Contents - Previous - Next

This is the old United Nations University website. Visit the new site at

Response to Surface Moisture Availability

General Considerations The ground hydrological processes in conjunction with the characteristics of the soil and vegetation determine the surface moisture availability, which controls the partitioning between sensible and latent heat fluxes. In moist conditions about 80% of the net radiation is transferred in the latent form, so that a change from moist to arid conditions can increase the direct heating of the atmosphere by about 100 Wm-2. This gain may of course be offset by a loss of latent heating if precipitation is affected by the lack of evaporation. The ample evidence that this occurs is discussed later.

A calculation of the effects on relative humidity of such repartitioning is instructive. If an evaporation rate of 3 mm d-1 is reduced to 1 mm d-1 by limited surface moisture availability, the effect on the tropospheric mean relative humidity of the repartitioning is about 12% per day, including an increase in sensible heat flux of 58 Wm-2. Such a large difference could affect the precipitation from an initially moist air mass within a day or less. It is interesting that the increase in saturation vapour pressure due to the warming causes nearly half the decrease in relative humidity. Mintz (1984) has discussed the summer moisture budget over the eastern United States to demonstrate the important role of local evaporation. He suggests that because surface evaporation moistens the boundary layer it can stimulate convective instability, which leads to the condensation of moisture advected from the ocean as well as that evaporated from the land.

Another way of assessing the relative roles of advected moisture and local evaporation on continental rainfall is to consider the non-divergent (on average) flow of air across a land mass at constant temperature. For precipitation to be likely over the whole land traverse, the relative humidity, and therefore also (at constant temperature) the total water content of the air (Q in g cm-2), must not change much. If we assume a decrease of Q of 0.5 g cm-2 (about 10% in relative humidity) as the largest consistent with this assumption, we may calculate the rainfall, assuming an evaporation rate of 3 mm d-1 appropriate for the moist tropics, and a mean flow V of 5 m s-1 of the moisture bearing air. Table 4 shows that for an island of 100 km horizontal scale (Lo), the prescribed moisture flux convergence leads to heavy precipitation, local evaporation playing a very minor role. However, for Lo ~ 1,000 km, a scale typical of peninsular India, local evaporation is already the larger contributor, whilst for Lo ~ 3,000 km, characteristic of Africa, the contribution from the horizontal moisture flux convergence is relatively small. This argument needs modification for areas with, on average, low-level mass convergence.

TABLE 4. Relative roles of moisture advection and evaporation

Size of landmass L0(km) Contribution of horizontal moisture convergence Vd Q/L0 (mm d-1) Evaporation (mm d-1) Rainfall (mm d-1)
100 22.0 3 25.0
1,000 2.2 3 5.2
3,000 0.6 3 3.6

Note: V = 5 m s-1; d Q = 0.5 g cm-2

GCM Experiments. Several experiments have been made with GCMs to assess the impact of variations in surface moisture availability. Generally these have used the formulation (E = b Ep), described by equations (5) and (6). Because in this scheme potential evaporation (Ep) increases as the surface dries, evaporation (E) decreases less rapidly with b than the linear dependence would suggest. Rowntree and Bolton (1983) found Ea b 1/2: for a range of b from about 0.05 to 1.

(A) Regional anomalies. There have been two model experiments with anomalously low soil moisture levels over a region similar to the Sahara, one with a cyclic channel model (Walker and Rowntree 1977), one with the MO I l-layer global model (Cunnington and Rowntree 1986). The results of these may be summarized as follows:

  1. Evaporation is decreased (from about 3 to 0 mm d-1).
  2. The surface warms (by ~ 20 K) and sensible heat flux increases (from about 30 to 120 W m-2).
  3. The lower layers (up to about 600 mb) warm, with cooling ( ~ 2 K) at higher levels.
  4. Surface pressure decreases while at 700 mb and above, geopotentials are higher with a more anticyclonic circulation.
  5. Precipitation decreases (from ~ 5 to c 1 mm d-1), so that the initial anomaly in soil moisture persists.
  6. Ascent of air above the anomaly becomes limited to a shallow layer, so that it is insufficient with dry air to give significant rainfall.

It should be noted that the radiative schemes in these experiments were relatively simple. Walker and Rowntree prescribed constant surface heating and atmospheric cooling rates so that none of the feedbacks discussed in the previous section could operate. Cunnington and Rowntree used a more complete radiative calculation but still prescribed clouds. The impact of moistening of the soil on the surface radiation budget was considerable. Averaged for days 7 to 16, the net radiation was 110 Wm-2 with initially dry soil and 164 Wm-2 with initially wet soil. With model-dependent cloud, this contrast would probably have been reduced and the persistence of the anomaly could have been less marked.

