3. Model description and general experimental design

Section 2.4 discussed a theoretical relationship between the continental scale land surface energy and hydrologic cycles and the remotely forced large-scale forcing. This chapter will be devoted to discussion of models used for the work in the rest of this thesis. In section 3.1, the models used will be described. Section 3.2 discusses issues regarding interpretation of model results. Section 3.3 will present the general experimental philosophy used for the balance of this work. More complete descriptions of specific experiments are found at the beginning of Chapters 4, 5, and 6.

3.1 Model descriptions

3.1.A The General Circulation Model (GCM)

The general circulation model used for these experiments is the GEOS-1 GCM documented by Suarez and Takacs (1995). It is a grid-point GCM with 13 levels, using the discretized primitive equations to calculate temperature, zonal and meridional winds, and moisture. Horizontal grid spacing is 4° latitude by 5° longitude. The short-wave radiation is calculated using a slight modification of Harshvardhan (1987), and sub-grid scale convection is determined using the relaxed Arakawa-Schubert scheme (RAS, Moorthy and Suarez 1991). Surface fluxes are calculated using the Richardson-number dependent scheme of Louis et al. (1982); they are thus dependent on the stability of the near-surface atmosphere. To simulate ocean effects, GEOS-1 may be coupled to either an ocean model, or forced through prescribing the SSTs at mean climatological or anomalous values. The prescribed forcing method is used in the experiments discussed in this work. The GCM has been forced using SSTs from the 1979-88 period, and was found to adequately simulate the global climate (Koster and Suarez, 1994, 1995a, Suarez and Takacs 1995).

To simulate land surface forcing, GEOS-1 can also be coupled to a realistic LSM or forced with prescribed land surface boundary forcing. In the model results presented in this thesis, both methods are used, to allow for or disable hydrological land-atmosphere interactions and feedbacks.

3.1.B The Land Surface Model (LSM)

The land surface model used is the Version 1 of the Mosaic LSM of Koster and Suarez (1992, 1994, 1995a). Full documentation of its parameterization methods and numerics may be found in Koster and Suarez (1995b). Mosaic is an SVAT model which parameterizes the effects of vegetation type and sub-grid scale vegetation heterogeneity on the surface energy budget and hydrologic cycle. Nine surface types are prescribed, including ice and bare soil, each with its own parameterizations of surface resistance to evaporation, sub-surface soil hydrology, and vegetation canopy (where appropriate). There are four permanent land surface moisture reservoirs; a vegetation canopy reservoir, and surface, root zone, and long term storage soil moisture reservoirs.

Snow cover represents a fifth moisture storage reservoir when present. When snow is present, evaporation from it is assumed to take place at the potential rate. Vertical movement of moisture between soil layers is calculated using the bulk form of the diffusion equations of Richards (Clapp and Hornberger 1978). Horizontal diffusion of sub-surface soil moisture is not modeled explicitly, but moisture in excess of storage in the top soil layer is removed as runoff, after accounting for vertical diffusion and moisture removal from the soil through the various moisture pathways.

Mosaic, as implied by its name, uses a different strategy from other SVAT models to parameterize sub-grid scale vegetation-soil heterogeneity. Rather than prescribe both ground cover and vegetation canopy at a single grid square, the grid itself is subdivided into a number of tiles or 'mosaics' with homogeneous vegetation. These mosaic tiles do not interact directly with each other at the surface, but do so indirectly through area weighted averaging of the surface fluxes from the mosaic tiles for calculation of the GCM variables at the grid point. The GCM inputs to the land surface, in turn, are distributed to each land surface tile at each time step for calculation of the land surface response. The mosaic strategy has been compared to the more traditional Simple Biosphere (SiB) formulation (Sellers et al., 1986, Xue et al. 1992) with multiple-canopy grid squares, and was found to give similar results to the more computationally expensive formulation (Koster and Suarez, 1994).


