Land - surface modelling

DECEMBER 1997 3279S C H L O S S E R E T A L .

q 1997 American Meteorological Society

18-Year Land-Surface Hydrology Model Simulations for a Midlatitude GrasslandCatchment in Valdai, Russia

C. ADAM SCHLOSSERGeophysical Fluid Dynamics Laboratory/NOAA, Princeton, New Jersey

ALAN ROBOCK AND KONSTANTIN YA. VINNIKOVDepartment of Meteorology, University of Maryland at College Park, College Park, Maryland

NINA A. SPERANSKAYAState Hydrological Institute, St. Petersburg, Russia

YONGKANG XUEDepartment of Geography, University of Maryland at College Park, College Park, Maryland

(Manuscript received 23 May 1996, in final form 24 April 1997)

ABSTRACT

Off-line simulations of improved bucket hydrology and Simplified Simple Biosphere (SSiB) models areperformed for a grassland vegetation catchment region, located at the Valdai water-balance research station inRussia, forced by observed meteorological and simulated actinometric data for 196683. Evaluation of the modelsimulations is performed using observations of total soil moisture in the top 1 m, runoff, evaporation, snowdepth, and water-table depth made within the catchment. The Valdai study demonstrates that using only routinemeteorological measurements, long-term simulations of land-surface schemes suitable for model evaluation canbe made. The Valdai dataset is available for use in the evaluation of other land-surface schemes.

Both the SSiB and the bucket models reproduce the observed hydrology averaged over the simulation period(196783) and its interannual variability reasonably well. However, the models soil moisture interannual vari-ability is too low during the fall and winter when compared to observations. In addition, some discrepancies inthe models seasonal behavior with respect to observations are seen. The models are able to reproduce extremehydrological events to some degree, but some inconsistencies in the model mechanisms are seen. The bucketmodels soil-moisture variability is limited by its inability to rise above its prescribed field capacity for the casewhere the observed water table rises into the top 1-m layer of soil, which can lead to erroneous simulations ofevaporation and runoff. SSiBs snow depth simulations are generally too low due to high evaporation from thesnow surface. SSiB typically produces drainage out of its bottom layer during the summer, which appearsinconsistent to the runoff observations of the catchment.

1. Introduction

Within the past decade, the range of complexityamong the land-surface parameterizations used in gen-eral circulation models (GCMs) has grown significantly.Current land-surface models are based on a wide rangeof theoretical frameworks, ranging from a simplebucket hydrology (Budyko 1956; Manabe 1969), toa more sophisticated biosphere or big-leaf scheme[e.g., the Simple Biosphere (SiB) model (Sellers et al.

Corresponding author address: Dr. C. Adam Schlosser, GFDL/NOAA, Princeton University, Forrestal Campus, U.S. Route 1, P.O.Box 308, Princeton, NJ 08542.E-mail: [email protected]

1986) and the BiosphereAtmosphere Transfer Scheme(BATS) model (Dickinson et al. (1986)]. Recent studieshave shown significant sensitivities to the choice ofland-surface parameterizations for GCM applications(e.g., Rind et al. 1992; McKenney and Rosenburg 1993).However, these studies had no observations for modelcomparison and evaluation.

In an effort to improve our understanding and para-meterizations of land-surface processes, a wide varietyof observational and diagnostic modeling studies havebeen conducted. One widely used methodology involvesoff-line simulations of the land-surface schemes. Theland-surface schemes are decoupled from the GCMs andforced by observational and/or simulated data (whenneeded). The prognostic outputs of the models are thencompared to observed values.

3280 VOLUME 125M O N T H L Y W E A T H E R R E V I E W

Calibration and verification studies using Amazondata (e.g., Sellers and Dorman 1987; Sellers et al. 1989;and Xue et al. 1996b) have addressed the parameteri-zations of land-surface processes for a tropical climate,but verification studies must also be performed for midand high-latitude environments. Off-line studies of theProject for the Intercomparison of Land-Surface Param-eterization Schemes (PILPS) (Henderson-Sellers et al.1993) have recently used data from the Cabauw site inthe Netherlands (Chen et al. 1996). The Cabauw datacontained estimates of observed latent and sensible heatfluxes for model evaluation. However, the forcing dataspanned only one year and the resulting spinup issuescomplicated the model analysis. In addition, an analysisof the interannual variability of the models could notbe performed.

Robock et al. (1995; hereafter referred to as RO) usedroutine meteorological and actinometric data from sixstations within midlatitude Russia spanning the years197883 to perform off-line runs with a standard GCMbucket model (Manabe 1969) and the Simplified SimpleBiosphere (SSiB) (Xue et al. 1991) model. The resultsrevealed some discrepancies in both of the models, andsubsequently, improvements to SSiBs runoff parame-terization have been made (Xue et al. 1996a). However,the study lacked any model comparisons to observationsof evaporation, runoff, and water-table depth, and thesix years of forcing for the model simulations, the lon-gest for off-line simulations to date, were still insuffi-cient for an extensive interannual analysis. Further stud-ies using different stations with more extensive dataspanning longer time periods were strongly recom-mended by RO.

Recently, Vinnikov et al. (1996) described a uniqueset of data obtained from the Valdai water-balance re-search site in Russia. An extensive set of meteorolog-ical, hydrological, and actinometric measurements wasgathered at three catchments within the research site forthe years 1960 to 1990. The three catchments, Tayozh-niy, Sinaya Gnilka, and Usadievskiy, each containeddifferent dominant vegetation types: a mature forest, agrowing forest, and grassland, respectively. Vinnikov etal. (1996) used routine soil moisture measurementstaken within each catchment to perform an extensivestatistical analysis of the characteristic temporal andspatial scales of soil moisture.

Our principal aim of this study is to demonstrate thatan effective seasonal and interannual diagnostic analysisof off-line land-surface hydrology simulationsneverbefore conductedcan be obtained using the Valdaidata. We use meteorological data taken near the Usa-dievskiy (grassland) catchment and simulated radiationdata to perform off-line simulations for a slightly mod-ified bucket hydrology model (similar to that of the ROstudy) and the SSiB model. Overall, the Valdai simu-lations for the bucket and SSiB models demonstrate thatvarious land-surface schemes are able to be forced withonly routine meteorological data over many years to

produce useful simulations for model evaluation. Withthe need for only meteorological data to force the mod-els, the potential exists to utilize many datasets such asValdai for off-line testing. The meteorological forcingdataset at Valdai spans 18 years (196683), allowingfor much longer model simulations than previous stud-ies. The Valdai dataset contains measurements of soilmoisture, water table depth, runoff, snowcover, and pre-cipitation within the catchment. Thus, the observed wa-ter balance for the catchment can be closed by calcu-lating evaporation as a residual. In addition, a subset ofevaporation estimates calculated by Fedorov (1977) isalso used as further validation. The models are allotteda spinup period of 1 year determined from previoussensitivity tests (Schlosser 1995). The remaining portionof the simulation period, 17 years, is then used to verifyagainst observations. The models interannual and sea-sonal variability is evaluated, and the models ability toreproduce extreme hydrologic events (e.g., droughts) isassessed.

In the next section, we describe the data of the Usa-dievskiy catchment at Valdai used for the simulationexperiments. In section 3, we describe some improve-ments made to the parameters and codes of each of themodels for the simulations, and outline the basic frame-work of the SSiB and bucket hydrologies. The resultsof the model simulations and a verification and inter-comparison analysis are presented in section 4. Finally,discussions, conclusions, and closing remarks are given.

2. Data

The Valdai research station is located in a forest re-gion south of St. Petersburg at 57.68N, 33.18E and con-tains an extensive set of water-balance observationcatchments of different dominant vegetation type (Fe-dorov 1977 and Vinnikov et al. 1996). A summary ofthe current data collection is given in Table 1. The entireobservational dataset obtained from Valdai spans theyears 196090. This study focuses on the model sim-ulations of the grassland catchment. The grasslandcatchment, Usadievskiy, is approximately 0.36 km2.

a. Hydrological data

Within the Usadievskiy catchment, 11 representativesites (out of more than 100) were chosen for total soilmoisture observations. Near the end of each month, arepresentative sample of soil cores is taken at each site,and the total soil moisture is measured by the thermo-stat-weight (gravimetric) technique (described in RO).The accuracy of the soil moisture measurements in thetop 1-m layer is 61 cm (RO). The seasonal variationsof total soil moisture in the top 1 m at each of the 11measurement sites are very similar and their differencesare mostly biases (Fig. 4 of Vinnikov et al. 1996). There-fore, we use the average value of the 11 measurement


TABLE 1. Valdai data collection.

