Soil Moisture Prediction over the Australian Continent

Transcription

1 Meteorol. Atmos. Phys. 63, (1997) Meteorology, and Atmospheric Physics 9 Springer-Verlag 1997 Printed in Austria 1 Centre for Advanced Numerical Computation in Engineering and Science, UNSW, Sydney, Australia 2 School of Mathematics, UNSW, Sydney, Australia 3 National Resource Information Centre, Canberra, Australia 4 Division of Water Resources, CSIRO, Canberra, Australia Soil Moisture Prediction over the Australian Continent Y. Shao I, L. M. Leslie 2, R. K. Munro 3, P. Irannejad 1, W. F. Lyons 3, R. Morison 2, D. Short 4, and M. S. Wood 3 With 12 Figures Received July 3, 1996 Revised October 31, 1996 Summary This paper describes an attempt to model soil moisture over the Australian continent with an integrated system of dynamic models and a Geographic Information System (GIS) data base. A land surface scheme with improved treatment of soil hydrological processes is described. The non-linear relationships between soil hydraulic conductivity, matric potential and soil moisture are derived from the Broadbridge and White soil model. For a single location, the prediction of the scheme is in good agreement with the measurements of the Hydrological and Atmospheric Pilot Experiment (HAPEX). High resolution atmospheric and geographic data are used in soil moisture prediction over the Australian continent. The importance of reliable land surface parameters is emphasized and details are given for deriving the parameters from a GIS. Predicted soil moisture patterns over the Australian continent in summer, with a 50 km spatial resolution, are found to be closely related to the distribution of soil types, apart from isolated areas and times under the influence of precipitation. This is consistent with the notion that the Australian continent in summer is generally under water stress. In contrast, predicted soil temperatures are more closely related to radiation patterns and changes in atmospheric circulation. The simulation can provide details of soil moisture evolution both in space and time, that are very useful for studies of land use sustainability, such as plant growth modelling and soil erosion prediction. 1. Introduction Soil moisture plays a major role in atmospheric and hydrological processes, since it influences the partitioning of both surface available energy into sensible and latent heat fluxes, and of precipitation into evaportranspiration and runoff. Soil moisture is also an important parameter for ecological processes, for instance, the availability of soil water directly affects plant growth. For studies of plant growth and land use sustainability, there is an increasing demand for detailed predictions of soil moisture at large scales. To date, however, there has been little dedicated work on this problem, and soil moisture predictions are usually obtained as a by-product of climate and weather prediction models coupled with a land surface parameterization scheme. Soil moisture simulation for a continental coverage poses three major challenges. Firstly, soil moisture predictions with land surface schemes will be limited by the empirical or semi-empirical nature of the parameterizations. Opinions differ on how reliable soil moisture predictions with land surface schemes are. An

2 , 196 u Shao et al. assessment of various schemes for soil moisture simulation with prescribed atmospheric forcing data and prescribed land surface parameters for soil hydraulic properties, aerodynamic properties and vegetation characteristics for a single point has been examined in Shao et al. (1994) and related studies (Shao and Henderson-Sellers, 1996; Mahfouf et al., 1996 and Wetzel et al., 1996). Figure la and lb show soil moisture simulations for a 1.6 m soil layer and a 0.5m soil layer, respectively, from 13 schemes driven by the same atmospheric forcing data from the Hydrological and Atmospheric Pilot Experiment (HAPEX). All schemes correctly predict the r,t] 220 ~, ~, I20 (a) 20 ' ' '~ m (b) ', 9~.--Q HAPEX / 9 J Time (Day) a I I I I O---OliAPEX... ::'"<:%:7...., ] F I, I, I ~ I! Time (Day) b Fig. I, Annual evolution of soil water simulated with 13 land surface schemes for HAPEX compared with observations. The simulation was carried out with observed atmospheric forcing data and pi%scribed land surface parameters (a) Soil water in the top 0.5 m soil layer and (b) in the 1.6 m soil layer basic annual pattern of soil moisture evolution, despite the large discrepancies among the schemes and between the predictions and observations (e.g., about 100ram for the 1.6m soil level). However, these schemes have a profoundly different partitioning of precipitation into runoff, drainage and evapotranspiration, as illustrated in Fig. 2. These differences can mainly be attributed to the different treatment of soil hydrological processes in the schemes. Secondly, as soil moisture evolution involves interactions between the atmosphere, soil, and vegetation, land surface schemes are usually complex. The prediction of soil moisture depends critically on the input parameters that describe soil hydrological properties, surface aerodynamic properties and vegetation features (e.g., leaf area index). Finally, the interactions between the land surface and the atmosphere involve complex feedback processes which are not yet well understood, but are known to have a significant impact on the climate variability (e.g., Shukla and Mintz, 1982; Polcher and Laval, 1994). In the case of soil moisture simulation, it appears that the uncertainties in the choice of land surface parameters and in the lower boundary condition of the soil layer exceeds those arising from the atmospheric data. This is illustrated with the sensitivity tests presented in Section 5. As will be shown, soil moisture evolution is sensitive to the choice of soil types (and thus the soil hydrological parameters) but is not as sensitive to the accuracy of atmospheric forcing data. The lower boundary issue has been addressed by Zhang et al. (1996). In current general circulation models, the land surface parameters contain significant uncertainties and the lower boundary is crudely treated. Therefore, it is likely that soil moisture predictions from current general circulation models are not sufficiently accurate, to facilitate meaningful analysis of land surface processes. Our intention in this study is to provide a simulation of soil moisture for the Australian continent. To this end there are three major tasks; the first one being the development of a new land surface scheme with an improved treatment of surface soil hydrology. The second task is to establish a set of up-to-date parameters for the land surface, including soil and vegetation over the Australian continent on the basis of GIS data,

3 Soil Moisture Prediction over the Australian Continent i i i i O model 1 9 model 3 model 2 9 Ill model 4 ALS1S 9 P-M estimated.,= 200 model 5 model 7 model 6 9 model model 9 9 model 10 9 model model 12 I I I [ I Evaporation (mm) model Fig. 2. Comparison of annual total runoff plus drainage against evaporation (mm) for 14 different schemes for HA- PEX, including ALSIS. The evaporation estimated from the Penman-Monteith equation is also shown and the third task is to couple the land surface scheme with an atmospheric model for the fourdimensional assimilation of soil moisture. As the emphasis of this study is not placed on the feedback process between the atmosphere and the land surface, a loosely linked system of an atmosphere model, a land surface scheme and a GIS database is used. The soil moisture is calculated offline with the Atmospheric Land Surface Interaction Scheme (ALSIS) driven by the output of the new University of New South Wales (UNSW) High-resolution Limited-area Atmospheric Model (HLAM) which itself has a simple land surface scheme. This allows a better examination of aspects of the parameterization of scheme and land surface parameters, without excessive concern over the complicated feedback processes between the atmosphere and the land surface. This approach is justifiable as soil moisture is not particularly sensitive to all atmospheric parameters, apart from precipitation (Section 5). The atmosphere-land surface feedback processes will be examined later in a companion paper. In this study, a brief description of ALSIS (Section 2) and HLAM (Section 3) is presented. In Section 4, the land surface parameters and the GIS database are described. The performance of ALSIS verified for a single point with HAPEX data and the sensitivity tests are discussed in Section 5. The results of soil moisture prediction over the Australian continent are described also in Section 5, and the conclusions of the paper are summarized in Section Atmosphere-Land Surface Interaction Scheme (ALSIS) The last decade has been a rapid development of sophisticated land surface schemes for atmospheric, hydrological, and ecological modeling (e.g., Dickinson et al., 1992; Sellers et al., 1986; Noilhan and Planton, 1989; Liang et al., 1994; Wetzel and Boone, 1995). A land surface scheme is composed of three major components: bare soil transfer processes, vegetation canopy transfer processes, and soil thermal and hydrological processes. Almost all land surface schemes are based on the one-dimensional conservation equations for temperature and soil moisture OT Ot 10G C Oz O0 Oq Ot Oz Sw (1) (2)

