Changes in desert area due to humans have important implications from a local, regional to global level. Here I focus on the latter in order to better understand estimated changes in desert dust aerosols and the associated iron deposition into oceans. Using 17 model simulations from the World Climate Research Programme's Coupled Model Intercomparison Project phase 3 multi-model dataset and the BIOME4 equilibrium vegetation model, I estimate changes in desert dust source areas due to climate change and carbon dioxide fertilization. If I assume no carbon dioxide fertilization, the mean of the model predictions is that desert areas expand from the 1880s to the 2080s, due to increased aridity. If I allow for carbon dioxide fertilization, the desert areas become smaller. Thus better understanding carbon dioxide fertilization is important for predicting desert response to climate. There is substantial spread in the model simulation predictions for regional and global averages.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Changes in desert area impact the human living in those areas substantially and can impact the surface albedo and feedback onto climate. In addition, the changes in desert area may lead to changes in desert dust source generation, which has important implications for biogeochemistry and climate [e.g., Martin, 1990; Miller and Tegen, 1998]. Previous studies have highlighted the potentially important role that human induced climate change will have on desert size and dust generation [e.g., Mahowald and Luo, 2003; Tegen et al., 2004; Woodward et al., 2005]. However, previous studies have used only one or two models in order to assess the impact of climate onto mineral aerosols [e.g., Mahowald and Luo, 2003; Tegen et al., 2004; Woodward et al., 2005; Mahowald et al., 2006a]. In this study I explore the predictions of climate change from 17 different general circulation models compiled in World Climate Research Programme's Coupled Model Intercomparison Project phase 3 (CMIP3) multi-model dataset and use the BIOME4 model to convert the precipitation, insolation and temperature changes from those models into predictions of desert. The goals of this study are 1) to identify desert dust source regions which may change (or already have changed) due to the impact of anthropogenic climate change, and 2) to look for the sensitivity of the results to the general circulation model used for the climate predictions.
 Carbon dioxide fertilization of arid plants is a potentially important mechanism for changing the size of desert dust source regions [e.g., Mahowald et al., 1999]. Atmospheric carbon dioxide concentrations influence plant stomatal conductance because as CO2 diffuses into the leaf, water diffuses out. This effect is particularly pronounced under conditions of low atmospheric humidity, where stomatal control is an important regulator of net CO2 uptake by plants. Under high ambient CO2 concentrations, more CO2 could be expected to diffuse into the leaf for unit water loss, increasing plant water use efficiency and thus productivity and survival under arid conditions. There is evidence that this mechanism exists in the real world, but the magnitude is controversial [e.g., Smith et al., 2000; Caspersen et al., 2000]. Therefore I include sensitivities studies of the impact of carbon dioxide fertilization effects in the simulations in order to determine the importance of this mechanism.
 Humans can also impact desert dust sources through direct perturbation of the soils during land use or water diversion. The importance of human land use in perturbing atmospheric dust is highly uncertainly and is estimated to be between 0–50% in the current climate [e.g., Mahowald and Luo, 2003; Tegen et al., 2004; Mahowald et al., 2004; Moulin and Chiapello, 2006; Mahowald et al., 2007]. The impact of water use on atmospheric dust has also not been assessed at a global level, but a recent study suggests no trend in dustiness downwind of the Aral Sea, which has substantially reduced in size during the last 50 years due to human water use [Mahowald et al., 2007]. While human land use is likely to be important for desert dust generation, it is beyond the scope of this study to consider.
 I limit my discussion here to changes in desert dust source area: calculating the dust source strength requires the diurnal cycle of friction velocity of winds which is not archived in the PCMDI archive [e.g., Luo et al., 2004]. In addition, there may be fluctuations in the transport and lifetime of the aerosols that are different in different climate models, which I cannot account for in this study. I also neglect the feedback of changes in desert or desert dust itself onto climate in this study [Yoshioka et al., 2007]. However, this study represents a significant step forward, as changes in the source areas can account for a large fraction of the changes seen in different climates [e.g., Mahowald et al., 1999, 2006a].
