.A., 1995b. Validation of humidity, moisture fluxes, and soil moisture in GCMs: report of AMIP Diagnostic Subproject 11; Part 1 - soil moisture. Proceedings of the First International AMIP Scientific Conference, WCRP-92, WMO/TD-No. 732, W. L. Gates, (Editor), World Climate Research Programme, Geneva, 85-90.

Robock, A., Vinnikov, K.Y., and Schlosser, C.A., 1997. Reply (Evaluation of land-surface parameterization schemes using observations). J. Climate, 10: 377-379.

Rodriguez-Iturbe, I., Vogel, G.K., Rigon, R., Entekhabi, D., Castelli, F. and Rinaldo, A., 1995. On the spatial organization of soil moisture fields. Geophys. Res. Lett., 22: 2757-2760.

Schemm, J., Schubert, S., Terry, J. and Bloom, S., 1992. Estimates of Monthly Mean Soil Moisture for 1979-1989. NASA Tech. Mem. 104571, National Aeronautics and Space Administration, Greenbelt, Maryland, 260 pp.

Schlosser, C.A., 1995. Land-Surface Hydrology: Validation And Intercomparison Of Multi-Year Off-Line Simulations Using Midlatitude Data. Ph.D. dissertation, Department of Meteorology, University of Maryland, College Park, 135 pp.

Schlosser, C.A., Robock, A., Vinnikov, K.Y., Speranskaya, N.A. and Xue, Y., 1997. 18-year land-surface hydrology model simulations for a midlatitude grassland catchment in Valdai, Russia. Mon. Weather Rev., in press.

Sellers, P.J., Mintz, Y., Sud, Y.C. and Dalcher, A., 1986: A simple biosphere model (SiB) for use within general circulation models. J. Atmos. Sci., 43: 505-531.

Sellers, P.J., Randall, D.A., Collatz, G.J., Berry, J.A., Field, C.B., Dazlich, D.A., Zhang, C., Collelo, G.D. and Bounoua, L., 1996a. A revised land-surface parameterization (SiB2) for GCMs. Part I: Model formulation, J. Climate, 9: 676-705.

Sellers, P.J., Los, S.O., Tucker, C.J., Justice, C.O., Dazlich, D.A., Collatz, C.J. and D.A. Randall, 1996b. A revised land-surface parameterization (SiB2) for GCMs. Part II: The generation of global fields of terrestrial biophysical parameters from satellite data, J. Climate, 9: 706-737.

Sellers, P.J., Dickinson, R.E., Randall, D.A., Betts, A.K., Hall, F.G., Berry, J.A., Collatz, G.J., Denning, A.S., Mooney, H.A., Nobre, C.A., Sato, N., Field, C.B. and Henderson-Sellers, A., 1997. Modeling the exchange of energy, water, and carbon between continents and the atmosphere. Science, 275: 502-509.

Sengupta, S. and Boyle, J.S., 1997. Comparing GCM Simulations, Ensembles and Model Revisions Using Common Principal Components, UCRL-JC-126621, PCMDI Report No. 40 Lawrence Livermore National Laboratory, Livermore, California, 32 pp.

Shao, Y. and Henderson-Sellers, A., 1996. Validation of soil moisture simulation in landsurface parameterisation schemes with HAPEX data. Glob. Plan. Change, 13: 11-46.

Slater, A.G., Pitman, A.J., andDesborough, C.E., 1997a. The simulation of freeze-thaw cycles in a GCM land surface scheme. Submitted to J. Geophys. Res.

Slater, A.G., Pitman, A.J. and Desborough, C.E., 1997b. The validation of a snow parameterization for use in general circulation models. Submitted to Int. J. Clim.

Spangler, W.M.L. and Jenne, R.L., 1984. Reference manual: World monthly surface station climatology. National Center for Atmospheric Research, Boulder, Colorado.

Thornthwaite, C.W., 1948. An approach toward a rational classification of climate. Geogr. Rev., 38: 55-94.

