CHAPTER 10

Global Change in Water Availability

Acceleration of the Water Cycle

Global warming involves changes not only in temperature but also in the rates of evaporation and precipitation. If the concentration of greenhouse gases increases in the atmosphere, the downward flux of longwave radiation increases at the Earth’s surface, as explained in chapter 1. Thus, temperature increases at the surface, enhancing evaporation from the surface, as saturation vapor pressure increases with increasing temperature according to the Clausius-Clapeyron equation of thermodynamics. The increase in the rate of evaporation in turn results in an increase in the rate of precipitation, thereby accelerating the water cycle of the entire planet.

A first attempt to evaluate the impact of global warming on the water cycle was made by Manabe and Wetherald (1975). The model used for their study was the simple 3-D GCM with highly idealized geography that was described at the beginning of chapter 5. In their numerical experiment, the equilibrium response of climate was obtained as the difference between two long-term integrations of the model with the standard and doubled concentrations of atmospheric CO₂. Toward the end of each run, the global mean rate of precipitation in each of these integrations was identical to that of evaporation, satisfying the water balance at the Earth’s surface as well as in the atmosphere. The global mean rates of precipitation (and evaporation) were 93 cm year⁻¹ for the control run and 100 cm year⁻¹ for the CO₂-doubling run. This implies that the water cycle of the model increased by about 7.4% in response to the doubling of the atmospheric CO₂ concentration. The magnitude of this increase is larger than might be expected, when keeping in mind that the warming resulting from CO₂ doubling was similar to that resulting from a 2% increase in solar irradiance, as noted in chapter 5. Why does the water cycle intensify by 7.4% in a simulation where the warming is comparable to one in which the incoming solar radiation increases by only 2%? To try to answer this question, we will examine the heat budget of the Earth’s surface.

TABLE 10.1 Mean heat budget of the Earth’s surface
	Control	2×CO2	Change (%)
Downward radiative flux (DSX − ULX)	102.6	106.1	+3.5 (+3.4%)
Net downward solar flux (DSX)	166.0	165.3	−0.7 (−0.4%)
Net upward longwave flux (ULX)	63.5	59.3	−4.2 (−6.6%)
Upward heat flux (LH + SH)	102.6	106.1	+3.5 (+3.4%)
Latent heat flux (LH)	75.4	81.0	+5.6 (+7.4%)
Sensible heat flux (SH)	27.2	25.1	−2.1 (−7.7%)
Heat budget obtained from the simple GCM described in chapter 5, section Polar Amplification. Units are W m−2. From Manabe and Wetherald (1975).

In the model described here, the Earth’s surface has no heat capacity. Thus, its heat balance must be maintained between the net downward fluxes of solar and longwave radiation and the net upward fluxes of sensible heat and latent heat of evaporation, as indicated in table 10.1. This table shows that the net downward flux of radiation increases by 3.4% in response to the doubling of atmospheric CO₂, mainly owing to the increase in the downward flux of longwave radiation. On the other hand, the net upward heat flux also increases by an identical amount. Thus, the Earth’s surface returns all of the radiative energy it receives back to the overlying troposphere through the upward heat flux. The upward flux of latent heat through evaporation increases by as much as 7.4% (or 5.6 W m⁻²), whereas the flux of sensible heat decreases by 7.7% (or 2.1 W m⁻²). Because of the partial compensation between these changes, the upward heat flux increases by 3.4% (or 3.5 W m⁻² ), which is equal to the percentage increase in the net downward flux of radiation, thus maintaining the heat balance at the Earth’s surface. In summary, the latent heat of evaporation increases by as much as 7.4%, even though the downward flux of radiation increases by only 3.4%. Why does the latent heat flux increase so disproportionately?

According to the Clausius-Clapeyron equation, the saturation vapor pressure of air increases with increasing temperature at an accelerating pace. This implies that the atmospheric vapor pressure over a saturated surface (e.g., the oceanic surface) also increases nonlinearly as surface temperature increases linearly. Thus, it becomes much easier to remove heat from the Earth’s surface through evaporation than through the sensible heat flux as temperature increases. This is an important reason why the rate of evaporation increases by 7.4%, thereby accelerating the pace of the hydrologic cycle, whereas the sensible heat flux decreases by 7.7% (see table 10.1).

As noted in chapter 5, the mean surface temperature of the model increases by 2.9°C in response to the doubling of the atmospheric CO₂ concentration. Given that the mean rates of both evaporation and precipitation increase by 7.4%, this implies that the water cycle intensifies by 2.6% per 1°C increase in mean surface temperature. The hydrologic sensitivity of the model thus obtained may be compared with that of other models constructed more recently. Allen and Ingram (2002) estimated the average hydrologic sensitivity of the models that were used in the IPCC Third Assessment Report. They found that the global mean rate of precipitation increases by about 3.4% per 1°C increase in surface temperature. Held and Soden (2006) estimated a hydrologic sensitivity of 2% per 1°C for a subset of the coupled atmosphere-ocean models used in the IPCC Fourth Assessment Report. The hydrologic sensitivity of 2.6% per 1°C for the model used by Manabe and Wetherald (1975) happens to lie between the average sensitivities obtained from the two sets of more recent models used for these IPCC assessments.