Hydrological measurements for three different catchmentstakenevery monthTotal soil moisture in the top 100 cm for 911 sites within each

catchment, 196385Catchment averaged total soil moisture in the top 20, 50, and

100 cm, 196090Runoff, 196090Depth of groundwater table, 196090Water equivalent snow depth,* 196090Precipitation, 196090

Radiative flux measurements10 day averages of 3-h totals takenduring growing seasonTotal solar radiation, 196090Net shortwave radiation for

Various agricultural fields, 196788Top of boreal forest, 196090Boreal forest under the tree canopy, 196790

Albedo forNatural meadow field, 196787Top of boreal forest, 196090

Meteorological data (including precipitation) taken every 3 h, 196683

* Measurements of snow depth also taken after snow events andfrequently during snowmelt.

FIG. 2. Annual averaged values of observed total soil moisture,depth of water table, snowmelt, and runoff for the three catchmentsat Valdai. In addition, annual averaged observed precipitation mea-sured near the grassland catchment is given.

FIG. 1. Map of the Usadievskiy catchment at Valdai. Water-tablemeasurement sites are indicated by filled circles. Soil moisture mea-surement sites are indicated by filled squares. Open circles withdashed lines indicate the snow measurement sites and routes, re-spectively. Runoff is measured at the stream outflow point of thecatchment located in the lower left-hand corner of the map and in-dicated by a bold bracket. The short dash line denotes the catchmentboundary. Hatched areas denote regions of swampy conditions. El-evation contours are in increments of 2 m.

sites for total soil moisture in the top 1 m to comparewith the model simulations.

Runoff measurements are obtained by streamflow ob-servations made at the outflow point of the catchmentregion (as indicated by a bracket in Fig. 1). Triangleweirs are used to measure the water level and are thentranslated into a water flux outflow from the catchment.A quantitative error analysis of the runoff measurementswas difficult to obtain. Generally speaking, a high de-gree of confidence can be placed on the measurementsmade during the warmer months. During the coldermonths when the stream is typically frozen and thespringtime when the stream typically overflows, the run-off measurements tend to be less accurate.

Numerous wells that measure water-table depth arelocated at various points within the catchment (Fig. 1).The water-table depth measurements obtained for thisstudy are an average of all the measurements taken ateach of the well sites within the catchment.

An example of the data available from the Valdairesearch site is given in Fig. 2. A similar figure was


FIG. 3. Monthly observations for the years 196783 of total soilmoisture, water-table depth, snow depth (water equivalent), and run-off for Usadievskiy. Shown also are monthly observations of pre-cipitation. Tick-mark labels correspond to January of the indicatedyear.

given by Vinnikov et al. (1996) with the following con-clusions. The results show that the interannual vari-ability of total soil moisture does not seem to differsignificantly between each one of the catchment sites.Observations of water-table depths would seem to sug-gest the forest catchment shows slightly less interannualvariability than the grassland and growing forest catch-ments. Looking at the runoff results, it appears that run-off in the grassland catchment is always higher than inthe forest and has more interannual variability, as mighthave been expected. But in the growing forest catch-ment, the runoff first parallels the grassland from 1960to 1970, showing that these measurements are indeedrepresentative, then parallels the forest for 5 years(197175). For the last 10 years (197685), the growingforests runoff is less than the mature forests, whichwould suggest that the growing forest is utilizing ad-ditional water to mature the canopy.

In 1975, the annual averaged values of total soil mois-ture and depth of the water table are at their lowest anddeepest, respectively. This would suggest that a signif-icant drying event had occurred. The monthly obser-vations (Fig. 3) reveal the dramatic drying period duringthe summer of 1975 with a very deep water table, verydry soil moisture stores, and no runoff during the latterpart of the year. In addition, a fairly robust summerdrying is also seen to occur during 1972. The dry con-ditions of the warmer months of 1972 and 1975 seemto be preceded by one or more months with below nor-mal precipitation between January and May. In addition,a very shallow winter snowcover occurs in 1972.

b. Meteorological dataRoutine observations of standard meteorological vari-

ables have been made at the Valdai research site for anumber of years (Table 1). The collection of meteoro-logical data used in this study spans a time period of196683. The meteorological instruments used to obtainthis data were located on a natural grassland plot. Theseobservations were taken at 3-h intervals, eight times perday (0000, 0300, 0600, 0900, 1200, 1500, 1800, and2100 local time) and include the following meteorolog-ical variables needed as forcing for the land-surfacemodel simulations: Ta air temperature at standard levelof measurement (8C), Td dewpoint temperature at stan-dard level of measurement (8C), P precipitation (mm),CL lower cloud-cover fraction, CT total cloud-cover frac-tion, V wind speed at standard level of measurement (ms21), and pa sea level pressure (mb). These variableswere then used to force the models directly or to supplyvalues to algorithmic arguments used within the models.

An additional set of unique rain gauge data was alsoutilized in this study. Near the Usadievskiy catchment,monthly totals of precipitation were obtained throughthe use of a modified rain gauge instrument (gauge 98;Golubev et al. 1995). The rain gauge was designed sothat the effects of wind biasing on the catchment of

snowfall and rainfall in the gauge were minimized bysurrounding the rain gauge with natural shrub-type veg-etation. The height of the shrub vegetation was main-tained at the same height as the top of the rain gauge.Precipitation totals were recorded twice daily for theyears 197483. The seasonal cycle of monthly precip-itation measured at the rain gauge 98 site, as comparedto the precipitation measured at the meteorological in-strument site, shows the measurements obtained for raingauge 98 are higher during the fall and winter months(Fig. 4). This would suggest that the rain gauge at themeteorological site was not capturing as much precip-itation as rain gauge 98 due to the aerodynamic effectson wind-blown snow (D. Yang et al. 1995).

Through the use of these rain gauge 98 data, monthlycorrection coefficients for the precipitation measure-ments at the meteorological instrument site were cal-culated in order to correct for wind biasing. Each month-ly correction coefficient mM is simply the ratio of therain gauge 98 monthly total to the meteorological raingauge monthly total, averaged for all the years:


FIG. 4. Seasonal cycles averaged for the years 197483 of monthly precipitation measuredwith rain gauge 98 and the gauge at the meteorological station.

TABLE 2. Monthly values of the precipitation correction coefficientmM. Note: For the remaining months, mM 5 1.0.

Jan Feb Mar Nov Dec

mM 1.2 1.1 1.1 1.1 1.2

1983 98PMOmetoP1974 Mm 5 , (1)M 10

where is the monthly total precipitation measuredmetoPMat the meteorological site, and is the monthly total98PMprecipitation measured at gauge 98. Correction valuesare greater than 1 during the fall and winter months andnot significantly different from 1 during the spring andsummer months (Table 2). Using the appropriate month-ly value of mM, the result is a corrected value of theprecipitation for each time step, PC, given by

PC 5 mMP. (2)

c. Evaporation as a residualUsing the monthly observations of soil moisture, run-

off, snowcover, and precipitation, a residual estimate ofmonthly evaporation is possible. By rewriting the basicwater-balance equation that describes changes in soilmoisture in the top 1 meter of soil,

dW5 P 2 E 1 M 2 R , (3)R 1mdt

where W is total soil moisture in the top 1 m of soil,PR is rainfall (measured precipitation that occurred whenTa . 08C), E is evapotranspiration, M is snowmelt, andR1 m is runoff from the top 1 m of soil, we can obtainan equation for an evaporation residual estimate

dWE 5 P 1 M 2 R 2 . (4)R 1m dt

Measurements of water equivalent snow depth weretaken during measurable snow events, near the end ofeach month and every few days in the spring duringrapid snowmelt. So we calculated monthly snowmelt Mas the total observed decrease in water equivalent snowdepth within the month. The runoff from the top 1 mof soil R1 m is comprised of the overland runoff RO andthe drainage of groundwater RD out of the top 1 m ofsoil:

R1 m 5 RO 1 RD. (5)The monthly runoff measurements made at Usadiev-

skiy reflect the total runoff from the catchment and notfrom the top 1 m of soil. To account for this inconsis-tency, variations in observed water-table depth wereused. If the water-table depth WT was observed to bedeeper than 1 m, a decrease in the depth of the watertable was assumed to reflect the drainage coming fromthe top 1 m layer of soil, RD, and an increase in thewater-table depth was assumed to contribute to catch-ment runoff (i.e., the observed runoff). Changes in wa-ter-table depth were equated to these water fluxes by anempirical constant mW. When the water table was ob-served to be in the top 1 m of soil, the total catchmentrunoff R was assumed to be representative of the runofffrom the top 1 m of soil:

dWTR 2 m , WT . 1 m (6)WR 5 dt1m R, WT , 1 m. (7)5

Through empirical evidence and basic soil physicsarguments, the most appropriate value for mW is 0.1(Fedorov 1977).