3.1.C Modeling the land surface control over latent heat flux

Usually, modeling of evaporation from land surface moisture reservoirs to the atmosphere is treated similarly to a series of electrical resistances along a wire. There are a number of different 'circuits' through which moisture can travel to reach the overlying atmosphere as latent heat flux. Figure 3.1 illustrates these different pathways for the simplified version of a Simple Biosphere (SiB) model which will be used for this study. The resistance functions are derived using empirical methods from field data and theoretical considerations. Table 3.1 gives definitions for the variables used in Figure 3.1.

Table 3.1: Variable definitions for Figure 3.1. Units are given for each.




vapor pressure at reference height for flux calculation, hPa


air temperature at reference height for flux calculation, K


vapor pressure of canopy and land surface, hPa


air temperature of canopy and land surface, K


saturation vapor pressure at temperature T , hPa


bare soil resistance to evaporation, sm-1


resistance imposed by plant vascular system, s


aerodynamic resistance between land surface and vegetation canopy, sm-1


resistance imposed by soil and root systems, s


soil moisture potential in root zone, m


leaf water potential, m

Total vegetation resistance r in land surface models is taken to be the product of resistance functions for different environmental factors, and is given by:



where VPD indicates the ambient vapor pressure deficit, T is canopy temperature, and yl is the leaf water potential between the leaves and the surface and root zone soil moisture reservoirs. Usually, the predominant effect on vegetation resistance is from leaf water potential, which is a function of soil moisture content in the surface and root zones. Also, since the canopy temperature and ambient vapor pressure are directly correlated with the soil moisture and evaporation rates during the warm season, those stresses will work in the same direction as that for leaf water potential. The leaf water potential resistance function is of the form:


where yl is the leaf water potential (LWP), y1 is the LWP below which wilting of the vegetation canopy begins, and y2 is the LWP below which vegetation metabolism ceases due to insufficient soil moisture. Above y1, and below y2, changes in leaf water potential, and thus soil moisture, have no effect on evapotranspiration. Vegetation resistance is at its minimum and maximum, respectively. Figure 3.2 illustrates the change in vegetation resistance for a hypothetical vegetation canopy as a function of leaf water potential, which in turn is a function of soil moisture. The change in vegetation resistance shows an exponential decay with decreasing leaf water potential as the wilting point is approached. Note that there are strong nonlinearities in vegetation resistance around the level where wilting begins. Largest unit change in resistance per unit change in leaf water potential (and thus soil moisture) is just below the beginning of wilting.

The disabling of land surface control over evaporation involves fixing the evaporability or function to its mean seasonal cycle. This function is similar to the function used for the bucket model, originally proposed by Manabe (1969). A full description the calculation of the mean seasonal cycle of b, and aspects of disabling land-atmosphere hydrologic coupling, is found in Koster and Suarez (1995a), and can be found in Appendix B. The model climatology of used for these experiments was developed from a 20-year simulation of the LSM/GCM described here, using Reynolds mean seasonal cycle SSTs as the ocean boundary condition.

Figure 3.1: Schematic of resistance network used for each tile in the Mosaic land surface model of Koster and Suarez (1992). Symbols are defined in Table 3.1.

Figure 3.2: The vegetation resistance function F(y), as a function of the leaf water potential yl, which is itself a function of soil moisture and the vegetation type. For a specific vegetation type, while 1/F(y) is linear in y, the vegetation resistance is proportional to F(y), and thus is highly non-linear around y1, the point at which the wilting of vegetation begins. Vegetation resistance to evaporation increases rapidly as (and thus soil moisture) decrease below y1, but above y1, vegetation resistance due to soil moisture stress is at its minimum.

3.2 Model limitations

As described in section 2.3, variability on the diurnal and intraseasonal time scales suggests that a complex interplay among surface energy and hydrologic variables determines the regional climate state. It also suggests that the temporal and spatial scaling of land surface characteristics, especially soil moisture and vegetation, may be important in determining the nature and scale of feedbacks between the atmosphere and land. The next section discusses some research on the scaling of soil moisture anomalies and the implications for land-atmosphere feedbacks.