So far, we have discussed the simulated area-averaged changes in precipitation and evaporation that accompany global warming. As shown in a review of the early studies conducted by Manabe and Wetherald (1985), for example, these changes are not uniform spatially. This is attributable in no small part to changes in the horizontal transport of water vapor by the large-scale circulation of the atmosphere. When tropospheric temperature increases in response to an increase in greenhouse gas concentration, it is expected that the absolute humidity of air would also increase, owing mainly to the increase in saturation vapor pressure (i.e., the moisture-holding capacity of air) with increasing temperature. The increase in absolute humidity in turn enhances the transport of water vapor by the atmospheric circulation. This is an important reason for the changes in the spatial distribution of the difference between precipitation and evaporation due to global warming, which thereby alters the availability of water at the continental surface.

In the remainder of this chapter, we will describe how the distributions of precipitation and evaporation change in response to the doubling and quadrupling of the atmospheric concentration of CO₂, thereby affecting the spatial patterns of river discharge and soil moisture at the continental surface. Our description of these changes will be based upon analysis of the two sets of numerical experiments conducted in the late 1990s using the coupled model described in chapter 8. The first set of numerical experiments simulates the changes in the distributions of precipitation and evaporation and associated changes in water availability (e.g., rate of river discharge and soil moisture) that could occur around the middle of the twenty-first century, when the CO₂-equivalent concentration of greenhouse gases is likely to have doubled. The second set simulates the changes in response to the quadrupling of CO₂-equivalent greenhouse gas concentration. Comparing the results from these two sets of experiments, we will attempt to identify the robust changes that are common to the two sets and elucidate the physical mechanisms that control these changes.

Numerical Experiments

The coupled atmosphere-ocean-land model used for this study consists of GCMs of the atmosphere and ocean and a simple model of the heat and water budgets over the continents. It is similar to the coupled model described in chapter 8, except that the grid size is halved from ~500 to ~250 km in order to better simulate the geographic distribution of precipitation. A water budget of the continental surface is computed at each grid cell for a simple “bucket” that has a globally constant moisture-holding capacity of 15 cm, representing the difference between the field capacity and the wilting point integrated over the root zone of the soil (Manabe, 1969). Evaporation is a function of soil moisture and potential evaporation, which is computed assuming a saturated land surface (Milly, 1992). Where the predicted water content of the bucket exceeds its capacity, any excess water is converted to runoff, which is then collected over river basins and transported to the ocean at each river mouth.

Plate 2a illustrates the geographic distribution of the annual mean precipitation obtained from the control experiment, in which the CO₂ concentration is held fixed at the standard value of 300 ppmv. For comparison, plate 2b illustrates the observed distribution compiled by Legates and Willmott (1990). Inspecting these two panels, one can see that the coupled model simulates reasonably well the large-scale distribution of precipitation. For example, the model places realistically the regions of heavy precipitation in the western tropical Pacific, tropical Africa, and the Amazon basin in South America. It also places well regions of meager precipitation not only over the subtropical oceans, but also in Australia, South Africa, the North American Great Plains, and Central Asia. Further inspection also reveals that the model substantially underestimates precipitation over the tropical oceans, probably owing to the failure of the model to resolve intense tropical storms, which preferentially generate very intense precipitation over the ocean. On the other hand, possibly as a consequence of the oceanic precipitation deficit, the model overestimates precipitation over the tropical continents.

Using this model, Wetherald and Manabe (2002) performed a set of several numerical experiments, in which the model was forced by gradually increasing the CO₂-equivalent concentration of greenhouse gases. The temporal variation of the CO₂-equivalent concentration of greenhouse gases used here follows approximately the IS92a scenario of IPCC (1992) and is shown by a solid line in figure 10.1. As this figure shows, the CO₂ concentration increases at a gradually increasing rate until 1990, after which it increases at the rate of 1% per year (compounded), doubling by the middle of the twenty-first century. The CO₂ growth curve lies in the middle range of the scenarios that were presented in the “Special Report on Emission Scenarios” constructed by the IPCC (2001). The effects of sulfate aerosols, which scatter incoming solar radiation and partially compensate the warming effect of increasing greenhouse gases, were also prescribed, based on Haywood et al. (1997). The sulfate concentration of aerosols used here was estimated and projected for the period 1865–2090—that is, the period of the numerical experiment described here.

The numerical experiment described above was repeated eight times, starting from slightly different initial conditions extracted randomly from the control run, to produce an ensemble of simulations with the same radiative forcing. An ensemble mean for the 30-year period 2035–65 was produced by averaging the model output across the eight ensemble members. This approach greatly reduces the influence of unforced interannual and interdecadal variations, which can have a substantial effect on the hydroclimate. The influence of CO₂ doubling was then estimated as the difference between the 30-year ensemble mean centered on 2050, when the prescribed CO₂ concentration has doubled, and the 100-year mean obtained from the control experiment in which the CO₂ concentration was held fixed at its standard value (1×C).