4 198 Y. Shao et al. where T is soil temperature, C is volumetric soil heat capacity, 0 is volumetric soil water content, Sw is a sink term which includes runoff and transpiration (it has been assumed that the temperature sink term is zero), G is soil heat flux, and q is soil water flux. A land surface scheme is the algorithm required to solve this system for a particular soil layer configuration. The parameterizations occur in the boundary conditions, in soil hydraulic and thermal properties, and in the treatment of the sink terms. The upper boundary conditions at the atmosphere and land surface interface include sensible and latent heat fluxes. In most land surface schemes, there is little conceptual difference in the formulation for atmospheric transfer, such as the calculation of sensible and latent heat flux. There is also little difference in the treatment of canopy (the 'big leaf' assumption), with a possible exception of the Simple Biosphere Model (Sellers et al., 1986). However, the treatment of soil hydrological processes, which can be reflected in the number of soil layers, can be significantly different. Depending on the number of computational soil layers, the schemes can be grouped into bucket-type single-layer schemes (Manabe, 1969), force-restore two-layer schemes (Noilhan and Planton, 1989) and diffusion-type multilayer schemes (Wetzel and Boone, 1995). Most schemes have less than three soil layers, as they are designed mainly for use in general circulation models where the demand on computational efficiency is important. Clearly, land surface schemes with a small number of soil layers represent soil moisture distribution poorly, and have shortcomings in the treatment of the functioning of plant roots and evaporation from bare surfaces. Although these schemes might be adequate for global climate models, for ecological modelling more soil layers are required. One of the major new ingredients of ALSIS is the improved treatment of soil surface hydrology. In ALSIS, the one dimensional (vertical) Richards' equation is used directly to describe the evolution of soil moisture. The parameters suggested by White and Broadbridge (1988) are used to characterize the hydraulic properties of various soils. The forms of their hydraulic functions have the unique property of making solutions of both differential and finite difference forms of Eq. (2) determinate under all conditions, including saturated soil and completely dry soil. This eliminates the numerical failure that have previously made routine numerical solution impracticable. Numerical speed is greater than that of a generalization of the Green-Ampt (1911) model with infinitely sharp wetting fronts (Short et al., 1995). This is due to a combination of determinacy and numerical strategy (e.g., Redinger et al., 1984; Ross, 1990), whereby mixed dependent variables are used directly in a Newton-Raphson solution scheme. This feature allows ALSIS to incorporate as many soil layers as required to provide a better vertical resolution of soil moisture and better treatment of heterogeneity (in the vertical) of soil hydraulic properties. This flexibility in choosing the number of soil layers also facilitates a more effective treatment of root activities. In ALSIS, the land surface is divided into areas of bare soil and areas covered by different types of vegetation. The energy transfer processes over bare soil surfaces are described using aerodynamic resistance laws, while the description of the canopy transfer processes is based on studies as summarized in Raupach (1988). A detailed description of ALSIS can be found in Irannejad and Shao (1996). This paper will concentrate on the treatment of the surface hydrology and the functioning of vegetation roots. Soil temperature and soil moisture are simulated with finite difference solutions of (1) and (2). Using a simple flux-gradient relationship for G, the simulation of soil temperature for different soil layers becomes the I-D diffusion equation or 0 Or Ot - Oz Dr Oz (3) where Dr is the thermal diffusivity which is a function of soil texture and soil moisture. Implicit numerical methods, such as the Crank-Nicholson scheme, are effective in solving this type of equation. Vertical water flux in an unsaturated soil, q, obeys Darcy's law q= K(1 0~-z~ ) 00 with D -- - =_K-D-- Oz (4)

5 Soil Moisture Prediction over the Australian Continent 199 where K is hydraulic conductivity, ~ is the matric potential of the soil water, z is soil depth and D is hydraulic diffusivity. A combination of Darcy's law and the soil moisture conservation Eq. (2) gives the Richards' equation Or- & -&z -sw (5) where Sw is a sink term to be parameterized, which takes into account the effect of runoff and vegetation transpiration on soil moisture variation. The Richards' equation is a highly nonlinear convection-diffusion equation owing to the dependency of K, D and 0 on ~b. Redinger et al. (1984) and Campbell (1985) applied the Kirchhoff transformation to the diffusive term of Richards' equation ( ou) 00_ 0 K-~ -sw (6) Ot Oz with U = Xd'(2 = DdO (7) I --oo l 0 ~ We follow Ross (1990) and Ross and Bristow (1990) in using the Kirchhoff transform with the intention of minimizing problems associated with non-linearity. The solution to the Richards' equation requires closure relationships between hydraulic conductivity, soil water diffusivity, matric potential and soil water content. These relationships depend on the morphology of soil pores, with the average pore size being an important indicator for soil types. Most of the land surface schemes use the Brooks-Corey model (Brooks and Corey, 1966), a simple power law relationship in its original form (e.g., Liang et al., 1994) or its Clapp and Hornberger (1978) modification (see also Cosby et al., 1984). The problem with the Brooks-Corey model is that it breaks down when soil moisture approaches saturation, making it inappropriate for infiltration representation. The model of Green-Ampt (Green and Ampt, 1911) could be used to treat infiltration independently. However, its assumption of an infinitely sharp wetting front cannot approximate realistic spatial distribution of soil water, except in coarse-texture soils. Further, it may also misrepresent significantly the infiltration flux at the surface, as White and Sully (1987) showed for Yolo light clay. The closure relationships used in ALSIS are based on the Broadbridge-White soil model. Broadbridge and White (1988) presented an analytical solution to the nonlinear diffusionconvection model in uniform soils, based on the Darcy-Buckingham approach to unsaturated water flow. The model gives simple functional forms for D and K with a single free parameter(c) and readily measurable soil hydraulic properties: K = Kr +AK0 2 C- 1 C-O (8) D=h 0 O (9) with h l) + B c(c- 1) 4(c- 1)+ 2B where A0 = Os - Or with 0, being the saturation water content and Or the residual (air dry) water content. (9 = (0-0r)/A0 is the relative water content, S = S(G, 0,.) is sorptivity, B = , and AK = Ks - Kr with K~ = K(G) and K,. = K(O,.) which is negligibly small. When Kr = 0, is = _11 C- 0 < 7 l 0/ with the capillary length scale A~ as h S 2 A~ _ (11) c(c- 1) zx0k The functional relationships ~b(0) and K(O) for different values of C and As are as shown in Fig. 3. The matric soil water potential, ~, and hydraulic conductivity, K can be respectively scaled with A, and K, to provide dimensionless functions with a parameter C ~.- ~b 1-(3 ~-C-I1 n C-O (12) e e(c- 1) K (c- 1)e 2 K, = -- = (13) Ks C-O