3.2.A Scales of soil moisture variability

In two landmark studies using a simple bucket land surface model (Manabe 1969) coupled to a GCM, Delworth and Manabe (DM) studied the temporal behavior of soil moisture and its effect on the near-surface atmosphere. In the first study (DM 1988), they theorized that the soil acts to integrate the white noise forcing provided by monthly precipitation anomalies and spring snow melt, thus perpetuating the effects of short term precipitation anomalies over longer periods than exist in the forcing itself. Soil moisture anomalies were damped by evaporation at a rate depending on the ratio of potential evaporation to the field capacity of the soil. The equation for the resulting time series resembles that of a first order Markov process:


where w represents soil moisture, Ep is the potential evaporation, WFC is the field capacity of the soil moisture reservoir, P is precipitation input (rainfall and snow melt), R is runoff of excess moisture, t is time, (t) indicates that a variable is a function of time, and d() indicates a differential quantity. Markov processes have time spectra characterized as 'red noise' (if time step to time step anomaly correlations are positive) or 'blue noise' (if those correlations are negative). Soil moisture has a red noise spectrum since anomalies of one sign tend to persist over extended periods.

The investigators then tested the theoretical results in coupled GCM/bucket model simulations (DM 1989). Soil moisture anomaly decay time scales were longer in winter and in colder climates generally, with higher one-month lag autocorrelations and significant persistence for up to one year; in the tropics, the time scale was on the order of one to three months. This is consistent with potential evaporation Ep having a strong surface temperature (and thus solar energy input) dependence. Since field capacity was fixed everywhere, the effect of differing soil types was ignored. They also found land-to-atmosphere feedbacks from soil moisture anomaly persistence; near-surface relative humidity, latent and sensible heat fluxes, and to a lesser extent precipitation all showed increased persistence when simulated soil moisture was allowed to interact with the overlying atmosphere. Near saturation of the soil (greater than 75% of WFC) effectively decoupled the bucket reservoir from the overlying atmosphere, since changes in the bucket contents had no effect on the evaporation rate at such moisture levels.

Examination of soil moisture data from the former Soviet Union (Vinnikov and Yeserkepova 1991) found that its temporal behavior fit the theoretical distribution of DM (1988) very well. This would seem to validate the use of the bucket model, at least in regions with similar climates and soil types as those spanning the areas studied.

In the tropics, however, there are differences between the assumptions made by DM and the known characteristics of soil moisture and precipitation. For example, DM did not parameterize sub-grid scale precipitation variability in their simulations, which would mean that they may have underestimated actual precipitation rates where precipitation actually was occurring over each grid square. In the tropics particularly, this could result in overestimated soil moisture and underestimated runoff, since convective precipitation (which is of small spatial scale) dominates, and soil porosity and land surface slope are not accounted for. Thus, there is generally too much evaporation in bucket models. Additionally, soil moisture field capacities are typically as much as 60 cm in tropical rain forests rather than the 15 cm typically used in bucket formulations. Large field capacity leads to slower decay of soil moisture anomalies, because it takes longer to remove a soil moisture surplus (or eliminate a deficit) through changes in moisture input (term P(t) in equation 3.3). Thus, soil moisture anomalies may have more time to impact the tropics than found in the coupled GCM/bucket hydrology formulation. Finally, the simulations of DM ignore the short time scale (diurnal) variability in land surface control provided by vegetation. Such fluctuations may affect the atmospheric response to the soil moisture, even if the soil moisture itself is unchanged (Koster and Suarez, 1994, 1995a).