In addition to the numerical experiments described above, another experiment was conducted (Manabe et al., 2004a) in which CO₂ concentrations were prescribed to increase even further. This experimental design, originally used by Manabe and Stouffer (1993, 1994), was motivated by the work of Walker and Kasting (1992), who conjectured that the atmospheric CO₂ concentration is likely to increase by a factor of three to six in a few centuries unless the combustion of fossil fuels is reduced markedly. In this experiment, the CO₂-equivalent concentration of greenhouse gases increases at a rate of 1% per year (compounded) until it reaches four times the initial (control) value, and remains unchanged thereafter, as indicated by the broad gray line in figure 10.1. According to this scenario, the CO₂-equivalent concentration would quadruple by the early twenty-second century. The effect of anthropogenic aerosols was not included in this experiment because it is likely to be relatively small on a centennial time scale, as emission controls on sulfur dioxide are strengthened. In order to reduce the effects of unforced variability on interannual and interdecadal time scales, the model output was averaged over the 100-year period between the 200th and 300th years of the experiment. The influence of CO₂ quadrupling was then estimated as the difference between this 100-year mean and a multicentury mean from the control experiment, in which the CO₂ concentration was held fixed at the standard value.

FIGURE 10.1 Time series of CO₂ concentration (logarithmic scale) used for the ensemble of eight experiments with gradually increasing CO₂ and the control (1×C) are indicated by solid black lines. The broad gray line indicates the time series for the CO₂-quadrupling experiment (4×C). On the ordinate, 2× and 4× denotes twice and four times the standard concentration of CO₂ (300 ppmv).

In the eight-member ensemble described above, the global mean surface air temperature increased by about 2.3°C and global mean precipitation increased by 5.2% by the middle of the twenty-first century, when the CO₂ concentration has doubled. In the CO₂-quadrupling experiment, the global mean temperature increased by 5.5°C and global mean precipitation increased by 12.7% a few centuries after the beginning of the experiment, when CO₂ has already quadrupled. In both cases, the hydrologic sensitivity was about 2.3% per 1°C increase in global mean surface temperature, which is similar to the sensitivity of 2.5% per 1°C that Manabe and Wetherald (1975) obtained using the simple model described in chapter 5.

Plate 3a illustrates the geographic distribution of the ensemble mean increase in surface air temperature for the middle of the twenty-first century (i.e., the time of CO₂ doubling), which was obtained from the first set of experiments. Plate 3b illustrates the pattern of surface temperature change obtained from the CO₂-quadrupling experiment. Although the latter is more than twice as large as the former, owing mainly to the much larger positive radiative forcing in the CO₂-quadrupling experiment, the two warming patterns are very similar. Both of them resemble quite well the multimodel mean warming pattern from the models used for the IPCC Fifth Assessment Report (see figure 12.10 of Collins et al. [2013]). The basic physical processes responsible for the warming pattern were the subject of detailed analysis presented in chapter 8.

Plate 3b indicates that the magnitude of the warming due to CO₂ quadrupling increases with latitude in the Northern Hemisphere and is more than 14°C over the Arctic Ocean. With the exception of the Southern Ocean, where the warming is delayed greatly as explained in chapter 8, it is comparable in magnitude to the difference in surface temperature between the mid-Cretaceous period (approximately 100 million years ago) and the present, which Barron (1983) estimated referring to various proxy signatures of past climate change. It is therefore likely that the water cycle of the mid-Cretaceous may have been as intense as that of the 4×CO₂ world that will be described in the remainder of this chapter.

We have already discussed how the warming of the Earth’s surface affects evaporation, which, in turn, affects precipitation. Analyzing the results from the two sets of experiments described above, the large-scale distributions of the changes in evaporation and precipitation obtained from the CO₂-quadrupling experiment were found to be similar to those from the CO₂-doubling experiment, although the magnitude of the former was about twice as large as the latter. In the following section, we show how the latitudinal profile of precipitation and that of evaporation change in the CO₂-quadrupling experiment, exploring the physical mechanisms that control these changes.

The Latitudinal Profile

Figure 10.2 shows the latitudinal distribution of the zonally averaged annual mean rates of precipitation and evaporation simulated by the model. Although both precipitation and evaporation tend to be larger at low latitudes than at high latitudes, their profiles are different. For example, the zonal mean rate of precipitation is larger than that of evaporation in the tropics and from middle to high latitudes, whereas the reverse is the case in the subtropics. The difference between the two profiles is attributable mainly to the meridional transport of water vapor by large-scale circulation in the atmosphere.

FIGURE 10.2 Latitudinal profiles of the zonally averaged annual mean rates of evaporation and precipitation from the control run, in which the CO₂-equivalent concentration was held fixed at the standard value. From Wetherald and Manabe (2002).