Vachaud G., Passerat de Silans, A., Balabanis, P., and Vauclin, M., 1985. Temporal stability of spatially measured soil water probability density function. Soil Sci. Soc. Am. J., 49: 822-828.

Vinnikov K.Y. and Yeserkepova, I.B., 1991. Soil moisture: empirical data and model results. J. Climate, 4: 66-79.

Vinnikov, K.Y., Robock, A., Speranskaya, N.A. and Schlosser, C.A., 1996. Scales of temporal and spatial variability of midlatitude soil moisture. J. Geophys. Res., 101: 7163-7174.

Viterbo, P. and Beljaars, A.C.M., 1995. An improved land surface parameterization scheme in the ECMWF model and its validation, J. Climate, 8: 2716-2748.

Watson, R.T., Zinyowera, M.C. and Moss, R.H. (Editors), 1996: Climate Change 1995-Impacts, Adaptations and Mitigation of Climate Change: Scientific-Technical Analyses, Cambridge Univ. Press, Cambridge, 880 pp.

Weare, B.C., Mokhov, I.I. and Project Members, 1995. Evaluation of total cloudiness and its variability in the Atmospheric Model Intercomparison Project, J. Climate, 8: 2224-2238.

Wetzel, P.J., Liang, X., Irannejad, P., Boone, A., Noilhan, J., Shao, Y., Skelly, C., Xue, Y. and Yang, Z.L., 1996. Modeling vadose zone liquid water fluxes: Infiltration, runoff, drainage, interflow. Glob. Plan. Change, 13: 57-71.

Xue, Y., Sellers, P.J., Kinter, J. L. and Shukla, J., 1991: A simplified biosphere model for global climate studies. J. Climate, 4: 345-364.

Xue, Y., Zeng, F.J. and Schlosser, C.A., 1996. SSiB and its sensitivity to soil propertiesa case study using HAPEX-Mobilhy data. Glob. Plan. Change, 13: 183-194.

Yang, Z.-L., Dickinson, R.E., Robock, A. and Vinnikov, K.Y., 1997. On validation of the snow sub-model of the Biosphere-Atmosphere Transfer Scheme with Russian snow cover and meteorological observational data. J. Climate, 10: 353-373.

Table 1. How To Categorize Soil Moisture Schemes

1. Calculated or prescribed soil moisture

2. Initialization: climatology or previous model solution

3. Available (bucket) or total (SiBling) soil moisture

4. Field capacity (12 - 212.5 cm)

5. Number of levels (0 - 6)

6. Vegetation: implicit ( x potential evaporation), or

explicit (canopy interception and stomatal resistance)

7. Runoff: bucket overflow

- plus immediate, as fraction of precipitation?

- plus subsurface (leaky bucket)?