FIG. 5. Monthly evaporation estimates using a residual calculation for the years 196683 andobservations from Federov (1977) for the years 19661973 at Usadievskiy catchment.

FIG. 6. Seasonal cycles averaged for the years 196673 for evaporation estimates at Usadievskiyfrom Fedorov (1977) and from residual calculations. The seasonal cycle (196673) of rainfallused in the residual calculations is also shown.

The calculated residual evaporation shows a well-defined seasonal cycle with a summer maximum, as onewould expect (Fig. 5). However, the residual calculationalso produces negative values of evaporation, especiallyduring the winter months, and these values are mostlikely erroneous.

Evaporation measurements and estimates for the Usa-dievskiy catchment were also made by Fedorov (1977)from 1960 to 1973 (Fig. 5). During the warmer months,May through October, weighing lysimeters were usedto measure monthly evaporation within the catchment.For the remaining months, November through April, anestimate of evaporation was calculated using the Bu-dyko (1956) algorithm for potential evaporation esti-

mates. A good agreement in the seasonal cycles betweenthe residual and the Fedorov estimates is seen duringthe warmer months (Fig. 6). However, the observationsshow a fair amount of disagreement during the fall andwinter months. The residual estimate appears to be sys-tematically higher than the Fedorov measurements dur-ing the months of September and October. This is likelycaused by the high amounts of rainfall that were mea-sured (Fig. 6).

Nonetheless, the relative agreement between the Fe-dorov observations and the residual estimates of evap-oration during the summer months, when evaporativefluxes are most significant, is encouraging. In light ofthe likely erroneous negative values during the winter


FIG. 7. Seasonal cycles (196683) of (a) observed solar radiationand simulated solar radiation at Valdai using the Berlyand (1961)algorithm and (b) observed albedo for a natural grassland vegetation(meadow) plot at Valdai.

months and the slight bias during the fall of the residualcalculations, the Fedorov estimates are likely the betterchoice to use as observations. However, both the Fe-dorov estimates and the residual calculations will beused in the validation analysis of the model simulations(section 4).

d. Radiation data

The data obtained from the Valdai research site didnot include any downward longwave measurements. Inorder to force the models, an algorithm was needed tosimulate the incoming longwave radiation. The resultsfrom Brutsaert (1982) show the Satterlund (1979)scheme to be the best algorithm for estimating clear-sky downward longwave radiation (especially for sub-freezing air temperatures). In addition, results from theRO study show that models forced with the Satterlundscheme agree well with observations in their simulationof net radiation. As a result, the Satterlund scheme wasused for the simulations to provide an effective estimateof incoming longwave radiation. Recently, other studies(e.g., Yang et al. 1997; A. Pitman 1996, personal com-munication) have chosen the Idso (1981) longwave al-gorithm. However, our sensitivity analysis (not shown)reveals only slight differences between the simulationsof both the bucket and SSiB models result using thetwo different longwave algorithms. To adjust for cloudyconditions, the Monteith (1961) formula was used withsome slight modifications (refer to RO for a completedescription).

Measurements of shortwave radiation were made atValdai during the growing season but were halted duringthe winter due to the extreme freezing conditions. Inaddition, the shortwave observations were supplied in10-day averaged diurnal cycles, rather than the routineformat of the meteorological data (every 3 h). As such,the shortwave radiation data were inappropriate to useas forcing for the models and an additional algorithmwas needed to simulate incoming solar radiation. Usingthe data from the RO study, Schlosser (1995) comparedsimulations of incoming solar radiation with an algo-rithm developed by Berlyand (1956) and Berlyand(1961) to the observations. In addition, test simulationsof the bucket and SSiB models for each of the stationsusing the solar algorithm as forcing rather than obser-vations show that Berlyands algorithm is quite suitableto be used as solar forcing for these simulations. Thealgorithm also agrees well with the observations at Val-dai (Fig. 7), and their differences are well within the 20W m22 instrument error (Vinnikov and Dvorkina 1970)of the Yanyshevski pyranometer used to measure thesolar radiation. The algorithm was also able to reproducethe observed diurnal structure of solar radiation ratherwell (not shown). In light of these results, the Berlyandalgorithm is quite suitable for the Valdai simulation ex-periments.

3. ModelsThe two models used for this study are the simple

bucket hydrology model (Budyko 1956; Manabe 1969)and the SSiB model (Xue et al. 1991). Some slightmodifications to both of the models were made for theValdai simulations based on the results of RO andSchlosser (1995). These modifications are discussed be-low and summarized in Table 3. The simple bucket hy-drology framework is still widely used for land-surfaceparameterization in GCMs, and the SSiB model rep-resents the recent efforts to provide more sophisticatedand explicit parameterizations of land-surface processesfor GCM applications. So it is useful to examine themechanisms of the models seasonal and interannualhydrologies, which differ in complexity, to each otherand to the Valdai observations.

a. Model hydrologiesThe bucket hydrology provides a simple representa-

tion of the plant-available soil-water budget in the top1 m of soil using one soil layer storage term:


TABLE 3. Modifications to bucket and SSiB models.

Bucket modelAlbedo for grassland set to 0.23 (default value is 0.18 for no

snow cover)Deep snow albedo set to 0.75 (default value is 0.6)Bucket capacity set to 15.6 cm of plant-available soil moisture

(default value is 15 cm)Potential evaporation correction (Milly 1992)

SSiB modelPorosity set to 0.401 (default value is 0.42 for grassland)New runoff scheme implemented (Milly and Eagleson 1982)Large-scale basin flow algorithm turned offMonthly albedo shifted during growing season

May 5 20.008June 5 10.012July 5 20.002August 5 20.078September 5 20.06

Snow albedo set to 0.75 (default value is 0.6)Wilting level parameter c1 set at 6.44 to match observed wilting

levels of the top 1 m of soil (default value is 5.8)Middle soil layer thickness set to 98 cm (default value 47 cm

for grassland)

dWa 5 P 2 E 1 M 2 R , (8)R b b bdtwhere Wa is available soil moisture in the top 1 m ofsoil, Eb is bucket evapotranspiration, Mb is bucket snow-melt, and Rb is bucket runoff. The bucket runoff Rb asoriginally formulated by Budyko (1956), consists of anoverland flow component, where a fraction of rainfallimmediately runs off, and a drainage component (bucketoverflow). However, for the Valdai simulations we chosethe typical GCM bucket assumption of no instantaneousrunoff from rainfall. The bucket will, therefore, onlyproduce runoff when it is saturated (at field capacity)and the infiltrated precipitation is assumed to drain rap-idly. The bucket evapotranspiration is given by

Eb 5 bEp, (9)where b 5 Wa/0.75Wfc is soil moisture stress to evapo-transpiration, b # 1. When snowcover is present or theground temperature is below freezing, no changes ofsoil moisture are allowed. Any precipitation that fallsin the form of rain is runoff, and evaporation is takenfrom the snowcover (at the potential rate). The budgetequation for the snowcover is then

dWsb 5 P 2 E 2 M , (10)s p bdtwhere Wsb is water equivalent snow depth and Ps issnowfall (precipitation that falls when Ta # 08C).