6 200 Y. Shao et al. 10 ~ 10 ~ 10 a E, ~ 10 D 10 -~ ~,=default ' IE 10 0 i ' * ' L " I " "'~~~--.: ::-.:_...~_...522~.32::_=_ = ::..::_::.:?.::... : = C=default... 1, , , , I 10 "2 10 -~ 10-" 7"--" 10 ~ r,/) glo V -I~ : : C =default ,1 9 I, I, I, I O(rn m-) Fig. 3. The effect of C parameter and that of As on r and K(O). Identify soil moisture Q drawn in t-axis with 0 as defined in the test The dimensionless soil water diffusivity is D C(C- i) D,--D -(C_0)2 (14) with (S/AO) Dr=hc(c_I)-A,t, and hs 2 ts - C(C-1)K~ where ts is the capillary time scale. Based on the choice of C, the Broadbridge- White model corresponds at one extreme, to the weakly nonlinear Burger's model and at the other extreme to a highly nonlinear Green-Ampt-like model. In the Broadbridge-White model soil water diffusivity remains finite as the soil becomes either very dry or saturated. Furthermore, the hydraulic functions can be scaled across all soils described by the model9 This guarantees a priori the numerical convergence of the finite difference solutions of Richards' Equation (Short et al., 1993, 1995; Dawes and Short, 1993). Because S and Ks are routinely measured, the only problem in applying the model is estimating the nonlinearity parameter C. White the Broadbridge (1988) presented different methods for estimating C. They also showed that the Broadbridge-White model satisfactorily describes the evolution of the soil water content profile and the surface soil water potential observed in field and laboratory experiments. Further, the soil water travelling wave approximation agreed with the observations for longer infiltration times.

7 Soil Moisture Prediction over the Austtaliart Continent The High-resolution Limited Area Atmospheric Model (HLAM) At the University of New South Wales (UNSW) there is a large research effort being directed at the development of a high resolution limited-area numerical weather prediction model (HLAM). The model has several versions, but the one used in this study is as documented by Leslie and Purser (1991). It has great computational economy in terms of both storage requirements and algorithm efficiency. It is a two-time-level scheme comprising a semi-lagrangian advection step followed by a number (usually 4 or 5) of adjustments steps. The adjustment steps use the forward-backward scheme (Mesinger, 1977). The temporal differencing is formally secondorder, and the interpolations in the semi-lagrangian step use bi-cubic splines. The model features are summarized in Table 1. The model has been tested extensively in both research and operational modes (Leslie and Skinner, 1994; Leslie and Purser, 1995). Standard statistical evaluation, averaged over 30 stations in the Murray-Darling Basin, the major agricultural region of Australia, has shown that the model performance is very good: for nearsurface air temperature predictions, the rms error is 2.1 K with a mean absolute error of 1.7 K; for near-surface wind speed the rms error is approximately 3 ms -t. The HLAM also has been used extensively in high resolution modelling of weather situations such as heavy rain events (Leslie and Speer, 1997), fire weather (Speer and Leslie, 1996) and tropical cyclone track prediction (Leslie and Table 1. Features of the High Resolution Limited Area Model Mode[ Feature Horizontal Resolution No. of Vertical Levels Numerical Scheme Analysis Scheme Initialization Orography Boundary Layer Scheme Radiation Scheme Convective Scheme Sea-Surface Temp HLAM 150 to 10kin 31 Split semi-lagrangian 4-D 6-hourly cycled statistical interpolation Dynamic 5' x 5' Mellor-Yamada (2.5) Fels-Schwarzkopf Modified Kuo Weekly Average LeMarshall, 1996), as well as being run routinely twice-daily at the Sydney Regional Forecast Centre. In this application, HLAM is run continuously over the Australian region at 20 km horizontal resolution and 31 levels in vertical. In order to resolve the boundary layer, there are 10 levels from 850 hpa to the surface, with the lowest level at 1 m. The time covered by the simulations is the 60 day period January l, 1996 to February 29, The HLAM derived its initial conditions and boundary conditions for this period from the Australian Bureau of Meteorology's general circulation model. 4. Geographic Information Database The land surface parameters required for soil moisture simulation are derived from the most recent GIS data available for the Australian continent, including soil types, vegetation and land use. The parameters required by ALSIS are as listed in Table 2, The parameters are derived from improved geographic data based on the Atlas of Australian Resources Volumes 1, 3 and 6 (1980, 1986, 1988), which resulted from many years of effort by a number of researchers. The spatial resolution of the data is 5 5 km, and the quality of the data has been significantly enhanced in recent years. This paper discusses the derivation of the important soil and vegetation parameters. According to the Atlas of Australian Resources based on many years of survey, Vol 1 (1980), Australian soils can be classified into 28 soil classes, with 21% being shallow permeable sandy soil (Cfl), 17% deep massive earths (Bb4), 11.2% cracking clay soils with low permeability when wet (Cbl), and 11% shallow loam soil (Cf3). Other relatively important soils are Cal (sandy soil, 8.4%), Cd2 and Ccl (duplex soils, 6.7%, 5.5%, respectively) and Be2 (calcareous earth, 5.4%); the rest of the soil types occupy 13.8%. For each soil class, there is a qualitative description of the soil properties and associated landforms. Soil hydrological parameters are derived from these qualitative data. Table 3 lists the hydraulic parameters used in this work for the Australian continent, where the 11 soil classes correspond approximately to the USDA soit texture classes ranging from sand to