Table 2. AMIP Modeling Groups

Acronym	AMIP Group	Location
BMRC	Bureau of Meteorology Research Centre	Melbourne, Australia
CCC	Canadian Centre for Climate Modelling and Analysis	Victoria, Canada
CNRM	Centre National de Recherches Météorologiques	Toulouse, France
COLA	Center for Ocean-Land-Atmosphere Studies	Calverton, Maryland (USA)
CSIRO	Commonwealth Scientific & Industrial Research Organization	Mordialloc, Australia
CSU	Colorado State University	Fort Collins, Colorado (USA)
DERF	Dynamical Extended Range Forecasting(at GFDL)	Princeton, New Jersey (USA)
DNM	Department of Numerical Mathematics(of the Russian Academy of Sciences)	Moscow, Russia
ECMWF	European Centre for Medium-Range Weather Forecasts	Reading, England
GFDL	Geophysical Fluid Dynamics Laboratory	Princeton, New Jersey (USA)
GISS	Goddard Institute for Space Studies	New York, New York (USA)
GLA	Goddard Laboratory for Atmospheres	Greenbelt, Maryland (USA)
GSFC	Goddard Space Flight Center	Greenbelt, Maryland (USA)
IAP	Institute of Atmospheric Physics(of the Chinese Academy of Sciences)	Beijing, China
JMA	Japan Meteorological Agency	Tokyo, Japan
LMD	Laboratoire de Météorologie Dynamique	Paris, France
MGO	Main Geophysical Observatory	St. Petersburg, Russia
MPI	Max-Planck-Institut für Meteorologie	Hamburg, Germany
MRI	Meteorological Research Institute	Ibaraki-ken, Japan
NCAR	National Center for Atmospheric Research	Boulder, Colorado (USA)
NMC	National Meteorological Center	Suitland, Maryland (USA)
NRL	Naval Research Laboratory	Monterey, California (USA)
RPN	Recherche en Prévision Numérique	Dorval, Canada
SUNYA	State University of New York at Albany	Albany, New York (USA)
SUNYA/	State University of New York at Albany/	Albany, New York (USA)/
NCAR	National Center for Atmospheric Research	Boulder, Colorado (USA)
UCLA	University of California at Los Angeles	Los Angeles, California (USA)
UGAMP	The UK Universities' Global Atmospheric Modelling Programme	Reading, England
UIUC	University of Illinois at Urbana-Champaign	Urbana, Illinois (USA)
UKMO	United Kingdom Meteorological Office	Bracknell, United Kingdom
YONU	Yonsei University	Seoul, Korea

Table 3. Categories Of Models And Initialization

BUCKET WITH MODEL INITIALIZATION (13)
Field capacity^**
12 cm	CNRM
15 cm	DNM*, GFDL, IAP, LMD, SUNYA, YONU
16.2 cm	CSIRO
variable	CCC^V, MGO, UIUC, UKMO^V
200 cm	MRI
BUCKET WITH CLIMATOLOGY INTIALIZATION (8)
15 cm	BMRC, CSU, DERF, NMC^V
20 cm	MPI^V
26 cm	ECMWF^V, UGAMP
212.5 cm	SNG (LSX)^V
SiBLING WITH MODEL INITIALIZATION (2)
variable	GLA (SSiB)^V, GISS^V
SiBLING WITH CLIMATOLOGY INITIALIZATION (2)
variable	COLA (SSiB)^V, JMA (SiB)^V
PRESCRIBED SOIL MOISTURE (5)
	GSFC, NCAR, NRL, RPN^§, UCLA^§

^V with explicit vegetation (canopy interception and stomatal resistance)

^** cm of plant-available soil moisture for the entire soil depth

* with prescribed deep soil moisture

^§ not in Standard Output

Table 4. Number Of Layers Of Soil Moisture Models

0	1	2	3	4	5	6
GSFC	BMRC	CNRM	COLA	MRI		GISS
NCAR	CCC	CSIRO	ECMWF			SUNYA/NCAR
NRL	CSU	DNM	GLA
RPN	DERF	MGO	JMA
UCLA	GFDL		UGAMP
	IAP
	LMD
	MPI
	NMC
	SUNYA
	UIUC
	UKMO
	YONU

Table 5. Vegetation Parameterization In Soil Moisture Models

		Explicit (canopy interception,
Implicit (b x PE)		stomatal resistance)
BMRC	CCC	COLA	ECMWF
CNRM	CSIRO	GISS	GLA
CSU	DERF	JMA	MPI
DNM	GFDL	NMC*	SUNYA/NCAR
GSFC	IAP	UKMO
LMD	MGO	* no canopy interception
MRI	NCAR
NRL	RPN
SUNYA	UCLA
UGAMP	UIUC
YONU

Table 6. Runoff Parameterization In Soil Moisture Models

None	Bucket Overflow	Bucket Overflow	Bucket Overflow
(Prescribed W)	Only	plus Immediate Fraction	plus Immediate Fraction
		from Precipitation	from Precipitation
			plus Deep Subsurface
GSFC	BMRC	CSIRO	COLA
NCAR	CCC	ECMWF	GISS
NRL	CNRM	IAP	GLA
RPN	CSU	MGO	JMA
UCLA	DERF	MRI	MPI
	DNM	UIUC	SUNYA/NCAR
	GFDL	YONU	UKMO
	LMD
	NMC
	SUNYA
	UGAMP