The SSiB model hydrology scheme contains three soillayers and a canopy layer. The budget equation for thecanopy water storage is given by

dWc 5 P 2 D 2 E , (11)c cdt

where Wc is canopy interception water store, Pc is pre-cipitation intercepted by the canopy, D is drainage rate,and Ec is evaporation of canopy water store. For SSiBsthree soil layers, the water budgets can be expressed as

dW1 5 P 2 Q 2 E 2 r E 2 R (12)i 12 bs 1 t sdtdW2 5 Q 2 Q 2 r E (13)12 23 2 tdt

anddW3 5 Q 2 Q 2 r E , (14)23 3 3 tdt

where Wi is total soil moisture in the ith soil layer, Piis precipitation infiltration, Qij is water flux between theith and jth soil layers (positive denotes a downwardflux), Ebs is bare soil evaporation rate, Et is transpirationrate, Rs is surface runoff rate, Q3 is drainage rate fromthe lowest soil layer, and ri is root-distribution factorfor the ith soil layer. When the top soil layer is unsat-urated, Pi is equal to the precipitation that was not in-tercepted by the canopy (throughfall) and Rs 5 0; uponsaturation, any excess precipitation is assumed to besurface runoff Rs. The interlayer water flux terms arecalculated by solving for the diffusion equation usingthe finite-difference method. During frozen soil condi-tions the infiltration rates, interlayer water fluxes, anddrainage rates are diminished by decreasing the meanhydraulic conductivity of each soil layer linearly withsoil temperature (described in a later section). A detaileddescription of each of the terms in (12)(14) is givenby Xue et al. (1996a). The budget equation for SSiBswater equivalent snow depth Ws is given by

dWs 5 P 1 F 2 M 2 E , (15)s s sdtwhere Ms is snowmelt, F is freezing of water on canopy,and Es is evaporation from snow. Any water stored onthe canopy that freezes is added to the water equivalentsnow depth [the F term in (15)]. Similar to the bucketframework, during snowcover conditions SSiB does notallow any evapotranspiration from the soil layers. Itshould be noted that the Es term was erroneously leftout in the text of Xue et al. (1996a).

b. Bucket model modificationsThe RO study found that for some of the station cases,

the bucket model was limited in its soil moisture sim-ulations by the assumed 15-cm capacity of its plant-available soil moisture store (hereafter referred to asplant-available field capacity). Sensitivity tests ofSchlosser (1995) with the bucket model using observedplant-available field capacities of the RO stationsshowed significant improvements to the simulations. Inlight of these results, the field capacity of the bucket


TABLE 4. Soil characteristics at Usadievskiy. Coverage and water-holding characteristics for each soil type (Fedorov 1977). Percentcoverage is based upon areal coverage within the catchment, W isthe wilting level of the top 1 m of soil, Wf is the field capacity ofthe top 1 m of soil, and W0 is the total water-holding capacity of thetop 1 m of soil. Catchment averages are weighted with respect topercentage of soil-type area coverage.

Soil type LoamSandyloam Sandy

Catchmentaverage

Percent coverageW (cm)Wf (cm)W0 (cm)

5613.430.837.4

2810.925.344.7

165.8

17.541.2

11.527.140.1

model was adjusted for the Valdai simulations. An av-erage field capacity for the Usadievskiy catchment wascalculated using observations of soil type coverage overthe catchment and measurements of field capacity andwilting level available from Valdai for the different soiltypes.

Table 4 lists the measurements obtained for each soiltype and the catchment averaged values of total waterholding capacity W0, field capacity W f, and wilting levelW

*, for Usadievskiy (Fedorov 1977). The averaged field

capacity estimate for Usadievskiy is 27.1 cm and theaveraged wilting level is 11.5 cm. Field capacity andwilting level measurements were taken according to astandard technique designed for the Russian water-bal-ance station network. Field capacity is measured by sat-urating a covered and enclosed plot of soil (in order toprevent evaporation loss and lateral movement of soilwater, respectively) and monitoring the decay of rapiddrainage from the soil. The wilting level is calculatedon an agricultural plot using an oat crop. Vinnikov andYeserkepova (1991) give a more detailed explanationof these measurement techniques. Since the bucket mod-el simulates plant-available soil moisture, W

*is sub-

tracted from W f to obtain an appropriate field capacityfor the bucket model of 15.6 cm. When comparing withobservations of total soil moisture, W

*is then added

back to the buckets simulated available soil moistureto produce a bucket total soil moisture.

Sud and Fennessy (1982) and Milly (1992) noted aninconsistency in the bucket model parameterization ofpotential evaporation used in GCMs. For the Valdai sim-ulations, the bucket model was modified to correct forthis inconsistency so that the potential evaporation rateEp was calculated for the surface temperature at com-pletely saturated soil conditions (rather than for the ac-tual surface temperature):

Ep 5 rCD|V|( 2 qa),q*ts (16)where r is air density, CD is drag coefficient, |V| is windspeed at standard measurement height, is specificq*tshumidity of air temperature of a saturated surface, andqa is specific humidity at standard level of measurement.We use the standard value of CD (50.003) equal to thatof the GFDL GCM for all surfaces (RO) and within the

acceptable range of values for grassland vegetation(Hartmann 1994). It should be noted that the value ofCD, which partly depends upon the roughness of thecanopy surface (Stull 1991), can be varied according toobservations of a particular site (as in Chen et al. 1997)or to more consistently represent a smoother or roughervegetation. So in theory, the bucket model can representa wide range of vegetation types in its heat flux param-eterization.

Finally, the bucket model albedo was adjusted for theValdai simulations. Observations of albedo for a mead-ow (i.e., natural grassland vegetation) plot at Valdaiwere taken for the years 196683 during the warmermonths. These observations were then used to adjustthe buckets albedo. The standard bucket assumes a val-ue of 0.18 (RO). Figure 7 gives an averaged seasonalcycle of the albedo measurements during the growingseason. The observed albedo has an average value forthe growing season of approximately 0.23. The standardbucket scheme assumes a constant value, thus the ob-served monthly variations of albedo (Fig. 7) cannot bereproduced. As such, the buckets albedo was set to 0.23as the best approximation. Unfortunately, no observa-tions of albedo during snowcovered conditions weremade at Valdai. However, the results of the RO studysuggest that the standard bucket snow albedo is too lowfor off-line simulations of natural grassland vegetation(Fig. 14 of RO). Using the observations of the RO study,a snow albedo of 0.75 was prescribed to the bucketmodel for the Valdai simulations.

c. SSiB model modificationsThe results of the RO study suggested that during the

springtime SSiB showed an inconsistent partitioning ofsnowmelt into runoff (Fig. 8 of RO). In light of theseresults, a new runoff scheme was tested with SSiB. Thenew runoff scheme includes modified mean hydraulicconductivity calculations that are based upon the par-ameterizations of Milly and Eagleson (1982). This run-off scheme is also included in the latest test version ofSiB2 (Sellers et al. 1996). The most notable differencefrom the original runoff scheme is during frozen soilconditions. The mean hydraulic conductivity Kav is de-creased to limit the infiltration of snowmelt, interlayerexchanges of soil moisture, and gravitationally drivendrainage. The mean hydraulic conductivity for frozensoil conditions Kfrz is given by

Kfrz 5 FadjKav, (17)where

T 2 263soil, 263 , T , 273soilF 5 10 (18)adj 0, T # 263 soil

andT , infiltration and upper-layer moisturesT 5 fluxes (K)soil T , lower-layer soil moisture fluxes (K). deep


FIG. 8. Annual averaged values of observed and simulated totalsoil moisture, snowmelt, evaporation [observed residual denoted byan open circle and (Fedorov 1977) by a crosshair], and runoff forthe years 196783 at Usadievskiy. For the residual evaporation es-timates, (4) was used with annual values for the right-hand-side terms.In addition, the annual averages of observed precipitation are given.