8 202 Y. Shao et al. Table 2. List of Parameters Required by ALSIS Symbol Physical Meaning Dimension Ks O~ On C ol Ow Dhs OZ s L h~ LAI R st,min pr(z) O~ v Saturated Hydraulic Conductivity Saturated Volumetric Water Content Air Dry Volumetric Water Content Macroscopic Capillary Length Scale Soil Hydraulic Characteristic Parameter Volumetric Water Content at Field Capacity Volumetric Water Content at Wilting Point Heat Diffusivity for Dry Soil Soil Surface Albedo Fraction of Vegetation Cover Height of Vegetation Leaf Area Index Minimum Vegetation Stomatal Resistance Root Fraction in Different Soils Vegetation Aibedo ms-1 m3m-3 m3m-3 m m3m-3 m3m-3 m2s-1 m sm -1 Table 3. Soil Hydraulic Parameters for the Australian Continent. Note that fly and rlw are Normalised Field Capacity and Wilting Point with Os, Respectively Soil Keys Ks Os Or ~ C rlf 7lw Bcl, Cal, Cdl, Cfl, Cf e Cf e Ba3, Bb e A1, Bc e Bal, Bbl, Cd e Ba2, Bb e A2, Bb4, Bb e Cf e Bdl, Cbl, Cb e Bd2, Ccl, Cel 1.447e Bd3, Cf e clay. The distribution of soil hydrological parameters Ks, Os- Or, A and C over the Australian continent are shown in Fig. 4a, 4b, 4c and 4d, respectively. Among the five soil hydrological parameters, there exists a reasonable understanding of Ks, Os and Or, supported by experimental data such as those reported by Clapp and Hornberger (1978) Cossby et al. (1984) and Wetzel and Chang (1986). The estimation of A and C is more difficult. According to White and Broadbridge (1988), A can be estimated by h S 2 A = C(C - 1) (0s - Or)K, (15) where S is the sorptivity and h(c) ~ C(C-1)[Tr(C-1) + B]/[4(C-1) + 2B] with B = The quantities S and Ks can be measured routinely. White and Broadbridge (1988) also discussed three different techniques of estimating C from laboratory and field measurements, and made estimations of A and C for several Australian soils. Additional data are published in subsequent studies of White (1988) and white and Sully (1988). However, the estimation of these parameters cannot be made on the basis of experimental data, and subjective choices must be made. A good understanding of the impact of C on soil moisture simulation can be obtained from numerical tests, as shown in Short et al. (1995), and these tests indicate that it is plausible to assume C between and 2. In the present study, subjective choice of C and )~ were made for some soils after a careful examination of the qualitative description for

9 Soil Moisture Prediction over the Australian Continent 203 3O OO E E (a/ ~ LID ~ (b) """ "" ".% 9 Observations] w J L p, f i i, H Time, Julian day l Fig. 5. Comparison of simulated soil moisture with ALSIS with HAPEX observations for the 0.5 m (a), top lm (b) and top 1.6 m (c) all soil classes. Currently, a major effort of the authors is being devoted to the improvement of the data base of soil hydrological parameters. Vegetation influences the surface energy budget, evapotranspiration and surface aerodynamic properties, such as roughness. The vegetation data sets provide a range of parameters such as vegetation height, fractional vegetation cover, leaf area index (LAD, minimum vegetation stomatal resistance, vegetation albedo and root distribution. The source of vegetation data is the Atlas of Australian Resources, vol. 6 (1988) complemented with a data set for land use. In this data set, vegetation was divided into 35 classes according to height, density and number of canopy layers. Among the 35 vegetation types, the most extensive vegetation cover is tall shrublands in its sparse open form ($2 and S1, 31.5%). Low woodlands (L2) and low open woodlands (L1) occupy nearly 27%, while other medium and short vegetation (M2, M1, Z1, G2 and G3) covers collectively about 22% of the continent. From the vegetation data base, a reasonable estimation can be made for quantities such as vegetation height, fraction vegetation cover, vegetation albedo and minimum vegetation stomatal resistance. It is worthwhile to mention that the vegetation data set is being continuously improved. A new vegetation data set is being compiled on the basis of an amalgamation of 3 major existing data sets in Australia. Data on forest and woodland is being derived from the National Forest Inventories Forest and Woodland Database, using a Specht (1970) classification on forest type, dominant overstory cover and height and dominant understory cover the height. Data on the 'active agriculture' regions are being derived from the variance within a composite 3 year ( ) AVHRR image. The remaining areas of vegetation for Australia are being derived from Carnahan (1976). The new data base will be used in a similar study after testing. Plant root systems differ according to species, age and various environmental factors such as soil nutrient, texture, compactness, salinity, frequency and amount of precipitation (o1" irrigation) and position of water table (Gerwitz and Page, 1974; Cullen et al., 1972). However, for soil moisture simulation on large areas, it is not possible to take all these factors into consideration in calculating the root fraction in the soil layers. Studies (Kalisz et al., 1987; Sainju and Kalisz, 1990) show that the concentration of

10 204 Y. Shao et al. roots is generally greatest in the surface soil layer and decreases with depth. The usual approach to describe the root distribution is using an exponential or a polynomial relationship between root density and soil depth (Verseghy, 1991; Abramopoulos et al., 1988). In the present work, the exponential relationship proposed by Gerwitz and Page (1974) is used R(z) = e -fz (16) where f is the parameters, corresponding to the reciprocal of the depth above which 63% of roots exist. In specifying the root distribution, the 35 vegetation classes over the Australian continent are regrouped into three general categories: trees and shrubs, tall grass (including crops) and short to medium grass. For these three categories, f is set to 0.033, 0.05 and 0.067, respectively. The root fraction for a given grid is a weighted average according to the percentage of different vegetation categories. The estimation of LAI for the simulation period draws on the remotely sensed NDVI (Normalized Difference Vegetation Inde) data. NDVI data are derived from AVHRR (Advanced Very High Radiometric Resolution) satellite records of reflective radiation in the red region ( gm) and the near infrared region ( gm) of the electromagnetic spectrum. A composite of satellite images over a two week period in February 1996 was used in this study. For major vegetation types, empirical relationships between NDVI and LAI have been previously established by McVicar et al. (1996) using field observations. For the Murray-Darling Basin, for instance, the following relationship between LAI and NDVI, determined by McVicar et al. was used 1 +NDV1 LAI = NDVI (17) This type of empirical relationship has been used to estimate the LAI for the whole continent. Figure 4e shows the distribution of LAI for the Australian continent in February Aerodynamic parameters, such as surface roughness length and zero-displacement height can be estimated from leaf area index and vegetation height, according to Raupach (1994). The distribution of surface roughness length is shown in Fig. 4f. 4.1 Lower Boundary Condition The lower boundary condition, specified either by given soil moisture content or given soil moisture flux, can significantly influence soil moisture prediction, as recently discussed in Zhang et al. (1996). There are numerous techniques in parameterizing the lower boundary condition, but none of these parameterizations will be effective unless there is sufficient knowledge of the soil hydraulic properties in deep soils. A considerable effort of our group is being devoted to derive a data set for the permeability in deep soils, but at this stage the data set is not yet complete. In the present work, a deep bottom soil layer (3 to 4 meters) is used and at the lower boundary of this deep layer, soil is always assumed to be at saturation. 5. Results 5.1 Verification of ALSIS with HAPEX Data Before applying the ALSIS scheme for an extensive area, the scheme has been verified against several single point data sets. Presented here are comparisons of the model predictions against the HAPEX data, as this data set has also been used to examine many other models (Mahfouf, 1990; Mahfouf and Noilhan, 1991 and Shao et al., 1994). HAPEX was conducted in 1986 in southern France and has been well documented by a series of papers (e.g., Goutorbe et al., 1989; Goutorbe and Tarrieu, 1991 and Goutorbe, 1991). The HAPEX data used here were compiled by Mahfouf and Noilhan (see Shao et al., 1994) from HAPEX-MOBILHY at Caumont (SAMER No. 3, 43 ~ 41 t North, 0 ~ 6 r West and a mean altitude of 113 m). Detailed information on the SAMER network and the site can be found in Goutorbe (1991). The chosen location is a soya crop field that germinates in May and is harvested at the end of September. Although HAPEX was conducted in a heterogeneous area, the immediate surroundings of Caumont can be considered uniform across several hundreds of meters. The forcing data consist of the measured values of wind speed, specific humidity and air temperature at the screen height (2 m) and precipitation, solar radiation, long wave downward