While they must be considered, such arguments regarding the basic behavior of the land surface change only the details, rather than the fundamental results, of DM. Since the forcing for soil moisture is the difference between precipitation and evaporation (P--E), the scaling of soil moisture anomalies depends on the precipitation anomaly scaling, which particularly in convective regions like the tropics is typically of sub-grid scale, absent some sort of large scale forcing. Therefore, the land can exhibit spatial correlations beyond the theoretical spatial scales proposed by DM (1988, 1989) if there is large scale coherent atmospheric forcing. Such coherence in large scale forcing can result in coherence in the precipitation itself. This forcing would include hydrologic (precipitation) forcing, and radiative forcing due to anomalous cloud cover.

Forcing as a result of internal atmospheric variability typically would not persist long enough to create a significant feedback over a large enough region for a long enough time if the forcing were the result of internal atmospheric dynamics alone, which has a memory of a few weeks or less without persistence in lower boundary forcing. Therefore, it must be persistent boundary forcing, over either land (in particular, snow mass and soil moisture) or ocean (SSTAs) or some combination/relationship between the two, which results in temporal/spatial scales of atmospheric forcing sufficient to result in significant coherent land surface state anomalies. This issue will be addressed further in the remainder of this work, in terms of large scale SSTA and soil moisture anomaly forcings.

3.2.B Sensitivity of monsoon simulation to horizontal model resolution

A number of modeling studies have been published regarding the sensitivity of different phenomena to the horizontal resolution of a numerical model of the atmosphere (need some references here). Sperber et al. (1994) compare simulation of the Indian monsoon and East Asian monsoon in the cycle 33 ECMWF forecast model (version from late 1989) at four different spectral truncations: T106, T63, T42, and T21 (corresponding to approximately 1°x1°, 2°x2°, 3°x3°, and 6°x6° grid point resolutions, respectively). For the East Asian monsoon region, they find that for the spectral ECMWF model, T42 resolution is the minimum required to resolve the features of the East Asian monsoon advance. The best simulation of East Asian monsoon jumps, as described in the East Asian monsoon phenomenology section, is found at the highest resolution. EOF analysis of precipitation by Sperber et al. (1994) reveals that the Mei-Yu mode of Lau et al (1988) is best simulated spatially and temporally by the highest spectral resolution, but that the T42 resolution captures the essential features of both the Mei-Yu and the July dry down of China (Sperber et al. 1994, their Fig. 13, p 2479). This is true of both westerly jet jumps and precipitation jumps. An East Asian summer monsoon model climatology will be presented for the models used in this study in Chapter 4, to demonstrate the adequacy of the simulations for the current work. An East Asian summer monsoon climatology will be presented for the models used in this study in Chapter 4, to demonstrate that these characteristics are also captured by the GEOS-1 GCM coupled to the Mosaic LSM.

3.3 General experimental design

We discussed the influence of SSTAs on variability of the Asian monsoon and East Asian summer monsoon in the introduction. Further details of the relative influences of land surface and SSTAs on the variability of the modeled Asian monsoon system may be found in Lau and Bua (1998). In the subsequent chapters, we will limit ourselves to examination of the land surface portion of the hypothetical relationship between the local, fast adjusting land surface forcing and the large scale, slowly varying atmospheric circulation.

To do so, a series of sensitivity experiments is performed with a coupled LSM/GCM. Ensembles are created with four ensemble members each, with each ensemble testing sensitivity to a different aspect of the initial state. Each ensemble member is a five-month global simulation starting on May 1, with different initial conditions.

A pair of ensembles is run to test model sensitivity to a suite of model generated initial atmospheric conditions and soil moisture conditions. The results from these experiments will be discussed in Chapter 4. Additional four-member ensembles are run to test sensitivity of the Asian monsoon and East Asian monsoon variability to the diurnal cycle in vegetation control and diurnal cycle in solar radiation. Results for these ensembles are found in Chapter 5. Finally, two ensembles are run with differing initial atmospheric states, but with initial Eurasian soils set either to total wilting level for the modeled vegetation or to zero. These soil moisture perturbation experiments are discussed in Chapter 6.