Other questions' such as the dependence of the sensitivity on geographical location and horizontal scale, can only be assessed properly by an extensive series of experiments. However, some bounds can be placed on the question of location by considering the response to a global-scale anomaly. If even with all land wet a region has little rainfall, we can expect little sensitivity there; conversely, with all land dry a region that maintains its rainfall must also be rather insensitive.

(B) Global scale anomalies. Mintz (1984) has reviewed several experiments with such anomalies. These are listed in table 5. The experiments with non-interactive soil moisture are simplest to interpret as conditions are fixed. However, long experiments with interactive soil moisture, such as those of Carson and Sangster, are useful in assessing the persistence of anomalies; they are also useful in assessing the response to b = 1, since with large initial b , b will remain greater than I (giving E = Ep) for some tens of days. Also with initial b = 0 if rainfall is substantial after, say, 30 days, the initially dry soil has clearly little effect on rainfall.

The relevance of these extreme experiments to changes in vegetation cover deserves some discussion. The condition represented by b = 1 can (with difficulty) be achieved through irrigation but so long as some rainfall occurs, b = 0 is not possible all the time. However, it may occur for a much greater fraction of the time if water is drained away quickly. Vegetation can limit the efficiency of drainage by (i) allowing the interception of rain-water by the foliage, (ii) obstructing surface runoff by grass stems, tree roots, etc., and (iii) tapping subsurface water sources through the root system. Any consequent reduction in drainage (or runoff) must be compensated by increased evaporation, though it should be noted that some of the surface runoff may be replaced by percolation and subsurface runoff.

Alternatively, the problem may be considered from the viewpoint of evaporation. With no vegetation, a condition that may occur either with arable farming or due to soil degradation or drought, only the water in a shallow layer of the soil is available for evaporation and this can dry out at the high potential evaporation rates typical of the tropics after a few days. In contrast, with deep-rooting vegetation some evaporation is likely to continue even through a dry season lasting several months.

For further discussion, see Dickinson (1980) and Lockwood and Sellers (1982). The effect of vegetation on the surface roughness, discussed by Dickinson (1980), is considered in the next section.

(C) Experiments with b = 0 (e.g. fig. 2, lower). In the three experiments with b fixed at zero, rainfall is small (< 2 mm d-1) over much of the land. However, in all three rainfall exceeds 2 mm d-1 over parts of Africa at 5 to 15N, Central and northern South America, and Asia south of a line from northwest India to Korea.

TABLE 5. GCM soil moisture experiments

Reference Model Initial b Soil moisture interactive? Averaging period
Manabe 1975 GFDL (265 km) * 3 months
  1 -  
Kurbatkin et al. 1979 GFDL (spectral) * 2 months
  0 -  
Suarez and Arakawa (see Mintz 1984) UCLA 0 - 1 month
  1 -  
Shukla and Mintz 1982 GLAS 0 - 1 month
  1 -  
Carson and Sangster 1981 MO (5-layer) 0 30 days

Note: * = Model-generated soil moisture used

These are evidently regions of moisture convergence. The wet regions are less extensive in the UCLA model and more extensive in the GLAS model, which is also wet over much of the eastern USA. Mintz (1984) attributes the differences to the boundary layer parametrizations in the models. In the UCLA model there is a variable depth, wellmixed boundary layer, which allows large horizontal convergence of water vapour that does not precipitate but subsequently mixes with the free atmosphere. Carson and Sangster's experiment with initial b = 0 gave days 21-50 mean rainfall of similar character to Shukla and Mintz's, though generally less than with initially wet soil, exceptions being South America north of the equator and parts of China and the eastern USA. Kurbatkin, Manabe, and Hahn (1979) obtained similarly located increases, while Shukla and Mintz had increases over South America at 0-15S and over South-East Asia and India. Mintz (1984) attributes the latter to a change in circulation induced by soil moisture, altering the orientation of the Dow relative to the Himalayas. The models thus suggest that in the cores of July tropical continental rain belts rainfall depends only slightly on soil moisture, moisture convergence from the oceans being dominant. Elsewhere continental evaporation is very important for rainfall.