Here, Tdeep is the deep soil temperature. This algorithmreplaces the rather simple criteria of the original runoffscheme where no snowmelt infiltration, interlayer soilmoisture exchanges, and drainage are allowed if thedeep soil temperature is below freezing. With the newrunoff scheme, SSiBs soil moisture during the spring-time shows a more realistic peak as seen in the obser-vations for four of the six RO stations (Xue et al. 1996a).While the test results of SSiB are not conclusive, it doesappear that the new runoff scheme of SSiB shows animproved simulation of runoff partitioning during thespringtime. As a result, the new runoff scheme waschosen to be used for the SSiB simulations of Valdai.

Another slight modification was made to SSiBs run-off parameterization. An empirical algorithm used tosimulate large-scale river baseflow was turned off sincethe Valdai simulations were for a much smaller scale (acatchment).

In the RO study, the standard GCM values for a grass-land vegetation physiology were used. However for theValdaiUsadievskiy simulations, a few of these param-eter values were replaced with observed values. Thearea-averaged total water-holding capacity (i.e., poros-ity) of the top 1-m of soil was estimated to be 40 cm(Table 4). So a soil porosity of 0.40 for the catchmentwas used in place of the standard porosity value forgrassland of 0.42. In addition, SSiBs wilting level wasadjusted to fit observations. Following the relationshipdescribed in detail by Xue et al. (1996a), we adjustedSSiBs wilting level parameter, c1, to match the wiltinglevel for the top 1 m of soil at Usadievskiy. The newand default values of c1 are given in Table 4. SSiB hasno explicit field capacity parameter to set to observa-tions, but rather a drainage term that varies exponen-tially with soil wetness (Xue et al. 1991). Using themethod described by Xue et al. (1996a), we find thatwhen we set SSiBs soil moisture stores equal to theobserved field capacity of the catchment (Table 4), theSSiB drainage rate becomes very small (less than 0.25mm day21), and so SSiB is essentially at its field capacityas well. Thus, no further parameter modifications wereneeded to adjust SSiBs implied field capacity.

For the remaining soil and vegetation parameters thatcould not be assigned observed values, the standardGCM values for grassland vegetation type were used.These standard GCM values of the parameters fall with-in the realistic range of values for the observed soil andvegetation types of the catchment (Table 4). However,the values of soil parameters, such as saturated hydraulicconductivity, are not a function of vegetation type (asis done for SSiB in a GCM) and can vary by orders ofmagnitude for a given soil type (Domenico andSchwartz 1990). In addition, the lack of observationsand the resulting uncertainties for some of the vegetationparameters such as the leaf area index (LAI) can havean impact on SSiBs evaporation calculations (Xue etal. 1996b). We performed sensitivity runs and foundthat SSiBs latent heat flux calculation is sensitive tochanges in LAI for the Valdai simulations. The resultsof these tests will be discussed in the concluding sectionof the paper.

SSiBs simulations of soil moisture will be comparedwith measurements of total soil moisture made in thetop 1 m of soil. As a result, SSiBs soil layer depthswere modified so that its top two root-active layers werecontained in the top 1 m of soil. Only the middle soillayer thickness was changed from the standard value of0.47 to 0.98 m. The top and bottom layer thicknesseswere kept at their standard values of 0.02 and 1.00 m,respectively.

SSiBs albedo was also adjusted according to obser-vations. However, the vegetation parameter values werenot modified, as quite a few vegetation parameters canaffect SSiBs albedo calculations such as greenness,LAI, vegetation cover fraction, leaf angle distribution,and reflectances. Thus, it would be difficult to choose


TABLE 5. Statistics of observed and simulated hydrologies. Aver-ages and standard deviations for the years 196783. Standard devi-ations are computed using annual means (Fig. 8). For evaporation,light font indicates statistics computed for the years 196783 (ob-servations are the residual estimates), and the bold font indicatesstatistics computed for the years 196773 [observations are the Fe-dorov (1977) data].

Total soil moisture(cm)

Evaporation(mm day21)

Runoff(mm day21)

Observed

Bucket

SSiB

25.5 6 1.1

25.2 6 0.8

25.1 6 0.9

1.0 6 0.21.2 6 0.041.0 6 0.11.1 6 0.071.2 6 0.11.3 6 0.08

0.8 6 0.3

1.0 6 0.2

0.6 6 0.2

which parameter to modify, especially when no obser-vations of these values were made at Valdai. Instead, asubroutine was constructed, which would shift the cal-culated albedos by a prescribed constant value for eachmonth. Using the standard values of vegetation param-eters for grassland, SSiBs albedos were calculated andthen compared to the observations shown in Fig. 7.Monthly albedo adjustment values were prescribed sothat SSiBs albedos matched the observations (Table 3).For snowcover, SSiBs snow reflectances were set to0.75 according to the observations of the RO stations.

d. Spinup of modelsA set of simple experiments, similar in design as de-

scribed by Z. L. Yang et al. (1995), was performed withthe data of the RO study in order to provide a morethorough assessment of the bucket and SSiB modelsinherent spinup for off-line simulations using the Rus-sian data (Schlosser 1995). The specific findings of thesetests are beyond the scope of this paper. However, theoverall results suggest that all simulation biases due tothe initialization can be removed for the bucket andSSiB models if they are allotted a spinup integrationperiod of 1 year and their soil moisture stores initializedas completely saturated. In light of these results, theValdai simulations begin at 1966 with completely sat-urated soil moisture stores. The simulated output from1967 to 1983 of the integration is then used to comparewith observations.

4. Evaluation and intercomparison of modelsimulations

a. Annual means and interannual and seasonalvariability

Generally speaking, the models do reasonably wellat simulating the observed averaged hydrology over thesimulation period and the interannual variability (Table5). However, while both models show the soil moisturedrying in 1972 and 1975 similar to the observations,the observations show 1975 as the driest year and the

models have 1972 as the driest year (Fig. 8). The largestdisagreement between the models simulated annual av-eraged total soil moisture in the top 1 m and the ob-servations occurs during the years when their annualaverages are most variable (197176). SSiB also tendsto produce less annual snowmelt (i.e., lower snowdepths) and runoff with respect to observations and thebucket. The reason for SSiBs snowmelt discrepancy isdiscussed in the seasonal cycle analysis. The bucketmodel tends to produce higher amounts of annual runoffthan both the observations and SSiB. In 1977, the ob-served runoff rates for Usadievskiy during the fall areinconsistently low with respect to the runoff of the othertwo catchments (not shown) and to the high precipita-tion rates (Fig. 3), and thus appear questionable. As aresult, the observed runoff was omitted from the inter-annual analysis (Fig. 8 and Table 5).

Qualitatively, both the models and observations showsimilar traits for their seasonal cycles of total soil mois-ture, snowdepth, evaporation, and runoff (Fig. 9). Soilmoisture is at its lowest during the summer months andis replenished during the fall. The models and obser-vations also agree fairly well on the magnitude of thesummer drying of the soil. Evaporation is at its maxi-mum during the summer. Runoff peaks during the springsnowmelt. Closer inspection however reveals notabledifferences between the modeled and observed seasonalvariability. The bucket tends to make July the driestmonth, while observations show August as the driestmonth on average. SSiBs summer drying of the soilagrees with the bucket. Another notable disagreementis seen during the spring snowmelt. Observations showthe soil moisture on average to peak above its catchmentaveraged field capacity of 27.1 cm (Table 4). The bucketmodel simulation keeps its soil moisture at field capacityon average from late fall into early spring with no peakoccurring due to the snowmelt. Similarly, SSiBs soilmoisture appears to be fairly constant at just below thebucket and observed field capacity from late fall to earlyspring.

The bucket models soil moisture is limited by itsfield capacity and thus cannot reproduce the observedspring peak of soil moisture. Since the bucket has lesssoil moisture than the observations in the spring and thehighest evaporation rate with respect to SSiB and theobservations in May (Fig. 9), it is able to dry out quickerduring the ensuing summer.

SSiB, in theory, can account for the observed springbehavior of soil moisture and groundwater. Observa-tions reveal that the water-table depth rises above 1 mevery year during the springmelt (Fig. 3). To be con-sistent with observations, SSiBs lowest soil layershould be saturated (i.e., wetness is equal to 1.0) duringthe spring snowmelt. However, on average SSiBs low-est soil layer is not saturated (Fig. 10) and this willcause soil water from SSiBs upper layers to drain rap-idly (by gravity). As such, the quick drainage of soilwater from SSiBs upper layers during the snowmelt


FIG. 9. Seasonal cycles of observed and simulated total soil mois-ture, snow depth (water equivalent), evaporation (residual calcula-tion), and runoff for the years 196783 at Usadievskiy. In addition,the seasonal cycle of observed precipitation (196783) and the sea-sonal cycle of observed evaporation from Fedorov (1977) for theyears 196773 is given.