11 1.5 ' ' ~, j Soil Moisture Prediction over the Australian Continent 205 I ,.. io, I,!2,5,>... I ~/. ~..,.,,1!,,,!,!9, (51>,,(6z,) ' 0.0 ~,,...,..., i ~ E l o '\ " -(176) (~- 1.5 " i%'" ' ' ,...! ~ (241 (246) F I t,,,.i is '--~,,.,-,,,, i:, (... r,t i I trl... [~'i. i i,t,,,,i... r... 0,5 1.0 (297) '... &O,,.,, ~,,,,i... r,~, j ~... L,,,~,,,,I... t... J O, , , E).I Volumetric Water Content, m m Fig. 6. Comparison of simulated soil moisture profile (dashed) with HAPEX observations (solid) for 39 weeks, valid for individual days (numbers in parenthese) radiation and surface atmospheric pressure at half hour intervals. The validation data consists of annual changes of soil moisture and 38 days of the surface energy components. The measurements of soil moisture were conducted weekly at every 0.1 meter from the surface down to 1.6m using neutron sounding probes. Net radiation, sensible heat flux and ground heat flux are measured every 15 minutes during the Inentsive Observation Period (28 May-3 July) at Caumont. Latent heat flux is calculated as the residual term to close the surface energy balance. The accuracy of the measurements is around 10%. For detailed information about the values of soil and vegetation parameters for HAPEX-MOBILHY site see Shao et al., (1994).

12 206 Y. Shao et al. E E o 10 5 ~1.1T T' ~ 1.1u' u' t- O E o O (a) -10 Cl.m e3 " , ~ -5 ~ 1.'1 q, q (c) -10, r,,,,,,, J ~20 ~80 24o 3oo 36o Time, day T' & u',.9 q I (d) " ~ W Time, day Fig. 7. Sensitivity of soil moisture simulation to the accuracy of atmospheric forcing data. HAP- EX atmospheric forcing data are used. Temperature, specific humidity and wind speed are varied by 10 and 20% in the tests E -213 E 2o "--" 10 9 "E 13 O -113 E ~ ,., g 0 "~ o --.,e o o ~,o o,,. k t'~ 20 O ', " = -- C+0.1(C-1) o o z,,~ -0.2 : : lax o o z~ A = 1.1 O, o o 0.9 9,t 1.2 ~ , r o -e ~, zx ' ' ' O3 20 ~ ~ ' ~ ' _ ' ~ K~ 10 i ~, ~, o o a a Time, day Fig. 8. Sensitivity of soil moisture simulation to the accuracy of soil hydraulic parameters, C, A, Os, Or and K~. In the first sensitivity tests, the soil hydraulic parameters are varied by 10 and 20% (in case of (a), the variation is applied to C- 1) The 1.6 m soil layer is divided into 5 layers with a depth of 0.05, 0.15, 0.3, 0.5 and 0.6m, respectively. Following the descriptions of the experiment, the soil is considered as loam with the Broadbridge-White soil parameters set as Ks = m 2 s -1, Os = m 3 m -3, Or = 0.15m3m -3, )~s = 0.4m and C = Figure 5 compares the simulated annual cycle of water content in the top 0.5 m, top 1.0 m, and top 1.6 m of soil with observations at HAPEX- MOBILHY. Considering that the observational error is about 10%, the model simulations of soil moisture are in good agreement with the observations. A comparison of Fig. l a and Fig. lb reveals that the capability of ALSIS to correctly model the soil water is superior to most schemes examined in Shao et al. (1994). Figure 6 compares the observed soil moisture

13 Soil Moisture Prediction over the Australian Continent ~,,, ~,, ~,,,~... i c,... J... -i ,2 I ~E 0"II (8.) ~'E 0.5,,~,,~,,,_f o,~_ o :> , F Year f (c) Time, days from January 1 Fig. 9. Effect of initial conditions on soil moisture simulation. Soil moisture prediction for three successive years are shown. (a) Predicted soil moisture for layer m; (b) for layer m; (c) for layer l-l.6m. For year 2 and year 3, there is no difference in soil moisture evolution. For the first month of year 1 and year 2, there is a discrepancy profile with that simulated with ALSIS over 39 weeks. For about 28 weeks, the simulation is in excellent (e.g., day 140) or good (e.g., day 169) agreement with the observations and for the remaining 11 weeks, the agreement is unsatisfactory (e.g., day 297). There are two main reasons for this discrepancy. Firstly, vertical homogeneity was assumed for soil hydraulic properties. Although the model is capable of dealing with vertically heterogeneous soils, no information was available about deep soil layers. In reality, cultivated lands such as the site of HAPEX are subject to natural heterogeneity. Further, the top meters soil is usually ploughed, causing a substantial structural change in the surface soil, such as increased permeability. Thus, applying the same set of hydraulic parameters may cause deviations from observations in different directions for the top and the deep soil layers. Secondly, there may be errors in the observational data, such as inconsistencies in soil moisture and precipitation measurements. Examples are days 25, 331 and 338, when measured profiles show a wet surface, typical for rain events, but no significant precipitation was recorded in the atmospheric forcing data. On the other hand, profiles for days 92, 199, 252 and 297 are typical of a dry spell, despite precipitation being reported. A more detailed examination of ALSIS, including simulation of surface energy fluxes and other physical variables are reported in Irannejad and Shao (1996). From Figs. 5 and 6, it can be concluded that soil moisture simulations with ALSIS agree satisfactorily with the HAPEX observations. 5.2 Sensitivity of Soil Moisture Simulation to Atmospheric Forcing and Soil Hydraulic Properties Unless the interaction between the atmosphere and the land surface significantly changes precipitation and the surface radiation balance, the choice of land surface parameters, especially soil hydrological parameters, dominates the reliability of soil moisture simulations. This hypothesis is the basis for placing the emphasis of the current work on the land surface scheme itself and the establishment of a reliable land surface description derived from a GIS. To support this hypothesis, four sensitivity tests again with the HAPEX data set were

14 (a) (b} ~~ 013O O.OO4 O.OO5 0`ooa i i ill iihllllll i i lih illlih (c) (d) g.029 0,050 0,075 O,IO `2OO 0, , ,049 1,100 I,~00 I,~g9 I~500 i (e) H i i ih I,IF~ N ~ < Fig. 4. Distribution of some key land surface parameters over the Australian continent. (a) saturation hydraulic conductivity in 10-4ms 1; (b) maximum available soil moisture 0s - 0~ (m3m-3); (c) capillary length scale A (m); (d) the C parameter; (e) leaf area index and (f) roughness length z0 (m)

