A simulation of the effect of climate change–induced desertification on mineral dust aerosol



[1] Vegetation and climate fields from a coupled carbon-cycle – climate model integration, which included the feedback of vegetation on climate, have been used to drive the HadAM3 AGCM incorporating the Hadley Centre mineral dust scheme in experiments to investigate future dust concentration and forcing. Comparison of 2000 with 2100 simulations shows the global annual mean atmospheric dust load increases from 4 × 104 to 1.3 × 105 mg m−2, due to the combination of desertification and climate change. The global mean radiative forcing due to dust increases from 0.04 to 0.21 Wm−2 at the top of the atmosphere and from −0.74 to −1.82 Wm−2 at the surface.

1. Introduction

[2] Mineral dust aerosol is produced by the action of wind on dry soils, primarily from the world's deserts [Pye, 1987], and is found throughout the troposphere, where it interacts directly with both shortwave and longwave radiation [Sokolik and Toon, 1999]. The global mean top of the atmosphere (ToA) radiative forcing due to dust is small due to the partial cancellation, in the mean, of areas of strong positive and negative forcing. The global mean surface forcing is always negative [Miller and Tegen, 1998; Woodward, 2001].

[3] Most GCM studies of climate change exclude feedbacks between climate and biosphere, though such feedbacks do exist: approximately half of anthropologically emitted CO2 is currently absorbed by terrestrial and marine ecosystems [Schimel et al., 1995], through mechanisms which are sensitive to CO2 concentrations [Betts et al., 1997] and to climate [Sarmiento et al., 1998; Cao and Woodward, 1998]. A pioneer study to have used a fully coupled carbon-cycle – climate model was Cox et al. [2000] (C00), which showed significant acceleration of global warming compared with uncoupled models and predicted major vegetation changes, including dieback of the Amazonian rainforest, leading to a significant increase in bare soil.

[4] The changes predicted by the C00 experiments would be likely to lead to enhanced dust production and hence increases in direct radiative forcing as well as further changes due to enhanced nutrient delivery to marine ecosystems, ice nucleation changes and climate feedback mechanisms which, together with increased dust storm frequency, could adversely impact human quality of life. The aim of this work is to investigate potential future dust loading and direct radiative forcing, using a dust scheme within an AGCM constrained by vegetation and climate fields from the C00 simulation at 2000 and 2100. This method will allow the simulations of dust fields to include the effects of carbon-cycle – climate feedbacks. These have been neglected in previous estimates of future dust load, which have varied between −63% [Mahowald and Luo, 2003] and +24% [Tegen et al., 2004], with different methods and models and different assumptions. This work also gives a first estimate of future dust forcings.

2. Experiments

[5] The climate model used for this study was based on HadAM3 [Pope et al., 2000], with a resolution of 2.5° × 3.75° × 19 levels, incorporating the Hadley Centre dust scheme as described by Woodward [2001]. In this scheme dust is produced from the bare soil fraction of each gridbox. Its radiative properties are derived from data for dust from various sources in an attempt to give values representative for the whole globe, rather than specific to one desert.

[6] The model was run for two periods of 10 years (+1 year spin-up in each case) for the years 2000 and 2100. Vegetation and climate data, such as SSTs, sea-ice and trace-gas concentrations, were obtained from 10 year mean data from the C00 transient simulation in which a fully coupled carbon-cycle – climate model (HadCM3 coupled to TRIFFID and HadOCC) had been run for the years from 1860 to 2100 using the IS92a scenario [Cox et al., 2000]. A “double radiation call” method was used, whereby the radiation code was called twice in each model timestep, the first call including dust effects, the second excluding them and progressing the model. The dust forcing was obtained from the differences between the fluxes calculated in each call. Two additional perturbation experiments were run using vegetation data for 2000 and climate data for 2100 and vice versa.

3. Results

3.1. Atmospheric Dust Load

[7] The bare soil fractions for 2000 and 2100 from C00 are shown in Figure 1. Global mean results are summarised in Table 1.