FIG. 10. Top frame: seasonal cycle for the years 196783 of water-table depth observations for the Usadievskiy catchment. Bottomframe: seasonal cycle for the years 196783 of SSiBs simulated soilwetness in its lowest layer.

makes it difficult to maintain total soil moisture in itstop two layers above field capacity (Fig. 9).

Differences between the modeled and observed hy-drologies are also seen in the seasonal variability ofevaporation (Fig. 9). The buckets evaporation peaks inMay, while the observations show evaporation to peakin June. SSiBs evaporation peak appears to extendthrough June and July. In addition, the magnitude ofboth modeled peaks in evaporation is lower than theobserved peak. The buckets earlier peak in evaporationcould be a result of its soil moisture store drying outearlier in the summer, which causes its soil moisturestress to become high and limit evaporation for the re-maining summer months. The models tend to agree withthe Fedorov observations during the fall rather well, andthe higher residual observations are likely a result ofits fall bias described earlier (Fig. 6). In addition, SSiB

produces more evaporation than the bucket from latefall into early spring.

For runoff, both the models and observations on av-erage show a runoff maximum associated with thespring snowmelt. In March, the bucket model shows thelargest decrease of snowcover and thus the highestamount of runoff, since the bucket is at saturation (Fig.9) and all of its snowmelt goes into runoff. In April,the observations have the highest amount of snowmelt(Fig. 9) and a water-table depth less than 1 m (Fig. 10),so the higher runoff rates than the models would beexpected. During the summer, SSiB tends to producemore runoff than both the bucket and observations. Thisdiscrepancy is caused by the drainage term, Q3, fromits bottom soil layer (14) and not by surface runoff (12),as W1 is always well below saturation during the summerfor the Valdai simulations (not shown). The bucket run-off rates tend to agree rather well with observations fromlate spring to early fall. During the late fall, the bucketproduces the highest amount of runoff due to increasedprecipitation and a saturated soil moisture store. It isdifficult to determine whether SSiB or the bucket pro-duces a better runoff simulation during the fall. In the


FIG. 11. Seasonal cycles (196783) of simulated evaporation from the snow cover for thebucket and SSiB models.

winter, both the modeled and observed runoff rates arelow and in good agreement.

On the average, the bucket model produces a bettersimulation of snow depth than the SSiB model (Fig. 9).While both models tend to underestimate the snowcover,SSiBs low bias appears significant and may be the causeof SSiBs low runoff rates during the spring snowmelt.The snow depth simulation results are similar to thefindings of the RO study. We have determined thatSSiBs lower snow depths are a result of the highamounts of evaporation from the snow surface producedin the winter (Fig. 11). In our preliminary results, lowerevaporation rates from the snow surface can be achievednot only by tuning SSiBs internal parameterizations butalso by decreasing the downward longwave forcing(which is external to the model). However, the test re-sults also show that while a decrease in downward long-wave improves the snow depth simulation, SSiB thenproduces an erroneous delay in the timing of the snow-melt. Nonetheless, the need of a physically consistentmodification to correct this feature is uncertain at thistime and is still under investigation.

The temporal standard deviations (i.e., interannualvariability) of the modeled and observed seasonal cycles(Fig. 12) show notable discrepancies in their interannualvariability of soil moisture. Both the bucket and SSiBmodels show a marked seasonality to their interannualvariability of soil moisture. However, the observationsshow no such seasonality in the interannual variabilityof soil moisture. The buckets interannual variability ofsoil moisture is largest in June and July and becomes

almost zero from the late fall to the early spring. How-ever, since the bucket is typically saturated from the latefall to early spring (Fig. 9), one would expect very littleinterannual variability of soil moisture. However, evenfor unsaturated conditions during the fall and winter,the buckets soil moisture variability would also be lim-ited by its assumption of no changes of soil moistureduring frozen conditions. SSiB has its maximum inter-annual variability of soil moisture during August. Fromthe late fall to the early spring, SSiBs variability of soilmoisture is low and remains fairly constant, except fora slight peak in March. SSiBs low soil moisture vari-ability during the late fall and winter is a result of itslow hydraulic conductivity values during freezing con-ditions (see section 3), which limit the fluxes of waterin and out of its soil layers.

The bucket and the observations seem to be in fairagreement as to the seasonality of interannual variabilityof evaporation, with higher values during the summerand lower values during the winter (Fig. 12). The ob-servations show a peak in variability in August whilethe bucket peaks in April and June. SSiB has its highestdegree of variability of evaporation during March andMay. In addition, SSiBs evaporation has a higher de-gree of variability than the bucket and observations dur-ing the winter.

The seasonality of the interannual variability of sim-ulated and observed runoff corresponds well with thevariability of precipitation (Fig. 12). Relative maximaof runoff variability occur during the spring, late sum-mer, and fall. The notable differences are SSiBs low


FIG. 12. Temporal standard deviations of the seasonal cycles ofobserved and simulated total soil moisture, evaporation (Fedorov1977), and runoff for the years indicated at Usadievskiy. In addition,standard deviations of observed precipitation are given.

FIG. 13. Monthly results of observed and simulated total soil mois-ture, snow depth (water equivalent), evaporation (observed residualcalculation only), and runoff for the year 1974 at Usadievskiy. Ob-served monthly precipitation (bars) and the seasonal cycle of pre-cipitation (dotted line) for the years 196783 are also given.

variability from October through January, also in March(possibly due to very little snowmelt; Fig. 9), and thebuckets high bias in August and April.

b. Wet yearsWhile 1974 shows no significant increase in annual

averaged precipitation (Fig. 8), about twice as much rainfell than the average for July (Fig. 13). As a result,observations show the soil not to dry out as expectedduring the late summer (Fig. 9) and, in fact, soil moistureincreases during July.

The bucket model performs reasonably well at re-producing the observed soil moisture variations during1974 (Fig. 13), with the slight exception in June. Theobservations show a decrease in soil moisture, while thebucket tends to increase soil moisture due to a lowerevaporation rate. In July, the bucket produces more run-off than observations show. However, the bucket is al-most at field capacity for July and so most of the rainfallis partitioned into runoff.

In June, SSiB produces less evaporation but higherrunoff rates (from its drainage term) than the obser-vations (Fig. 13), which results in slightly less drying.SSiB, however, simulates a drying of the soil duringJuly, contrary to the bucket and observations, whichcould be result of its higher runoff (drainage) and evap-oration rates than the observations (Fig. 13).

c. DroughtsFor the entire time span of the Valdai data, 1972 has

the lowest annual averaged precipitation (Fig. 2). Inaddition, water-table depths averaged over the year areof the deepest of the record. During the summer, therainfall in June and August was less than half of theaverage (Fig. 14). After a peak of soil moisture inMarch, the soil is observed to go through a robust dryingperiod until midsummer when total soil moisture is closeto its catchment averaged wilting level (Table 4). The


FIG. 14. Monthly results of observed and simulated total soil mois-ture, snow depth (water equivalent), evaporation (observed residual,denoted by an open circle and (Fedorov 1977) by a crosshair), andrunoff for the year 1972 at Usadievskiy. Observed monthly precip-itation (bars) and the seasonal cycle (dotted line) for the years 196783 are also given.

FIG. 15. Monthly results of observed and simulated total soil mois-ture, snow depth (water equivalent), evaporation (observed residualcalculation only), and runoff for the year 1975 at Usadievskiy. Ob-served monthly precipitation (bars) and the seasonal cycle (dottedline) for the years 196783 are also given.

rise of soil moisture observed in August with little rain-fall appears questionable. With an increase in soil mois-ture and little rainfall, the residual estimate of evapo-ration for August become slightly negative, which islikely erroneous. The Fedorov evaporation observationfor August is approximately 1.5 mm day21.