 In order to determine desert area changes, the BIOME4 model was used [Kaplan et al., 2003]. This model determines the optimal plant functional type for each grid cell on a 0.5 × 0.5 degree cell based on which type out-competes the other plant types. This model will determine the equilibrium vegetation, which is attained in many ecosystems after hundreds of years. Because the vegetation in arid regions (like annual grasses) is likely to respond quickly to the evolution of the climate [Tucker and Nicholson, 1999], I can use the model for 20 year average time slices every 50 years to look at the evolution of the deserts as predicted by this set of models.
 Similar to previous studies, I use an anomaly method, where the monthly mean anomalies from the current climate simulation are added to observational datasets for the current climate [Harrison et al., 1998]. It is thought that anomalies from the current climate are more robust that the absolute value for different climates. Monthly averaged model results for 1880–1900, 1940–1960 and 1980–2000 were taken from the twentieth century runs from the CMIP3 dataset. Simulated monthly means for the 2040–2060 and 2080–2100 time period were taken from the A1b scenarios [Intergovernmental Panel on Climate Change, 2001]. Monthly anomalies were derived for each time period and used to drive the BIOME4 model to equilibrium for each grid box. Variables taken from the model output datasets were precipitation, temperature and cloudiness changes. The models included in this study are shown in Figure 1. I use the model designation of the desert biome for this discussion. The desert biome is generated in the BIOME4 model when the leaf area index is below 1.0 for woody desert plants, or when the net primary productivity is below 100.0 for cool conifer trees, grasses or woody desert plants.
 There is some data for the period 1940–1960 that can be used to drive the BIOME4 model for that time period, to compare against model output. Precipitation data from a merging of two gridded monthly time series from Chen et al.  and Dai et al. , as described by Dai et al.  are used. Precipitation data are also difficult to interpret because of the hetereogeneities in the spatial distribution of precipitation. However, this gridded dataset represents our best state of knowledge about precipitation at the global scale. Temperatures from the CRU dataset [Jones et al., 2003] are used for temperature changes, and the cloudiness is assumed not to change in the BIOME4 simulations using observations.
 In order to better understand where these changes in the dust source regions are likely to impact ocean biogeochemistry, I include a source apportionment study for the dust module [Mahowald et al., 2006a] in the National Center for Atmospheric Research (NCAR) Community Climate System Model (CCSM3). This 3-dimensional model simulates the source, transport and deposition of dust using 4 size bins, and has been compared to observations [Mahowald et al., 2006a]. For each source region, a separate tracer simulation is conducted for just that source region for one year, so that I can estimate where different sources dominate the deposition. The climate response of this model to various dust forcings is discussed by Yoshioka et al.  and Mahowald et al. [2006b].
3. Results and Conclusions
 In the global mean, the models predict drying of the desert regions due to warmer temperatures with an increase in greenhouse gases, as seen in the mean observations during the past 25 years [e.g., Dai et al., 2004]. With no carbon dioxide fertilization in the BIOME4 model, this means an increase in desert regions globally (Figure 1a), with a wide variation in the individual model predictions. If we take the median across the models instead of the mean, we obtain similar results (not shown). If carbon dioxide fertilization is included in the simulations, this produces a decrease in desert area, as the higher carbon dioxide levels allow the plants to respond to the increased aridity more effectively (Figure 1b). There is however, a wide spread in the predictions of different models, with the NCAR CCSM showing the greatest decrease in desert area in the future, and the GFDL and CSIRO models showing the biggest increase in desert in the future. The response of the CCSM3 in this study is similar to that shown by Mahowald et al.,  using the BIOME3 model, suggesting that differences from the IPCC models is more important than whether BIOME3 or BIOME4 is used.
 The averaged response of the CMIP3 models (average across all models used here) in different regions is shown in Figures 2a and 2b and Table 1 for the cases of no CO2 fertilization or with CO2 fertilization, respectively. With no CO2 fertilization, most desert regions increase in size in the future, with the exception of the desert extent in the Middle East, which decreases slightly. In the case of CO2 fertilization, some of the desert regions are predicted to increase in size (S. Africa, China/Mongolia, and N. Africa), while others decrease. Smaller desert areas tend to have the greatest percent change in area. The relative change in desert area predicted for the different models shows substantial spread (shown in Figure S1).
Table 1. Percent Change in Desert Area Size Using the Mean of All the Models and Available Observations for Precipitation and Temperaturea
With No CO2 Fertilization
Last Column for 1940.