An important factor that controls the latitudinal profiles of precipitation and evaporation at low latitudes is the Hadley circulation, with rising motion in the tropics and sinking motion in the subtropics. In the near-surface layer of the atmosphere, the trade winds carry moisture-rich air from the subtropics toward the intertropical convergence zone (ITCZ), where intense upward motion predominates and precipitation is at a maximum. In the subtropics, on the other hand, air sinks over a broad zonal belt. Because of adiabatic compression of this sinking air, relative humidity is low, enhancing evaporation from the oceanic surface. Thus, evaporation is at a maximum in the subtropics.

At middle latitudes, where extratropical cyclones are frequent, precipitation is at a maximum. These cyclones carry warm, humid air poleward and cold, dry air equatorward, yielding a net transport of water vapor from the subtropics toward the middle and high latitudes. Thus, atmospheric circulation transports water vapor from the subtropics, where evaporation exceeds precipitation, to the middle and high latitudes, where precipitation exceeds evaporation.

Changes in the latitudinal profiles of precipitation and evaporation in response to the quadrupling of the atmospheric CO₂ concentration are shown in figure 10.3. Figure 10.3a presents the latitudinal profiles of the zonally averaged annual mean evaporation rate from the control run (1×C) and the CO₂-quadrupling run (4×C). A comparison of these two profiles shows that the rate of evaporation increases at all latitudes with the increase in atmospheric CO₂ concentration. The magnitude of the increase is large in the tropics and decreases poleward, becoming small in high latitudes. Because the vapor pressure of air in contact with a wet surface (e.g., the ocean) increases with increasing surface temperature according to the Clausius-Clapeyron equation, the vertical gradient of vapor pressure above the surface also increases as long as the relative humidity of overlying air does not change substantially. This is the main reason why the increase in the rate of evaporation is largest at low latitudes, where the surface temperature is highest, and decreases with increasing latitude.

FIGURE 10.3 Latitudinal profiles of (a) zonally averaged annual mean rate of evaporation and (b) zonally averaged annual mean rate of precipitation, obtained from the control run (1×C) and CO₂-quadrupling run (4×C), and (c) differences in evaporation and precipitation between the two runs.

In response to the increase in evaporation described above, the zonal mean rate of precipitation also increases at most latitudes, as shown in figure 10.3b. The latitudinal distribution of the change in precipitation rate, however, is quite different from that of evaporation. As shown on a magnified scale in figure 10.3c, the rate of precipitation increases much more than that of evaporation in the tropics and in middle and high latitudes, whereas the reverse is the case in the subtropics. These changes are attributable mainly to the increase in the export of moisture from the subtropics toward other latitudes.

When temperature increases in the troposphere owing to global warming, absolute humidity also increases, keeping relative humidity almost unchanged, as discussed earlier. The increase in absolute humidity in turn results in an increase in the transport of tropospheric water vapor. For example, the poleward transport of water vapor by extratropical cyclones increases, thereby increasing the transport of moisture from the subtropics toward middle and high latitudes. Thus, precipitation increases more than evaporation poleward of 45° in both hemispheres, as shown in figure 10.3c. Meanwhile, the equatorward transport of water vapor by the trade winds also increases, increasing the supply of moisture toward the ITCZ, where precipitation increases much more than evaporation. On the other hand, because of the increase in the export of moisture toward both high and low latitudes, precipitation hardly changes in the subtropics, despite an increased supply of moisture through evaporation from the Earth’s surface. Held and Soden (2006) found similar hydrologic responses in their analysis of the climate change experiments used for the IPCC Fourth Assessment Report. The enhancement of the pattern of precipitation minus evaporation that they identified has been dubbed the “rich-get-richer” mechanism (Chou et al., 2009).

Because of the changes in the rates of precipitation and evaporation described above, water availability at the continental surface changes. For example, the rate of river discharge increases in middle and high latitudes and in the tropics, where precipitation increases more than evaporation. In contrast, over many arid and semiarid regions in the subtropics, soil moisture decreases substantially. Since the downward flux of longwave radiation increases in response to the increase in the concentration of greenhouse gases, the thermal energy available for evaporation increases at the Earth’s surface. On the other hand, precipitation hardly increases or decreases in much of the subtropics owing to the increased export of water vapor toward both high and low latitudes. For these reasons, it is expected that soil moisture would decrease substantially over arid and semiarid regions in the subtropics. In the remainder of this chapter, we will describe how the geographic distributions of river discharge and soil moisture change as a consequence of global warming.

River Discharge

When precipitation exceeds evaporation at the continental surface, soil moisture increases. Sooner or later, the soil becomes saturated with water and excess water runs off through rivers. The geographic pattern of annual runoff obtained from the control experiment is presented in plate 4. As expected, the mean annual runoff is usually large in those regions where the rate of precipitation exceeds that of evaporation. For example, the simulated runoff is large in tropical regions of heavy rainfall such as the Amazon basin in South America, the Congo basin in Africa, and river basins in Southeast Asia and the Indonesian islands. The rate of runoff is also large in certain midlatitude regions such as the Saint Lawrence and Columbia basins in North America, and river basins in western Europe. Although the rate of precipitation is not very large, the rate of runoff is large over northern Siberia and northern Canada, where the rate of evaporation is small mainly owing to low surface temperatures under weak solar radiation. On the other hand, runoff is small over many arid and semiarid regions of the continents, such as the Sahara, Central Asia, Great Plains, southwestern North America, much of Australia, and the Kalahari region of Africa, because of meager precipitation and intense incoming solar radiation available for evaporation.