15 SoiI Moisture Prediction over the Australian Continent 209 (a) Layerl Soil Moisture (b) Layer 2 Soil Moisture u rj.o~ o.(i~ 0,0~ p,o9 o.,l2 0.1~ ,~1 0..~'. o2"r (c) Layer 4 Soil Moisture (d) Total Soil Water in Tap Im (ram) V u ~d go 120 leo ;Z~ P.~ 2~0 82O Fig. lo. Soil moistm'e distribution c~ver the Australian continent for 15 Feb (a) Predicted soil moisture in m3m -3 for layer m; (b) for layer m; (c) for layer Ore and (d) total soil water in mm for the top lm soil conducted. In these tests, a variation was made in the atmospheric forcing data with all other parameters unchanged. In Test 1, Tesl 2 and Test 3, deviations of air temperature, specific humidity and wind speed from the corresponding annual mean were increased respectively by 10% and 20%. The results of these tests are summarized in Fig. 7. The differences in soil water for the top lm soil layer between the test runs and the control run (Fig. 5) are shown in Fig. 7a, 7b and 7c. For instance, the soil difference that resulted from a 10% change in T" = T~ - Ta is less than mm (Fig. 7a) and the relative difference is less than (1.5%) with the maximum occurring in October. A 20% change in T~ resulted in a soil water difference of tess than d:4 mm (Fig. 7a) and a relative difference of less than 3%. The increase in T~' resulted in a wetter sqt[ in winter and a drier soil in summer. In Test 2, the change in q', = qa - gla resulted in a soil water difference of a similar magnitude (Fig. 7b). In comparison with Test 1, the increase in q'a resulted in a drier soil in winter and wetter soil in summer. Changing the wind speeds in Test 3 produced an even less significant impact on soil water prediction as can be seen from Fig. 7c. In Test 4, T" and U' were increased by 10% and 20% and qt was decreased by 10% and 20% (so that the effect of temperature change and humidity change temperature was maximized, rather than conceled out, as indicated by Fig. 7a and Fig. 7b). The total effect of these changes remains rather insignificant, with the soil water difference between the test run and the control run being less than 10ram, and the relative difference less than 4%. A 20% change in these atmospheric variables is a significant one, possibly larger than the errors in the atmospheric forcing data obtained from HLAM used for soil moisture simulation over the

16 210 Y. Shao et al. Australian continent. The sensitivity tests imply that unless the interaction between the atmosphere and the land surface significantly changes precipitation and incoming solar radiation (through development of clouds), the accuracy in atmospheric forcing data does not seem to significantly influence the reliability of soil moisture simulation. From this point of view, the offline prediction of soil moisture can be considered as an effective approach. However, soil moisture simulation is more sensitive to the choice of land surface parameters, in particular the soil hydraulic parameters, as shown in Fig. 8. Among the five Broadbridge- White soil parameters, namely Ks, 0s, Or, As and C, the last three are generally less well defined. In Fig. 8, 0r and As are varied by 10 to 20%, while for C, C - 1 is varied by the same amount, and soil moisture simulations were compared with the control run. The sensitivity of soil water to the choice of these parameters is obvious. For instance, a variation in C- 1 of 20% produced a :k5mm difference in soil water simulation compared with the control run, corresponding to a 2% relative error. A 20% variation in A can result in up to zkl0mm, corresponding to about 5% uncertainty. A 20% variation in Or produced an uncertainty of similar magnitude, as can be seen in Fig. 8d. The parameter with maximum impact on soil water simulation is 0s, as can be seen in Fig. 8c (20% variation of 0s results in +50 mm, corresponding to a 20% velative error). If these parameters are changed simultaneously, the total uncertainty is significantly larger than that caused by the possible inaccuracy in the atmospheric forcing. A sensitivity test of soil moisture simulations, with a number of land surface schemes, varying the wilting point (similar to Or) has been examined in Shao et al. (1994), and it was shown that nearly all schemes are very sensitive to that parameter. Sensitivity tests can also be made using other sets of soil hydrological parameters, such as that of Clapp and Hornberger (1978) and Cossby et al. (1984). These sensitivity tests imply that unless soil hydraulic parameters can be adequately determined, there is little hope of a reliable soil moisture simulation. Based on this finding, considerable effort has been made in this study to obtain the most reliable soil hydraulic parameters available. 5.3 Soil Moisture Prediction Over the Australian Continent An accurate initialization of soil moisture for the whole continent is virtually impossible. Therefore, all soils are assumed to be saturated at the beginning of the simulation. The effect of this unrealistic initialization on soil moisture simulation is tested with the HAPEX data set. Figure 9 shows the evolution of soil moisture of different layers for the first 60 days of three successive years, under the same atmospheric forcing data. For the second and the third year, the differences between the predictions are trivial for the full year. This test reveals that the effect of the unrealistic initialization lasts for about a month. Based on these test, we conclude that the soil moisture prediction after one month of simulation are largely free of the influences of the initial conditions. A continuous simulation of soil moisture over the Australian continent over a two month period from 1 January to 1 March 1996 has been performed. The depth of the soil layer is 2 m and is assumed to be vertically homogeneous (this will be modified in a future study). The 2 m soil layer is divided into five layers with depth of 0.05, 0.15, 0.3, 0.5 and 1.0m, respectively. The atmospheric model was run over the Australian continent with a 20 x 20 km horizontal resolution and 30 vertical layers. The atmospheric data are stored every 30 min and averaged to a 50 x 50 km grid for soil moisture simulation. An example of the predicted soil moisture pattern is shown in Fig. 10, where soil moisture content of layer 1 ( m), layer 2 ( m) and layer 4 (0.5-1m) are illustrated together with the total soil water in the top lm, for 10:00 UTC 15 Feb 1996 (the 46th day of the simulation). It is expected that these results are largely free of the influences of the initial conditions. The basic soil moisture pattern is typical for the Australian continent in summer. As expected, in large areas in the north-western part of Australia, including the Great Sandy Desert, Gibson Desert, Great Victoria Desert and Nullabor Plain, the soil moisture content is very low for the time of the year (late summer in the Southern Hemisphere). Although, there is a slight increase in soil moisture toward deeper layers, for a considerable soil depth, the soil