Generally speaking, both the bucket and SSiB modelsreproduce the drying of the soil fairly well, but do notreproduce the observed spring peak in soil moisture(Fig. 14). The bucket does a good job reproducing theobserved runoff during the snowmelt. SSiB, however,shows no such runoff peak in response to the snowmelt.In fact, nearly all of the snowcover is melted by SSiBa month earlier than the observations and the bucket.After the snowmelt, the buckets drying is seen to par-allel the observations quite well until July, but then thebuckets evaporation decreases considerably, which dis-agrees with observations. As a result, the buckets dry-

ing diminishes. SSiB tends to dry out the soil a bit moreslowly than observed during the early spring. However,from late spring into summer, SSiBs drying parallelsthe observations quite well.

Much like the year of 1972, the annual averaged pre-cipitation indicates the year of 1975 to be one of thedriest years on record for the Valdai data (Figs. 3, 15).In addition, the water-table depth is seen to drop to itsdeepest level (Figs. 3, 4). A closer inspection revealsthat the year of 1975 suffered a significant drought dur-ing the fall (Fig. 15). As a result, soil moisture is ob-served to continue to decrease in September, which nor-mally shows a replenishing of soil moisture (Fig 9).

Looking at the bucket simulation (Fig. 15), soil mois-ture decreases during September similar to observations.However, a large positive bias of the bucket soil mois-ture with respect to observations is seen. This bias is aresult of a bucket simulation discrepancy that occurs inJuly. The bucket model shows an increase in soil mois-


ture in July, while the observations show soil moistureto remain fairly constant. The bucket increase in Julysoil moisture is likely a result of an erroneous simulatedevaporation rate that is lower than the observed residual.

SSiB does not seem to reproduce the observed hy-drology for 1975 much better than the bucket (Fig. 15).SSiB dries the soil out at a fairly constant rate throughJuly and produces additional runoff in June and July,contrary to the bucket and observations. During the falldrought, SSiBs evaporation rates are lower than ob-servations, and as a result, SSiB does not continue todry out the soil during the drought. In addition, SSiBproduces very little runoff in March due to very littlesnowmelt, contrary to the bucket and observations.

5. Discussion and conclusionsWith 17 years of model simulation (196783) and

observations of total soil moisture in the top 1 m, runoff,water table, evaporation, and snow depth, the Valdaisimulations are the first of their kind to effectively di-agnose the interannual and seasonal variability of theland-surface models simulated hydrologies. Both thebucket and SSiB were improved based on previous re-sults, and the experiments presented here go one stepfarther in the observationally based validation of thesemodels. From the results presented above, we can reachthe following conclusions.

Both the bucket hydrology and SSiB models simulatereasonably well the observed annual average hydrologyand its interannual variability. However, the models un-derestimate the interannual variability of soil moistureduring the late fall and winter. This result could be amodel inconsistency but also site specific and scale de-pendent. Further investigation is needed in order to ad-dress this issue.

The success of the bucket model and SSiB at repro-ducing extreme hydrological events is varied. During adrought, both the models were able to produce drier soilconditions, however, the other components of their sim-ulated hydrologies were not accurately simulated. Forexample, SSiB produced runoff, not seen in the obser-vations, and both the bucket and SSiB models evap-oration rates were low compared to observations. Toincrease our confidence in coupled GCM climate stud-ies, land-surface models must at least be able to accu-rately reproduce the hydrological response to precipi-tation anomalies in off-line tests.

The inability of the bucket model to rise above itsfield capacity can lead to errors in its simulation. Thebucket model cannot reproduce the observed peak oftotal soil moisture in the top 1 m during the springsnowmelt, since the total soil moisture in the top 1 mof soil rises above the bucket field capacity with thewater table in the top meter of soil. As a result, thebucket dries out the soil more quickly in the summerthan the observations and then produces lower evapo-ration rates than the observations during the midsum-

mer, since its soil moisture stores have been depleted.The bucket also produces more runoff than the obser-vations during the fall, since the bucket is at its fieldcapacity, and, as a result, all rainfall is partitioned intorunoff. The low degree of interannual variability of theseasonal cycle of total soil moisture during the fall andwinter of the bucket, contrary to observations, is alsoa result of a saturated bucket unable to rise above itsfield capacity. While the bucket discrepancy may be asite-specific feature, preliminary test results (not re-ported here) show that a modification to the bucket mod-el that would allow its soil moisture to rise above fieldcapacity and then drain gradually (rather than instan-taneously) could produce more realistic soil moisturevariations and improve its evaporation and runoff sim-ulations.

While the bucket intercomparison results of thisstudy and RO appear somewhat different from thePILPS results for Cabauw (Chen et al. 1997), thesedifferent intercomparison results can be attributed tothe bucket models sensitivity to the value of its dragcoefficient and are not a result of differences in thebucket model performance. In the bucket run for Ca-bauw reported in Chen et al. (1997), the value of thedrag coefficient is more than twice as large as the high-est value in the typical range for a grassland vegetation(Hartmann 1994). Using a more consistent drag co-efficient (calculated using the site-measured roughnesslength for heat and moisture fluxes), the buckets anom-alous result for the Cabauw simulations was removed(Chen et al. 1997). Our tests with the bucket modelusing the Russian data of RO (not shown) give sen-sitivities to the drag coefficient similar to the Cabauwtest results.

The most notable discrepancies of SSiBs simula-tions compared to observations are its lower snowdepths and higher summer runoff rates. SSiBs under-estimation of snowcover has been attributed to its highamounts of evaporation from the snow surface. We canimprove SSiBs snow depth simulation and also reducethe disagreement of SSiBs spring runoff with thebucket and the observations by lowering SSiBs evap-oration from the snow surface. A physically consistentmodification to correct this feature is currently underinvestigation (as discussed in the previous section) andwill be the subject of a future paper. It should be notedthat tests of the Bare Essentials of Surface Transfer(BEST) model (Pitman et al. 1991) with the Russianstations of RO show a similar sensitivity of simulatedsnow depth to changes in downward longwave radia-tion (A. Pitman 1996, personal communication). Dur-ing the summer, SSiBs excess runoff is caused bydrainage from the lowest layer rather than surface run-off from the upper soil layer as W1 is always well belowsaturation. However, the differences of SSiBs runoffsimulations could be due not only to an inconsistentparameterization, but also to inconsistent soil param-eters. It seems plausible that SSiBs runoff discrep-


ancies could be corrected by merely varying, within arealistic range of values, its soil parameters not pre-scribed by observations (Xue et al. 1996a). Moreover,SSiBs evapotranspiration calculations are sensitive tochanges in LAI, and LAI was not a measured quantityat Valdai (GCM values for grassland vegetation wereused). We can match SSiBs summer rates of evapo-ration to observations (Fig. 9) by increasing LAI inthe summer to values that still fall within a realisticrange for a grassland vegetation (Xue et al. 1996b),but it is uncertain whether this is a consistent improve-ment in SSiBs simulation due to the lack of LAI ob-servations for the catchment. We also did not includeany interannual variations of LAI that may occur innature (e.g., in response to precipitation anomalies) forSSiBs simulations, which is the standard SSiB appli-cation for GCMs. As LAI could vary from wet to dryyears, some of SSiBs discrepancies seen during theyears with extreme hydrologic events may be a resultof the lack of an interannually varying LAI. This ap-plies to SSiBs other vegetation parameters as well.

This study has shown, for the first time by diag-nosing the complete water budget at seasonal and in-terannual timescales, that land surface hydrology canbe successfully simulated for a midlatitude grasslandusing only routine meteorological observations as forc-ing. We continue to build datasets in other climateregimes to test these conclusions with different cli-mates and different vegetation types. Our current ef-forts include obtaining and utilizing data from addi-tional sites in Russia, Mongolia, China, and India. Todate, the Russian data have been used to improve anumber of land-surface schemes currently used inGCMs, such as the bucket (Schlosser 1995), SSiB (Xueet al. 1996a), Meteo-France (Douville et al. 1995),BATS (Yang et al. 1997), and SiB2 (Zhang et al. 1996,manuscript submitted to J. Geophys. Res.). The Valdaidata will be used in the next PILPS effort (Phase 2d)and allow for the first time an international intercom-parison study of multiyear land-surface simulations tocompare with observations of total soil moisture, run-off, water-table depth, evaporation, and snow depththe complete water balance. Through continued effortswith data and models such as these, our understandingand parameterizations of land-surface processes can begreatly enhanced. The data used in this study are avail-able to anyone wishing to use them. Please contact theauthors.