TABLE 10.2 Observed (historical) and simulated mean annual discharges of major rivers of the world
	Mean Discharge (103 m3 s−1)a		Change (%)b
River Basin	Historical	Control	2050	4xC
High Latitude
Yukon	6.5	10.1	+21	+47
Mackenzie	9.1	8.5	+21	+40
Yenisei	18.1	12.6	+13	+24
Lena	16.9	15.1	+12	+26
Ob’	12.6	6.4	+21	+42
Subtotal	63.2	52.7	+16	+34
Middle Latitude
Rhine/Elbe/Weser/Meuse/Seine	3.9	3.1	+25	+20
Volga	8.1	5.2	+25	+59
Danube/Dnieper/Dniester/Bug	8.5	6.7	+21	+9
Columbia	5.4	6.4	+21	+47
Saint Lawrence/Ottawa/Saint Maurice/Saguenay/Outardes/Manicouagan	11.8	12.4	+6	+12
Mississippi/Red	17.9	10.2	+0	−7
Amur		9.2	−1	+3
Huang He		16.7	+0	+18
Chang	28.8	53.5	+4	+28
Zambezi		31.1	−1	+2
Paraná/Uruguay		23.5	+24	+54
Subtotal	84.4	97.5	+8	+24
Low Latitude
Amazon/Maicuru/Jari/Tapajos/Xingu	194.3	234.3	+11	+23
Orinoco	32.9	28.2	+8	+1
Ganges/Brahmaputra	33.3	48.6	+18	+49
Congo	40.2	122.3	+2	−1
Nile	2.8	49.5	−3	−18
Mekong	9.0	28.6	−6	−6
Niger		58.3	+5	+6
Subtotal	312.5	469.8	+7	+13
TOTAL	460.1	661.7	+8	+16
From Manabe et al. (2004b). a Mean discharges shown from historical data and simulated from the control (1×C), where atmospheric CO2 concentration is kept unchanged at 300 ppmv. Subtotals and totals include only those basins with historical data. b Relative changes simulated to occur from the preindustrial period to the middle of the twenty-first century, when the CO2 concentration has doubled (2050), and those in response to a quadrupling of atmospheric CO2 (4×C). The percentage change from A to B is defined as 100 × (B − A)/A.

Table 10.2 includes a comparison of historical mean and model-estimated values of annual discharge for a number of important river basins throughout the world. The model estimates were obtained from the control time integration of the model, in which the atmospheric CO₂ concentration is kept unchanged at the standard value throughout the course of the integration. In high and middle latitudes, for example, about half of the basins have modeled discharges within about 20% of the observed value. The total discharge simulated for high and middle latitudes (52,700 and 97,500 m⁻³ s⁻¹) compares reasonably well with the observed values (63,200 and 84,400 m⁻³ s⁻¹). In low latitudes, however, river discharges are overestimated, particularly in tropical Africa and Southeast Asia. These are also regions where precipitation is overestimated substantially (plate 2a and 2b). In general, however, the model reproduces reasonably well the annual discharge from many of the major rivers of the world.

The comparison of observed and modeled discharge presented here is affected by temporal sampling error in the observations, because observational records do not always cover a long enough period to provide precise estimates of climatic means. Additionally, the natural balance between runoff and evaporation is modified significantly as a result of irrigated agriculture and evaporation from artificial reservoirs. Nevertheless, both sampling error and the consequences of water resource development are small compared with the simulation errors in the basins for which comparisons are made in table 10.2.

The geographic distributions of the changes in the annual mean rate of runoff simulated for CO₂ doubling and CO₂ quadrupling are illustrated in plate 5a and b, respectively. As this figure shows, the patterns of the changes are remarkably similar between the two simulations, although the magnitude of the change in the latter case is about twice as large, as one might expect given the difference in the simulated warming. The similarity between the two patterns implies that the physical mechanisms involved are practically identical between the two simulations. It also implies that any effect of unforced variability is very small, largely because of the time averaging applied to the runoff obtained from the two experiments.

In both simulations, runoff increases in high latitudes, particularly over the northwest coast of North America, northern Europe, Siberia, and Canada. It also increases in the rainy regions of the tropics such as Brazil, the west coast of tropical Africa, Indonesia, and northern India. On the other hand, runoff decreases in many semiarid regions such as the zonal belt to the south of the Sahara, the southern part of North America, the west coast of Australia, the Mediterranean coast, and northeast China. The magnitude of the reduction, however, appears to be relatively small in absolute terms, although it may not be small in terms of percentage, as shown by Milly et al. (2008). In general, the geographic pattern of the change in runoff is similar to the multimodel average from the models used for the IPCC Fifth Assessment Report (see figure 12.24 of Collins et al. [2013]).