17 Soil Moisture Prediction over the Australian Continent 211 moisture content falls in the range between 0.05 to 0.25 m3m -3, with the total soil water in the top 1 m of soil being around 100ram. From the desert areas, there is a gradual increase in soil moisture both toward the east and west coasts. In large areas of the 'Channel Country' and the Murray-Darling Basin, typical values of soil moisture are around 0.25 m3m -~ in layer 1 and 2, and 0.3 m3m -3 in layer 4. Total soil water in the top lm of soil is around 250 ram. Further towards the east coast, soil moisture is over 0.3 m3m -3, under the influence of rainfall that occurred during the simulation period. The obvious feature of the spatial distribution of soil moistijre is that the spatial patterns are closely related to soil hydraulic properties, as shown in Fig. 4 (see also Table 3). For instance, the tow soil moisture content in the desert areas is characteristic of the predominant sandy soils in the region (Cf2 and C1). It is known that the region has little precipitation, and as the sandy soils have a high (saturated) hydraulic conductivity and low air dry soil moisture content, soil moisture is low for most of the time. The exception is immediately after rainfall (see also Fig. 11). In the desert areas, soil moisture is rapidly lost through drainage or strong evaporation. Soil moisture content is significantly higher toward the east coast, apart from patches of very dry areas in Queensland. For a very large area in the eastern parts of Australia, the volumetric soil moisture content is around 0.3m3m -3, as the soils in this region are predominantly sandy clay or silty clay with high values of 0r. For some of these areas, although the absolute soil moisture content is quite high when compared with that of sandy soils, the available soil moisture is not necessarily much higher, as shown in Fig. 12 (left column). The strong similarity between soil moisture patterns and soil type patterns supports the notion that the Australian continent in summer is generally under water stress. In the light of these soil moisture patterns, it is interesting also to consider the patterns of soil temperature. As shown in Fig. 11, the temperature of the first soil layer reflects weather patterns, but does not show a clear correlation with the patterns of a particular surface parameter. Soil temperature in the top soil layer varies considerably in space as can be seen in Fig. 11 a. For deeper soils, the variation of temperature is smaller due to thermal inertia, as may be expected. More importantly, Fig. 11c shows that the temperature has a weak South-North gradient. This distribution of soil temperature clearly reflects the distribution of solar radiation received at the surface. The simulation provides detailed information both in space and in time. Figure 12 shows the availability of soil moisture over the Australian continent for four successive days (16-19 Feb. 1996). For instance, in large areas of North Queensland, available soil moisture (0-0r) is very high for 18 Feb 1996, because of the rainfall caused by a tropical system. On 19 Feb 1996, rainfall in Western Australia resulted in extensive areas of high soil moisture regions. These changes in soil moisture are also consistent with the changes in soil temperature and many other atmospheric quantities. 6. Conclusions This paper described an attempt to model soil moisture patterns and their evolution over the Australian continent. The emphasis of the investigation was placed on the development of a new land surface parameterization scheme and the estimation of land surface parameters. The feedback process between the atmosphere and the land surface was not considered in the paper, but will be examined in a companion paper. The land surface scheme, ALSIS, differs from many other schemes in the treatment of surface hydrology and the numerical formulation of the scheme. The non-linear relationships between soil hydraulic conductivity, matric potential and soil moisture content are based on the Broadbridge and White (1988) soil model. The soil hydraulic parameters used to represent these relationships differ considerably from those of Clapp and Hornberger (1978), which are widely used in current land surface schemes. The scheme can accommodate as many soil layers as are required and the algorithms used in the scheme are numerically efficient. The scheme was verified against the HAPEX data with good agreement. It cannot be overstated that all land surface schemes are sensitive to the choice of soil hydrological parameters. Sensitivity tests presented in this paper and Shao et al. (1994) clearly

18 212 Y. Shao et al. (a) Layer i Temperature (K) (b) Layer 3 Temperature V {e) Layer 4 Temperature (d) Average Temperature Fig. 11. As Fig. 10, but for soil temperature. In (d) average temperature for the top lm soil is shown demonstrate that the limitation for soil moisture prediction lies in the choice of land surface parameters rather than the accuracy of atmospheric forcing data. Therefore, unless a high quality data set of land surface parameters can be derived, reliable soil moisture simulation is impossible. Consequently, the second emphasis of this study was place on the development of a suitable set of land surface parameters based on the best GIS data currently available. In comparison to GCMs, the land surface information used in this study is more detailed. This study provided a prediction of soil moisture evolution and distribution over the Australian continent in summer. Although there is not yet a comparison of the predictions through independent studies, apart from single point evaluations with observational data such as HAPEX, the results are in good agreement with expectations. The simulations showed that over the Australian continent in summer, the soil moisture pattern is closely related to the distribution of soil types. This implies that, apart from isolated areas and times under the influence of precipitation, the drying factors such as evaportranspiration and drainage are primarily responsible for soil moisture status. This phenomenon forms an interesting cor/_rast with the distribution patterns of soil temperature, which in the top soil layers are more closely linked to weather patterns and exhibit a South-North gradient in deep soil layers. This is caused by the gradient in the incoming solar radiation received at the surface. High resolution atmospheric predictions and geographic data were used in the simulation, and as a result a detailed spatial distribution and time evolution of soil moisture was obtained. This level of detail is useful for many practical purposes, such as plant growth modelling and soil erosion prediction. This paper represents the first attempt of an ongoing effort in 4-dimensional simulation of soil moisture. In addition to the feedback process