(D) Experiments with b = 1 (e.g. fig. 2, upper). The four experiments with wet soil show considerable similarity, the UCLA model again being least similar to the others with only western Australia having < 2 mm d-1. In the other models rainfall is < 2 mm d-1 only over parts of the continents at 10-30S, much of the northern Sahara, and parts of the area from south-west Russia south to Pakistan. Sensitivity is thus relatively low in these regions in the absolute sense, though not necessarily in terms of the fraction of normal rainfall. Other areas, including much of the mid-latitude continents and peripheries of the tropical rain belts, which have large variations in rainfall between the b = 0 and b = 1 cases, are highly sensitive.

FIG. 2 Precipitation (mm/d-1) in wet-soil case (top) and dry-soil case (bottom) in experiment of Shukla and Mintz (1982). (Precipitation greater than 2 mm d-1 is shaded.)

(E) Comparison to experiments with interactive soil moisture. In the tropics b = 0 (dry soil) gives results closer to observations than does b = 1, except perhaps in the UCLA model. Comparison of Kurbatkin, Manabe, and Hahn's (1979) b = 0 case, with their control run with interactive b , shows that differences (though as already noted, mostly negative) are surprisingly small over land, particularly in low latitudes. Differences exceeding 2 mm d-1 only occurred over north-western South America, where there was an increase, and near 60N and over South-East Asia and India, where there were decreases. There are, generally, increases in rates of ascent of air at 500 mb over land, which Kurbatkin, Manabe, and Hahn (1979) attribute to decreased cloud and the lack of evaporative cooling, and less ascent and rainfall over the oceans.

Carson and Sangster (1981) found that their wet anomaly persisted much longer than the dry anomaly. With initial b = 0, quasi-equilibrium was reached before 50 days, whereas with b = 1.5 initially, this took over 200 days in high latitudes and also in a belt from the Sudan to Pakistan, where differences at days 171 to 260 still averaged over 2 mm d-1 in places. This contrast between persistence of the dry and wet cases is of course to be expected with this formulation, for which evaporation reaches half its maximum value with b - 0.25, so that the necessary changes in are very different. The same is probably true of the real world with vegetation, though not without.

Response to Variations in Surface Roughness

Dickinson (1980) (see also Webb 1975) has pointed out the effect of turbulence on the partitioning between sensible and latent heat fluxes. Dickinson argued that the "large amount of turbulence produced by a forest canopy preferentially increases the flux of sensible heat with moist air," and thus deforestation could reduce evapotranspiration. The Penman-Monteith representation of these fluxes allows a quantitative assessment of this effect. From equation (16) for To ~ 26C.


so, using (8):


Table 6 shows values of (H/LE) for some typical values of raE and rs, assuming (RN - G) of 500 W m-2 and d 'q = 5 x 10-3 (moist daytime conditions). As expected by Dickinson, for relatively dry vegetation (rs > 40 s m-1 H/LE does increase as turbulence increases (raE decreases) because the (1 + rs/raE) term is dominant.

The general condition for H/LE to increase as turbulence is easily derived from (19) to be

TABLE 6. H/LE as a function of rs, raE (for To ~ 26C, RN - G = 500 W m-2, d 'q = 5 x 103)

raE (s m-1) rs (s m-1)
  0 20 50 100 200
10 -.33 0 .50 1.33 3.00
40 .07 .20 .40 .73 1.40
100 .2 1 .28 .36 .52 .82

TABLE 7. Values of rs (s m-1) above which (MILE) increases as raE decreases (for T0 ~ 26C)

RN - G (W m-2) d 'q(10-3)
  2 6 10
200 40 120 200
400 20 60 100
800 10 30 50


Some values of this critical rs are given in table 7. Note that during the day increased turbulence may increase (RN - G) by lowering surface temperature and so increasing RN (less upward long wave flux) and decreasing G.

Typical daytime de'q in the tropics are - 8 x 10-3, whilst on a clear day (RN-G) can be — 800 W m-2 at noon, giving a critical value of 40 s m-1 from table 7. Dry forests generally have larger rs than this (e.g. Webb 1975) with 60 s m-1 a representative value for adequately watered vegetation. However, this difference is small enough to accept Webb's conclusion that over a fairly wide range of (RN - G) "the estimated evaporation is not greatly affected by changes of surface roughness, wind speed, or atmospheric thermal stability."

Variations in roughness will also affect the surface momentum flux. The effects of changes in the treatment of this in the GISS model quoted earlier (Hansen et al. 1983) suggest the need for assessment of this with a realistic treatment of surface stress.

Contents - Previous - Next