With CO2 Fertilization
 The mean projected change in future desert area (2080–2100) is shown in Figure 3 for the case of no CO2 fertilization or including CO2 fertilization, respectively. In addition, it is useful to know whether the models agree or disagree on the change in deserts in any particular region—perhaps the response in that region is more robust. There is not strong consensus in the models about whether different desert regions will increase or decrease in size (Figure 3).
 For the period 1940–1960 relative to the present, there is enough temperature and precipitation data in arid regions to use gridded data to assess changes in desert areas using the BIOME4. The results are shown in Table 1, the overall patterns are similar to the mean of the models, in that 1940–1960s have slightly larger deserts than today, but the relative change in each region is not consistent between the CMIP3 model-derived and observational-derived changes in desert areas. This is especially true for North Africa, much of which change is driven by the Sahel drought, which may or may not be captured by the models accurately.
 One of the important implications of changes in desert size is its impact on the downwind dust deposition. A source apportionment study from a model simulation shows generally which source areas will predominately impact which ocean deposition regions. Plotted are the annual averaged deposition fluxes (Figure 4a) and the source region with the largest deposition flux (Figure 4b) from the model simulations given by Mahowald et al. [2006a]. This highlights the importance of understanding what climate change and humans will do to the small Southern Hemisphere sources in order to better predict deposition fluxes, and thus iron deposition changes to these important ocean regions. Looking at the mean changes in the desert size, and the downwind regions these impact, shows that the mean model prediction is that deposition to the Arabian Sea will decrease whether or not carbon dioxide fertilization is included. The average of the models predicts an increase in deposition in 2100 relative to today downwind of South Africa, Australia, Asia and North Africa whether or not carbon dioxide fertilization occurs. The prediction is for a small change for North Africa, but that due to the strength of the North African plume, this change is important for broad regions of the globe. For other regions, the mean of the model changes between 2100 and today changes sign if carbon dioxide fertilization is included. This includes the South American source, which dominates dust deposition to the equatorial south Pacific and broad regions of the Southern Oceans, regions which may be especially sensitive to changes in dust [Martin, 1990]. However, realize that there is substantial variation between models in the predictions, and the mean of the models does not match the observationally-based predictions for 1940–1960.
 For the impact of changes in desert dust area on aerosol optical depth and thus on climate, there is no information for each source area. But Mahowald et al. [2006b] show that there is a linear response in radiative forcing or temperature to optical depth. Making a very simple assumption that changes in source area will lead to proportional changes in aerosol optical depth (which is largely supported by Mahowald and Luo  and Mahowald et al., ), this suggests that the globally averaged aerosol optical depth could be 1.6% lower or 7.9% higher in 2100. These are much more moderate changes then predicted by Mahowald and Luo  or Mahowald et al. [2006a], more similar to Tegen et al.  because the CCSM tends to predict larger changes in deserts than the rest of the CMIP3 models.
 This paper highlights the importance of evaluating multiple models in assessing future predictions of changes in desert dust. Predicting precipitation in a general circulation is very difficult, and the models do not agree what will happen in the future. This leads to a wide disagreement from the models on which source areas will increase or decrease in the future. Unfortunately it is not known which model (if any) is correct—comparisons with desert changes between 1940–1960s and present, where there is data, suggests very different results than the mean (or median) of the models. Overall, changes predicted using the mean of all the models is more moderate than using one model prediction [cf. Mahowald et al., 2006a]. In addition, the uncertainty in the magnitude of the carbon dioxide fertilization mechanism can change the sign in the predictions. Potential interactions between humans land use and desert dust source strength are neglected in this study, although they are likely to be important. In addition, desert area changes and especially desert dust itself may significantly change climate [Yoshioka et al., 2007], and these impacts should be considered in future studies.
 This work was done at the National Center for Atmospheric Research, sponsored by NSF. I acknowledge the modeling groups for making their simulations available for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the CMIP3 model output, and the WCRP's Working Group on Coupled Modelling (WGCM) for organizing the model data analysis activity. The WCRP CMIP3 multi-model dataset is supported by the Office of Science, U.S. Department of Energy. The manuscript was improved by the comments of two anonymous reviewers.