19 Soil Moisture Prediction over the Australian Continent lb D 213 J 0 9 ~d D D P O o ej D D 9 P ~o a~ ~o ~d P ~Z Nd

20 214 Y. Shao et al. between the atmosphere and the land surface, there is considerable scope for further improvement of the database used in this study. Notably, the addition of topography, soil depth, the lower boundary for soil moisture prediction and the temporal changes in vegetation cover may improve the simulation. These necessary geographic data are currently being collected. References Abramopoulos, E, Rosenzweig, C., Choudhury, B., 1988: Improved ground hydrology calculations for global climate models (GCMs): Soil water movement and evaportranspiration. J. Climate, 1, Broadbridge, E, White, I., 1988: Constant rate infiltration: a versatile nonlinear model. 1. Analytic solution. Water Res. Res, 24, Brooks, R. H., Corey, A. T., 1966: Properties of porous media affecting fluid flow. J. lrrig. Drain., Div. Am. Soc. Civ. Eng., 92(IR2), Campbell, G. S., 1985: Soil Physics with Basic. New York: Elsevier, pp Carnahan, J. A., 1976: Natural Vegetation, Atlas of Australian Resources. Second Series, Dept. of Nat. Resources, Canberra. Clapp, R. B., Homberger, G. M., 1978: Empirical equations for some soil hydraulic properties. Water Resour. Res., 14, Cosby, B. J., Hornbeger, G. M., Clapp, R. B., Ginn, T. R., 1984: A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils. Water Resour. Res., 20, Cullen, R W., Turner, A. K., Wilson, J. H., 1972: The effect of irrigation depth on root growth of some pasture species. Pl. Soil, 37, Dawes, W. R., Short, D. L., 1993: The efficient numerical solution of differential equations for coupled water and solute dynamics: The WAVES model, CSIRO Division of Water Resources, Technical Memorandum 93/18. Dickinson, R. E., Henderson-Sellers, A., Kennedy, R J., Giorgi, E, 1992: Biosphere-Atmosphere Tansfer Scheme (BATS) Version le as coupled to the NCAR Community Climate Model, NCAR Tech. Note, Natl. Cent. Atmos. Res.. Green, W. H., Ampt, G. A., 1911: Studies in soil physics, I, The flow of air and water through soils. J. Agric. Sci., 4, Gerwitz, A., Page, E. R., 1974: An empirical model to describe plant root system. J. Appl. Ecol., 11, Goutorbe, J. E, 1991: A critical assessment of the SAMER network accuracy. In: Schmugge, A. (ed.) Land Surface Evaporation. New York: Springer, pp Goutorbe, J. R, Tarrieu, C., 1991: HAPEX-MOBILHY data base. In: Schmugge, A. (ed.) Land Surface Evaporation. New York: Springer, pp Goutorbe, J. R, Noilhan, J., Cuenca, R., Valancogne, C., 1989: Soil moistm'e variations during HAPEX- MOBILHY. Ann. Geophys., 7, Irannejad, E, Shao, Y., 1996: The Atmosphere-Landsurface Interaction Scheme (ALSIS): Description and Validation. CANCES Technical Report, No. 2, University of New South Wales. Kalisz, P. J., Zimmerman, R. W., Muller, R. N., 1987: Root density, abundance, and distribution in the mixed mesophytic forest of Eastern Kentuky. Soil Sci. Soc. Am. J., 51, Leslie, L. M., Purser, R. J., 1991: High-order numerics in a three-dimensionai time-split semi-lagrangian forecast model. Mon. Wea. Rev., 119, Leslie, L. M., Skinneer, T. C. L., 1994: Numerical experiments with the West Australian summertime heat through. Wea. and Forec., 9, Leslie, L. M., Purser, R. J., 1995: Three-dimensional massconserving semi-lagrangian scheme employing forward trajectories. Mon. Wea. Rev., 123, Leslie, L. M., Speer, M. S., 1997: Coastal ridging over southeastern Australia. Int. J. Climatol. (in press). Leslie, L. M., Le Marshall, J., 1996: The impact of spatial and temporal resolution of satellite observations on tropical cyclone data assimilation. Meteorot. Atmos. Phys. 60, Liang, X., Lettenmaier, D. R, Wood, E. E, Burges, S. J., 1994: A simple hydrologically based model of land surface water and energy fluxes for general circulation models. J. Geophys. Res., 99, 14, , 428. Mahfouf, J.-E, Noilhan, J., 1991: Comparative study of various formulations of evaporation from bare soil using in situ data. J. Appl. Meteor., 30, Mahfouf, J.-E, 1990: A numerical simulation of the sin-face water budget during HAPEX-MOBILHY. Bound. Layer Meteor., 53, Mahfouf, J.-E, Ciret, C., Irannejad, R, Noilhan, J., Shao, Y., Thornton, R, Xue, Y., Yang, Z-L., 1996: Analysis of results from PILPS-RICE Workshop, III, Transpiration. Global Planet. Change, 13, Manabe, S., 1969: Climate and Ocean circulation, 1, The atmospheric circulation and the hydrology of the earth's surface. Mon. Wea. Rev., 97, McVicar, T. R., Walker, J., Jupp, D. L. B., Pierce, L. L., Byrne, G. T., Dallwitz, R., 1996: Relating AVHRR Vegetation Indices to In Situ Measurements of Leaf Area Index. CSIRO, Division of Water Resources, Technical Memorandum Mesinger, F., 1977: Forward-backward scheme and its use in a limited-area model. Contrib. Atmos. Phys., 50, Noilhan, J., Planton, S., 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117, Polcher, J., Laval, K., 1994: A statistical study of the regional impact of deforestation on climate in the LMD GCM. Climate Dynamics, 10, Ranpach, M. R., 1988: Canopy transport processes. In: Steffen, W. L., Denmead, O. T., (eds.) Flow and Transport in the Natural Environment: Advances and Applications. Springer, pp Raupach, M. R., i994: Simplified expressions for vegetation roughness length and zero-plane displacement as func-

21 Soil Moisture Prediction over the Australian Continent 215 tions of canopy height and area index. Bound.-Layer Meteor., 71, 2ll-216. Redinger, G. J., Campbell, G. S., Saxton, K. E., Papendick, R. I., 1984: Infiltration rate of slot mulches: measurement and numerical simulation. Soil Sci. Soc. Am. J., 48, Ross, R J., 1990: Efficient numerical methods for infiltration using Richards' equation. Water Resour. Res., 26, Ross, R J., Bristow, K. L., 1990: Simulating water movement in layered and gradational soils using the Kirchhoff transform. Soil Sci. Soc. Am. J., 54, 15t Sainju, U. M., Kalisz, R J., 1990: Characteristics of "coal bloom" horizons in undisturbed forest soils in eastern Kentucky. Soil Sci. Soc. Am. J., 54, Sellers, E J., Mintz, Y., Sud, Y. C., Dalcher, A., 1986: A simple biosphere model (SiB) for use within general circulation models. J. Atmos. Sci., 43, Shao, Y., Anne, R. D., Henderson-Sellers, A., Irannejad, E, Thornton, E, Liang, X., Chert, T. H., Ciret, C., Desborough, C., Balachova, O., 1994: Soil moisture simulation, A report of the RICE and PILPS Workshop, IGPO Publication Series, No 14. Shao, Y., Henderson-Sellers, A., 1996: Modelling soil moisture: a project for intercomparison of land surface parameterization schemes phase 2(b). J. Geophys. Res., 101, Short, D. L., Dawes, W. R., White, I., 1993: Guaranteeing numerical convergence and stability of numerical schemes for solving the highly nonlinear convectivediffusion equation for flow of water in soils. In: McAleer, M., Jackman, A. (eds.) Proceedings of International Congress on Modelling and Simulation. University of Western Australia, pp Short, D. L., Dawes, W. R., White, I., 1995: The practicability of using Richards' equation for general purpose soil-water dynamics models. Environment International, 21, Shukla, J., Mintz, Y., 1982: Influence of land-surface evapotranspiration on the earth's climate. Science, 215, Speer, M. S., Leslie, L. M., 1996: The Sydney Australia wildfires of January 1994: Meteorological conditions and high resolution numerical modeling experiments. Int. Journal of Wildland Fire, 6, Specht, R. L., 1970: Vegetation. In: Leeper, G. W. (ed.) The Australian Environment. Melbourne: CSIRO and Melbourne Press, Verseghy, D. L., 1991: CLASS - a Canadian land surface scheme for GCMs, I, Soil model. Int. J. Clim., 11, Wetzel, R, Chang, J. Y., 1986: Concerning the relationship between evapotanspiration and soil moisture. J. Climate AppL Meteor., 26, Wetzel, R, Boone, A., 1995: A parameterization for landatmosphere-cloud exchange: Testing a detailed process model of the partly cloudy boundary layer over heterogeneous land. J. Clim., 8, Wetzel, E, Liang, X., Irannejad, R, Boone, A., Noilhan, J., Shao, Y., Skelly, C., Xue, Y., Yang, Z-L., 1996: Modeling vadose zone liquid water fluxes: Infiltration, runoff, drainage, interflow. Global Planet. Change, 13, White, I., Broadbridge, R, 1988: Constant rate rainfall infiltration: A versatile nonlinear model, 2. Applications of Solutions. Water Resour. Res., 24, White, I., Sully, M. J., 1987: Macroscopic and microscopic capillary length and time scales from field infiltration. Water Resour. Res., 23, Zhang, L., Dawes, W. R., Hatton, T. J., 1996: Modelling hydrologic processes using a biophysically based model- Application of WAVES to HFE and HAPEX-MOBILHY. J. Hydrology, 185, Authors' address: Dr. Yaping Shao and R Irannejad, Centre for Advanced Numerical Computation in Engineering and Science, The University of New South Wales, Sydney, 2052, Australia; L. M. Leslie and R. Morison, School of Mathematics, UNSW, Sydney, Australia; R. K. Munro, W. E Lyons and M. S. Wood, National Resource Information Centre, Canberra, Australia; D. Short, Division of Water Resources, CSIRO, Canberra, Australia.