Figure 1.

Bare soil fractions simulated by coupled carbon-cycle – climate experiment.

Table 1. Summary of Experimentsa
ExperimentBare Soil Area, m2Dust Load, TgDust Optical Depth (at 550 nm)Surface Forcing Due to Dust, Wm−2ToA Forcing Due to Dust, Wm−2
  • a

    All values are global annual means.

2000 vegetation and climate4.7 × 101320.40.023−0.740.04
2100 vegetation and climate5.5 × 101368.60.050−1.820.21
2100 vegetation, 2000 climate5.5 × 101335.70.035−1.190.08
2000 vegetation, 2100 climate4.7 × 101322.60.024−0.840.05

[8] The simulated present-day annual mean dust loading (Figure 2a) is quite realistic: the dominant Saharan dust plume is well simulated, as is dust from other sources, though dust production from Chinese desert areas is too low.

Figure 2.

Simulated atmospheric dust load (kg m−2).

[9] In the 2100 experiment (Figure 2b) dust loading is increased almost everywhere and the global mean is approximately three times larger than in the 2000 experiment. A major new dust source is evident in Amazonia, in an area where forest die-back had occurred in the C00 simulation in response to a substantial reduction in precipitation. This was caused mainly by perturbations to the Hadley-Walker circulation resulting from particular patterns of sea-surface temperature anomaly in the Atlantic and Pacific Oceans [Harris, 2005], with the drying enhanced by feedbacks from the forest loss itself and by reduced plant stomatal conductance under higher CO2 [Betts et al., 2004]. Other vegetation models have been shown to respond in a similar manner to such a drying climate [Cramer et al., 2001]. The low precipitation, and the high surface wind speeds resulting from the reduced aerodynamic roughness of the deforested land, have provided ideal conditions for dust production. The Australian dust source is much larger than in the 2000 simulation, due to drying and loss of vegetation, but the vegetation in Australia was very temporally variable in the C00 integration, so this result may not be so robust. Dust production is also increased in other areas, such as in the southwestern Sahara where a region of drier soil overlaps with an area of increased windspeed, and southern Africa, where reduced soil moisture is associated with a loss of vegetation. As most of these new sources are south of the equator, the most significant increases in dust loading are confined to the Southern Hemisphere. Dust loading in the 2100 run exceeds that in the 2000 run by more than a factor of 10 over much of the Southern Hemisphere, including most of the southern Pacific, South America and Australia. In contrast, equivalent changes over most of the Northern Hemisphere are less than 20%.

[10] Data from the perturbation runs indicate that it is a combination of the change in vegetation and the change in climate which lead to the large increase in dust load (see Table 1). When compared with the 2000 experiment, the run with 2000 vegetation and 2100 climate shows a 10% increase in dust loading, and the run with 2100 vegetation and 2000 climate a 74% increase. The 2100 run shows an increase of 233%, indicating that the two effects are not additive.

[11] These results are very different from previous studies: Tegen et al. [2004] estimated future changes in natural dust emissions of −19% using HadCM3 and BIOME4, and +9% using ECHAM4 and BIOME4, and Mahowald and Luo [2003] estimated changes between −20% and −63%, depending on method, using NCAR CSM and BIOME3. Those studies used climate model reanalyses to drive the dust schemes, whereas here the HadAM3 model is constrained by climatological fields from the earlier C00 integration, and the dust scheme is an integral part of the AGCM and called with fields calculated at each timestep. This difference and the use of different vegetation and climate models will strongly influence the predictions. More importantly we have driven the AGCM with climate and vegetation fields derived from a transient experiment with full vegetation – climate feedback, which gives a very different future simulation from those produced by integrations which ignore this feedback [Cox et al., 2000]. In addition, because the dust scheme is run within an AGCM, the ability of climate to respond to vegetation changes is included in our results: e.g. the dieback of vegetation will affect local windspeed and precipitation, which will further influence dust production.