Table 7. Effects of Initialization for the Region 20-60°E, 55-65°N

Models with climatological initialization (Phillips, 1994), but which appear to be balanced initially:

GISS, ECMWF, SUNYA/NCAR, UGAMP, BMR, MPI, NMC

Models with initialization from a previous model solution (Phillips, 1994), but which appear to not be balanced initially:

Time to reach

Model Initial imbalance (cm) equilibrium (years)

CCC -17 0.5

SUNYA -11 0.5

IAP +3 1.0

MGO +3 1.5

Models with climatological initialization (Phillips, 1994) and which are imbalanced initially:

CSU -7 0.5

DERF +6 1.5

JMA +12 8.0

COLA +28 8.0

GLAS +37 8.0

FIGURE CAPTIONS

Figure 1. Schematic diagram of hydrological and meteorological scales of soil moisture variations. r is the autocorrelation function. The scales are determined by the slopes of the curves: for time scales, , t is time and T is the time scale; for spatial scales, , d is distance and L is the length scale.

Figure 2. Observed available soil moisture (cm) in the top 1 m, averaged for 1979-1988 for Russia. Station locations are indicated by the circles. The box shows the region used for the comparison of models and observations in Figs. 12-13.

Figure 3. Observed available soil moisture (cm) in the top 1 m, averaged for 1979-1988 for Illinois, USA. Observations started in 1981. Station locations are indicated by the circles. The box shows the region used for the comparison of models and observations in Figs. 12-13.

Figure 4. Comparison of Mintz and Serafini (1981, 1989, 1992) and Schemm et al. (1992) "data sets" with each other and with observations.

Figure 5. Annual average of prescribed soil moisture fields for the 3 models for which this variable was provided to the AMIP Standard Output. Note that NCAR and NRL are in unspecified units, presumably on scales of 0-10 and 0-1 of wetness fractions. GSFC is in cm of plant-available soil moisture in the top 1 m of soil.

Figure 6. Observed field capacity for Russian stations. The stations with red circles often have measured available soil moisture that exceeds the field capacity, while for those circled in blue, the field capacity always exceeds the measured available soil moisture.

Figure 7. Annual average of plant-available soil moisture fields for 6 of the models with standard 15-cm buckets. Units are in cm of plant-available soil moisture in the top 1 m of soil.

Figure 8. Annual average of plant-available soil moisture fields for 4 of the models with standard 15-cm buckets, one with a 12-cm bucket and one with a 16.2-cm bucket. Units are in cm of plant-available soil moisture in the top 1 m of soil.

Figure 9. Annual average of wetness (plant-available soil moisture divided by field capacity) for 3 of the models which only provided soil moisture for the top layer, and for 2 models with large field capacities. Shown are wetness only for those top layers (MRI, SUNYA/NCAR, UGAMP) and for the entire soil layer (ECMWF, MPI).

Figure 10. Annual average of total soil moisture (cm) for the 3 SiBlings and for GISS. The depth of the soil layer varies spatially. For the COLA model, the plant-available soil moisture is also shown.

Figure 11. Annual average of plant-available soil moisture fields for 4 of the models with unknown variable field capacities. Units are in cm of plant-available soil moisture.

Figure 12. Seasonal cycle of soil moisture (cm) averaged for the last 9 years of the AMIP simulations for all models and for observations for 2 different boxes where we have observations, a) Russia (Fig. 2) and b) Illinois (Fig. 3). For the models, the monthly average is plotted. For the observations, data from the first day of the month are plotted.

Figure 13. Soil moisture anomalies (cm) with respect to the monthly means for 1980-1989 for all models and for observations, averaged for same regions as in Fig. 12.

		Time to reach
Model	Initial imbalance (cm)	equilibrium (years)
CCC	-17	0.5
SUNYA	-11	0.5
IAP	+3	1.0
MGO	+3	1.5