Acknowledgments. We thank Chris Milly for valuablediscussions. This work was supported by NOAA GrantsNA36GPO311 and NA56GPO212; and NASA GrantsNCC555 and NAGW-5227; and NASA. The views ex-pressed herein are those of the authors and do not nec-essarily reflect the views of NOAA and NASA.

REFERENCESBerlyand, M. E., 1956: Forecasting and Control of the Heat Regime

Surface Air Layer. Gidrometeoizdat, 435 pp.

Berlyand, T. G., 1961: Distribution of Solar Radiation on Continents(in Russian). Gidrometeoizdat, 227 pp.

Brutsaert W. H., 1982: Evaporation into the Atmosphere. D. Reidel,299 pp.

Budyko, M. I., 1956: Heat Balance of the Earths Surface (in Rus-sian). Gidrometeoizdat, 255 pp.

Chen, T. H., and Coauthors, 1997: Cabauw experimental results fromthe Project for Intercomparison of Land-Surface Parameteriza-tion Schemes. J. Climate, 10, 11941215.

Dickinson, R. E., A. Henderson-Sellers, P. J. Kennedy, and M. F.Wilson, 1986: Biosphere-Atmosphere Transfer Scheme (BATS)for the NCAR Community Climate Model. NCAR Tech. NoteNCAR/TN-2751STR, 69 pp. [Available from National Centerfor Atmospheric Research, P.O. Box 3000, Boulder, CO 80307.]

Domenico, P. A., and F. W. Schwartz, 1990: Physical and ChemicalHydrogeology. J. Wiley and Sons, 824 pp.

Douville, H., J.-F. Royer, and J.-F. Mahfouf, 1995: A new snow pa-rameterization for the French community climate model. ClimateDyn., 12, 2135.

Fedorov, S. F., 1977: A Study of the Components of Water Balancein Forest Zone of European Part of the USSR (in Russian).Gidrometeoizdat, 264 pp.

Golubev, V. S., V. V. Kokneava, and A. U. Simonenko, 1995: Resultsof atmospheric precipitation measurement with Canadian, Rus-sian, and US standard instruments (in Russian). Meteor. Hydrol.,2, 102110.

Hartmann, D. L, 1994: Global Physical Climatology. Academic Press,411 pp.

Henderson-Sellers, A., Z. L. Yang, and R. E. Dickinson, 1993: TheProject for Intercomparison of Land-Surface ParameterizationSchemes. Bull. Amer. Meteor. Soc., 74, 13351349.

Idso, S. B., 1981: A set of equations for full spectrum and 814 mmand 10.512.5 mm thermal radiation from cloudless skies. WaterResour. Res., 17, 295304.

Manabe, S., 1969: Climate and the ocean circulation, 1. The atmo-spheric circulation and the hydrology of the earths surface. Mon.Wea. Rev., 97, 739774.

McKenney, M. S., and N. J. Rosenburg, 1993: Sensitivity of somepotential evapotranspiration estimation methods to climatechange. Agric. For. Meteor., 64, 81110.

Milly, P. C. D., 1992, Potential evaporation and soil moisture ingeneral circulation models. J. Climate, 5, 209226., and P. S. Eagleson, 1982: Parameterization of moisture andheat fluxes across the land surface for use in atmospheric generalcirculation models. Dept. of Engineering, Massachusetts Instituteof Technology Rep. 279, 159 pp. [Available from MassachusettsInstitute of Technology, Cambridge, MA 02139.]

Monteith, J. L., 1961: An empirical method for estimating longwaveradiation exchanges in the British Isles. Quart. J. Roy. Meteor.Soc., 87, 171179.

Pitman, A. J., Z.-L. Yang, J. G. Cogley, and A. Henderson-Sellers,1991: Description of bare essentials of surface transfer for theBureau of Meteorology Research Centre AGCM. BMRC Res.Rep. 32, 130 pp. [Available from Bureau of Meteorology Re-search Center, GPO Box 1289K, Melbourne, Victoria 3001, Aus-tralia.]

Rind, D., C. Rosenzweig, and R. Goldberg, 1992: Modelling thehydrological cycle in assessments of climate change. Nature,358, 119122.

Robock, A., K. Ya. Vinnikov, C. A. Schlosser, N. A. Speranskaya,and Y. Xue, 1995: Use of midlatitude soil moisture and mete-orological observations to validate soil moisture simulations withbiosphere and bucket models. J. Climate, 8, 1535.

Satterlund, D. R., 1979: An improved equation for estimating long-wave radiation from the atmosphere. Water Resour. Res., 15,16491650.

Schlosser, C. A., 1995: Land-surface hydrology: Validation and in-tercomparison of multi-year off-line simulations using midlati-tude data. Ph.D. dissertation, University of Maryland, 135 pp.


[Available from University of Maryland, College Park, MD20742.]

Sellers, P. J., and J. L. Dorman; 1987: Testing the simple biospheremodel (SiB) using point micrometeorological and biophysicaldata. J. Climate Appl. Meteor., 26, 622651., Y. Mintz, Y. C. Sud, and A. Dalcher, 1986: A simple biospheremodel (SiB) for use within general circulation models. J. Atmos.Sci., 43, 505531., W. J. Shuttleworth, J. L. Dorman, A. Dalcher, and J. M. Roberts,1989: Calibrating the simple biosphere model for Amazoniantropical forest using field and remote sensing data. Part I: Av-erage calibration with field data. J. Appl. Meteor., 28, 727759., D. A. Randall, G. J. Collatz, J. A. Berry, C. B. Field, D. A.Dazlich, C. Zhang, G. D. Collelo, and L. Bounoua, 1996: Arevised land surface parameterization (SiB2) for atmosphericGCMs. Part I: Model formulation. J. Climate, 9, 676705.

Stull, R. B., 1991: An Introduction to Boundary Layer Meteorology.Kluwer Academic Publishers, 666 pp.

Sud, Y. C., and M. J. Fennessy, 1982: An observational-data basedevapotranspiration function for general circulation models. At-mos.Ocean, 20, 301316.

Vinnikov, K. Ya., and M. D. Dvorkina, 1970: Some problems ofradiation stations network planning (in Russian). Meteor. Hy-drol., 10, 9096.

, and I. B. Yeserkepova, 1991: Soil moisture: Empirical data andmodel results. J. Climate, 4, 6679., A. Robock, N. A. Speranskaya, and C. A. Schlosser, 1996:Scales of temporal and spatial variability of midlatitude soilmoisture. J. Geophys. Res., 101, 71637174.

Xue, Y., P. J. Sellers, J. L. Kinter, and J. Shukla, 1991: A simplifiedbiosphere model for global climate studies. J. Climate, 4, 345364., F. J. Zeng, and C. A. Schlosser, 1996a: SSiB and its sensitivityto soil propertiesA case study using HAPEX 3 Mobilhy data.Global Planet. Change, 13, 183194., H. Bastable, P. Dirmeyer, and P. Sellers, 1996b: Sensitivity ofsimulated climate to changes in land surface parameterizationsA study using ABRACOS data. J. Appl. Meteor., 35, 386400.

Yang, D., 1995: Accuracy of the Tretyakov precipitation gauge: Re-sults of WMO intercomparison. Hydrol. Processes, 9, 877895.

Yang, Z. L., R. E. Dickinson, A. Henderson-Sellers, and A. J. Pitman,1995: Preliminary study of spin-up processes in land surfacemodels with the first stage data of Project for Intercomparisonof Land Surface Parameterization Schemes phase 1(a). J. Geo-phys. Res., 100, 16 53316 578., , A. Robock, and K. Ya. Vinnikov, 1997: On validationof the snow sub-model of the BiosphereAtmosphere TransferScheme with Russian snow cover and meteorological observa-tional data. J. Climate, 10, 353373.

Land - surface modelling

Documents

Transcript of Land - surface modelling