3.2. Radiative Forcing

[12] Considerable changes in radiative forcing between the two runs are seen. Mostly these are caused by differences in dust, but some aspects of differences in climate are also inevitably included: e.g. change in surface albedo due to vegetation changes, or change in cloud amount and albedo. However, the mechanisms of indirect forcings due to dust are excluded by using the double radiation call method, so the results shown here might best be described as the direct radiative forcings due to mineral dust produced at the two periods.

[13] Annual mean dust radiative forcing at the surface in the 2000 run is negative everywhere, with a distribution broadly following the pattern of loading (Figure 3a). It is the sum of negative shortwave and positive longwave components with similar distributions. The global mean is −0.74 Wm−2, but peak values in excess of −10 Wm−2 are found at the centre of the Saharan plume. In the 2100 simulation, the pattern of surface forcing is again similar to that of the loading (Figure 3b), but additional dust causes the global mean to change to −1.82 Wm−2, and produces extra forcing in excess of −10 Wm−2 below the major dust plumes.

Figure 3.

Simulated radiative forcings (Wm−2) at the surface and top of atmosphere (ToA) due to mineral dust.

[14] The ToA dust forcing is much more spatially heterogeneous. The ToA longwave forcing is positive and broadly follows the loading pattern. The shortwave forcing is more complex, generally producing warming over bright surfaces and cooling over dark ones. Shortwave and longwave forcings combine to give a small annual global mean of 0.04 Wm−2 in the 2000 simulation (Figure 3c) increasing to 0.21 Wm−2 in the 2100 run, which shows areas of strong additional forcing over the major plumes (Figure 3d).

[15] Not only the magnitude but also the sign of the additional ToA forcing varies, primarily due to differences in the underlying albedo and the dust particle size distribution and hence single scattering albedo. Both albedo and size distribution are controlled by the climate: the former is a direct function of surface cover and cloud, whilst the latter depends most strongly on windspeed and also soil particle size distribution in source areas as well as on other variables governing dust production and deposition.

[16] The changes in dust forcing might be expected to produce a positive feedback. In dust production areas the large negative surface forcing is likely to reduce vegetation growth and, additionally, the cooling of the surface and heating aloft would increase convection and windspeeds, thus enhancing dust production. However, these effects are likely to be secondary to the large changes produced by the vegetation – climate feedbacks.

4. Conclusions

[17] These results provide first estimates of future dust loadings and forcings calculated with vegetation – climate feedbacks included. The inclusion of such feedbacks can have a very significant impact on the simulated climate and vegetation, both of which are controlling factors for dust production. We have investigated the effect of this by using fields from the C00 integration to constrain our AGCM and dust scheme in simulations for 2000 and 2100. Comparison of the future with the present-day run shows that new dust source areas are created and that global mean atmospheric dust load increases by more than a factor of three. We estimate that global annual mean surface forcing due to dust more than doubles and ToA forcing increases by approximately a factor of five, with much larger regional effects. Such increases have the potential to feed back into the climate change process.

[18] Comparison of this work with previous studies suggests that the inclusion of the effects of vegetation - climate feedbacks has a major impact on the estimates of future changes in dust concentrations and, therefore, forcings. By considering the effects of vegetation – climate feedbacks, by constraining the AGCM with vegetation and climatological fields which are consistent because they come from the same climate simulation, and by using a dust scheme which is an integral part of the AGCM, these experiments are intended to give a more realistic estimate of future dust loadings than has hitherto been possible, as well as a first estimate of future dust forcings. As many of the processes parameterized in these experiments are highly model dependant and, in some cases, not yet fully understood, the results necessarily have considerable uncertainties associated with them. However, it has been shown that future changes may be large and they could potentially impact on climate and on human life. Further work is now needed to achieve better understanding and reduce uncertainties.


[19] This work forms part of the Climate Prediction Program of the UK Department of Environment, Food and Rural Affairs (contract PECD 7/12/37).