A notable exception occurs, however, in the Amazon basin, where runoff increases substantially in the result presented here but decreases in the multimodel mean. One can speculate that the discrepancy is largely attributable to the difference in the rate of precipitation. As indicated in Flato et al. (2013, fig. 9.4b), the multimodel mean rate of precipitation in the basin is substantially smaller than the observed rate, whereas a similar bias is not evident in plate 2a and 2b for the model presented here. In view of the close relationship between rainfall and runoff, it is likely that the rate of runoff is going to increase in the Amazon basin owing to global warming.

Averaged zonally, changes in the rate of runoff bear some similarity to changes in the difference between precipitation and evaporation. Runoff increases substantially in the tropics and also in middle and high latitudes. In contrast, the magnitude of the change in runoff is small in the subtropics. An increase in the export of water vapor from the subtropics to both higher and lower latitudes is the primary cause of these changes, as discussed earlier in this chapter.

The percentage changes in the rate of river discharge from the major rivers of the world for the CO₂-doubling and CO₂-quadrupling experiments are also shown in table 10.2. As this table indicates, the changes in the quadrupling experiment are about twice as large as in the doubling experiment. For example, the discharge from Arctic rivers such as the Mackenzie and Ob’ increases by ~20% with CO₂ doubling and ~40% with CO₂ quadrupling. The large increase in discharge from these Arctic rivers is attributable mainly to the increase in poleward transport of water vapor, as discussed earlier in this chapter. Recently, Peterson et al. (2002) analyzed the time series of the discharges from several major Arctic rivers in Siberia. They found that the total discharge from these rivers has a statistically significant positive trend, in qualitative agreement with the results presented here.

In the middle latitudes, the percentage change in discharge from European rivers such as the Volga is large. The response from these rivers is similar to those in high latitudes. The discharge from Columbia also increases as precipitation increases in the Rocky Mountains. On the other hand, the change in combined discharge of the Paraná and Uruguay rivers is relatively high, reflecting essentially a tropical response within the runoff source region for this system.

In the tropics, the discharge from the Amazon River increases by 11% and 23% in response to the CO₂ doubling and CO₂ quadrupling, respectively. Although the discharges from the Ganges-Brahmaputra and Congo increase greatly in response to the doubling and quadrupling of CO₂, they should be regarded with caution in view of the gross overestimate of the annual discharges obtained from the control experiment. Similar caution may also be applicable to the changes in river discharge from the Congo, Mekong, and Nile rivers.

The impact of climate change on river discharge has also been estimated using stand-alone models of river discharge by Alcamo et al. (1997) and Arnell (1999), and by using climate model output from Vörösmarty et al. (2000) and Arnell (2003). For example, the consensus pattern of change obtained by Arnell (2003) on the basis of data from several climate models is broadly consistent with the pattern described here, with the major exception of the Amazon River, where runoff decreases in his analysis. In sharp contrast, runoff increases substantially in the experiments described here. In view of the substantial underestimation of rainfall in the Amazon basin in many models used in Arnell’s analysis, it may be premature, however, to conclude that the discharge from the basin is going to decrease owing to global warming, as noted earlier.

Soil Moisture

A meaningful and direct comparison of modeled soil moisture with observations is not possible because of difficulties in defining the plant-available-water-holding capacity of the soil and because of the extreme heterogeneity of soil moisture, soil properties, and vegetation rooting characteristics. Nevertheless, soil moisture in this model is an excellent indicator of soil wetness. Plate 6 illustrates the distribution of annual mean soil moisture simulated by the model. The model reproduces reasonably well the large-scale features of soil wetness. For example, the regions of very low soil moisture simulated by the model approximately correspond with the major arid regions of the world: the Gobi and Great Indian deserts of Eurasia, the North American deserts, the Australian desert, the Patagonian Desert of South America, and the Sahara and Kalahari deserts of Africa. Furthermore, the model places reasonably well the semiarid regions adjacent to many of the major arid regions in Africa, Australia, and Eurasia. Although the semiarid region in the western plains of North America is simulated by the model, it extends eastward too far, particularly in the southern United States, where precipitation is underestimated substantially (compare plate 2a with 2b). On the other hand, soil moisture is large in Siberia and Canada, located in high northern latitudes where precipitation substantially exceeds the relatively meager evaporation. As expected, soil moisture is also large in heavily precipitating regions of the tropics in South America, Southeast Asia, and Africa. In summary, the model places reasonably well the locations of arid, semiarid, and wet regions of the world.

Drying in Arid and Semiarid Regions

Global warming affects not only river discharge but also soil moisture. The geographic distributions of the change in annual mean soil moisture in response to the doubling and quadrupling of the atmospheric concentration of CO₂ are illustrated in plate 7. The changes are presented in terms of percentage change relative to the control experiment. The geographic pattern of the percentage change in soil moisture due to CO₂ doubling resembles the pattern due to CO₂ quadrupling, though the latter is about twice as large as the former. As noted with regard to the change in the rate of runoff, the similarity between the two patterns implies that the basic physical mechanism involved is practically identical between the two simulations. The percentage reduction in soil moisture is relatively large in many arid and semiarid regions of the world, such as western and southern parts of Australia, southern Africa, southern Europe, northeastern China, and southwestern North America. Although the percentage reduction is also large in the southeastern United States, this result should be regarded with caution because simulated precipitation is substantially less than observed (plate 2) and simulated soil moisture (plate 6) is unrealistically small in the region.

It is encouraging that the geographic patterns of the percentage changes in annual mean soil moisture in plate 7 resemble the multimodel mean pattern of changes from Collins et al. (2013, fig. 12.23) in the IPCC Fifth Assessment Report. However, a notable exception occurs in the Amazon basin. Although soil moisture changes only slightly in this basin as indicated in plate 7, it decreases substantially in the multimodel mean. We have previously noted that the multimodel mean precipitation in this basin is much less than observed. A similar discrepancy is not evident in plate 2, which compares the simulated and observed distribution of precipitation for the coupled model discussed here. It is therefore possible that the difference in the sign of soil moisture change may be attributable to the difference in the rate of simulated precipitation in the basin. Thus we are tempted to speculate that soil moisture may increase in the Amazon basin as global warming proceeds, as it does in the present model.

The seasonal dependence of soil moisture change (%) in response to CO₂ quadrupling is shown in plate 8 for each of the standard seasons: June–July–August (JJA), September–October–November (SON), December–January–February (DJF), and March–April–May (MAM). This figure shows that soil moisture decreases in many arid and semiarid regions, particularly during the dry season. For example, the percentage reduction is pronounced in southern Australia from JJA to SON, in and around the Kalahari Desert of Africa in JJA, in southern Europe in JJA, and in southwestern North America from DJF to MAM. Although it is also large in the southeastern United States in MAM, this should be regarded with caution because of the systematic underestimation of precipitation in this region. Although not shown here, the geographic pattern of the soil moisture change that occurs in response to CO₂ doubling resembles the pattern that occurs in response to CO₂ quadrupling.

Why does soil moisture decrease in many arid and semiarid regions of the world? As we have discussed previously, the downward flux of longwave radiation increases owing to the increase in the concentration of greenhouse gases, thereby increasing the radiative energy that is potentially available for evaporation. On the other hand, the magnitude of the change in precipitation is usually small in these regions. In order to maintain the water balance of the continental surface, it is therefore necessary to reduce evaporation as a fraction of potential evaporation. Because the ratio of evaporation to potential evaporation decreases as the soil dries, a reduction in soil moisture reduces the fraction of the radiative energy used for evaporation. This is the main reason why soil moisture decreases in many arid and semiarid regions of the world. That is not so say that changes in the precipitation rate are not important. Indeed, the percentage reduction of soil moisture tends to be large in those regions where the percentage reduction of precipitation is also large. (For the geographic distribution of the percentage change in precipitation, see Collins et al. [2013, fig. 12.22] from the IPCC Fifth Assessment Report.)

In many relatively arid regions of the world, soil moisture is small partly because water vapor is exported outward by the large-scale circulation in the atmosphere, as it is in many regions in the subtropics. Because absolute humidity of air usually increases with increasing temperature, it is expected that the rate of export is likely to increase as global warming proceeds, reducing the amount of water vapor available for precipitation in these regions. This is another reason why soil moisture decreases in such relatively arid regions.

So far, we have discussed the systematic change of soil moisture that occurs on multidecadal to centennial time scales in response to a gradual increase in the atmospheric greenhouse gas concentration. The temporal variation of soil moisture at interannual and decadal time scales is depicted in figure 10.4. The figure illustrates, for both a global warming and a control run, the time series of annual mean and 20-year-running-mean soil moisture over the semiarid region in southwestern North America. The systematic reduction of annual mean soil moisture in the global warming run is often overwhelmed by large natural interannual variability. By the latter half of the twenty-second century, however, the thin gray line signifying the global warming run dips below 3 cm more frequently than the thin black line does that indicates the time series of the annual mean soil moisture obtained from the control run. (In the simple bucket model, a soil moisture value of 3 cm indicates that plant-available water is only 20% of what it would be in saturated soil.) This result implies that the frequency of drought is likely to increase during the twenty-first century as plant-available water dips below 20% of saturation in many semiarid and arid regions of the world.

FIGURE 10.4 Time series of simulated soil moisture (cm) averaged over the semiarid region in southwestern North America, which is enclosed by 20° and 38° N latitude, 88° and 114° W longitude, and coastal boundaries. Thin and thick black lines show, respectively, the time series of annual mean and 20-year-running-mean soil moisture obtained from the control integration. Thin and thick gray lines show, respectively, the time series of annual mean and 20-year-running-mean soil moisture obtained from one of the eight-member ensemble of global warming experiments described in the section Numerical Experiments. From Wetherald and Manabe (2002).

Midcontinental Summer Dryness

As plate 8a shows, soil moisture decreases in summer over extensive midcontinental regions of both North America and Eurasia in the middle and high latitudes. This is in contrast to winter, when soil moisture increases in these regions, as indicated in plate 8c. Midcontinental summer dryness has been the subject of many studies (e.g., Cubasch et al., 2001; Gregory et al., 1997; Manabe and Stouffer, 1980; Manabe and Wetherald, 1985; Manabe et al., 1992; Mitchell et al., 1990). We will now explore this topic further, referring to the study of Manabe and Wetherald (1987), which was conducted using an atmosphere/mixed-layer-ocean model as described in chapters 5 and 6.

According to their analysis, the large percentage reduction of soil moisture in summer over the northern part of Siberia and Canada around 60° N is attributable mainly to the earlier termination of the snowmelt season. As surface temperature increases at the continental surface owing to global warming, the snowmelt season ends earlier in the spring, exposing the snow-free surface (with low albedo) to intense solar radiation. This is the main reason why the absorption of solar energy at the continental surface increases markedly and makes additional energy available for evaporation in the late spring, thereby reducing soil moisture in summer.

In much of the continental regions in middle latitudes, the summer reduction of soil moisture is attributable not only to the earlier termination of the snowmelt season described above, but also to the poleward shift in the latitudinal profile of precipitation from winter to summer. When temperature increases in the troposphere, the absolute humidity of air usually increases. Thus, the poleward transport of moisture by extratropical cyclones increases, as we have discussed previously. For this reason, the rate of precipitation usually increases substantially along the midlatitude cyclone track and on its poleward flank. In contrast, on the equatorward flank of the cyclone track precipitation hardly changes or decreases slightly, as shown, for example, in figure 10.3b. Because the cyclone track and its associated rain belt shift poleward from winter to summer, particularly over the continents, a midcontinental region located in the poleward flank of the cyclone track in winter would be in its equatorward flank in summer. Thus, precipitation increases substantially in winter, whereas it often decreases slightly in summer. On the other hand, the downward flux of longwave radiation increases at the Earth’s surface owing to the increase in atmospheric CO₂ concentration, making additional energy available for evaporation. The combination of increased evaporation and a decrease in precipitation leads to a decrease in soil moisture over midcontinental regions in summer. A similar mechanism also operates in southern Europe, where the percentage reduction in soil moisture is particularly large in summer, as shown in plate 8. There is an important difference, however, because soil moisture in southern Europe decreases not only in summer but also in the other seasons, in contrast to many other regions of middle and high latitudes, where soil moisture increases in winter and early spring. Although precipitation increases in southern Europe in these seasons, the magnitude of the increase is small and is responsible for the small percentage reduction of soil moisture because it is not large enough to compensate for increased evaporation.

FIGURE 10.5 Zonal mean change in soil moisture (cm) in response to the doubling of atmospheric CO₂ concentration as a function of latitude and calendar month. Black shading indicates latitude bands where there is very little land or the land is ice-covered. From Manabe and Wetherald (1987).

In sharp contrast to summer, soil moisture increases in winter over very extensive regions in both the North American and Eurasian continents in middle and high latitudes, mainly owing to the increase in precipitation. Since temperature is very low in these regions, the rate of evaporation is small and hardly changes in winter despite the increase in surface temperature due to global warming. For these reasons, soil moisture increases over very extensive regions in winter (DJF) and spring (MAM), as shown in plate 8c and d, although it decreases slightly in southern Europe in spring.

Figure 10.5 illustrates the latitude/calendar-month distribution of the equilibrium response of zonal mean soil moisture to the doubling of the atmospheric CO₂ concentration. Although it was obtained from the atmosphere/mixed-layer-ocean model described in chapters 5 and 6, it essentially encapsulates the results that are obtained from the model presented in this section and shown in plate 8. Soil moisture decreases in summer but increases in winter in middle and high latitudes of the Northern Hemisphere. In the subtropics, soil moisture decreases during much of the year, particularly in winter and spring when precipitation decreases. Although the magnitude of the reduction is small in other seasons, it is not necessarily small when expressed in terms of percentage reduction.

Implications for the Future

If the concentration of greenhouse gases continues to increase, following a so-called “business-as-usual” scenario, the reduction of soil moisture in many arid and semiarid regions of the world is likely to become increasingly noticeable during the twenty-first century. By the latter half of the twenty-second century, the reduction of soil moisture in these regions could become very substantial and the frequency of drought is likely to increase markedly. Unfortunately, the river discharge in these regions is not likely to increase significantly, or may actually decrease as global warming proceeds. It is therefore likely that the shortage of water in these regions could become very acute during the next few centuries. In contrast, an increasingly excessive amount of water is likely to be available through river discharge in many water-rich regions in high northern latitudes and in heavily precipitating regions of the tropics, where the frequency of floods is likely to increase markedly, as Milly et al. (2002) found in a numerical experiment. The implied amplification of existing differences in water availability between water-poor and water-rich regions could present a very serious challenge to the water-resources managers of the world. For further discussion of this subject, see the short essay by Milly et al. (2008).