Drivers, timing and some impacts of global aridity change

PET is an estimate of the evaporative demand of the atmosphere and changes in PET with a warming atmosphere are largely determined by temperature and RH [38]. PET is the upper limit of ET and the two are equal when soil is fully saturated. There are a number of proposed methods for estimating PET, each with varying degrees of accuracy depending on availability of data, climate type, time scale, and application of interest [17, 39, 40]. For example, for non-water stressed climates and/or months, an energy-only estimate of PET (a modification of the Budyko method) is found to out-perform other PET formulations [41, 42]. For locations where temperature and precipitation are the only well-observed climate variables, a modified version of the Hargreaves formulation does a reasonable job of estimating PET [43]. Following recommendations from the Food and Agriculture Organization (FAO), we implement the Penman–Monteith (PM) formulation for PET [44, 45]. This is a physically based algorithm, including both the surface energy balance and aerodynamic components of PET. As recommended, we apply the PM equation to a reference crop, which provides an estimate of PET for a well-watered crop with an assumed crop height of 0.12 m, a stomatal resistance of 70 s m⁻¹ and an albedo of 0.23. This formulation and parameterization is widely used in the literature [e.g. 5, 8], however a few important limitations should be noted. First, because the PM provides an estimate of PET over a well-watered reference crop, this will introduces biases over non-vegetated land surfaces, for crops with different parameterizations, or for crops in more arid regions. Second, and importantly, stomatal resistance is assumed to be constant over time, when in fact it is understood to increase with CO₂ concentrations [46]. Therefore, by holding stomatal resistance constant, we risk overestimating PET in a future with higher atmospheric CO₂ concentrations. We explore the potential biases in our estimates of aridity change using the PM formulation of PET, as applied in this paper, to other formulations of PET. In particular, the aridity index using the PM formulation is compared to the Penman equation [47] , as well as the PM equation where stomatal resistance is doubled (i.e. takes the value 140 s m⁻¹) by the end of the 21st century. This geographically uniform doubling of stomatal resistance is chosen as an illustrative example of the relative sensitivity of P/PET to stomatal resistance and does not capture a realistic or complete assessment of the potential biases in holding stomatal resistance constant. The results of the sensitivity of aridity change to these choices of metrics are provided in the supplement (figures S2 and S3 are available online at stacks.iop.org/ERL/13/104010/mmedia). The results indicate that the PM equation projects a widespread higher increase in aridity than the Penman equation by a global average of 3% (figure S2). Doubling the stomatal resistance in the PM equation results in a projected percentage change in aridity that is ubiquitously less than that with a constant stomatal resistance, with a global average difference of 23% (figure S3).

To implement the PM formulation of PET we use GCM output of monthly air surface temperature, precipitation, relative humidity, sensible heat flux, and latent heat flux, and daily values of wind speed, which are averaged to the monthly time scale. All GCM output are interpolated to a uniform grid with a 2° latitude by 2° longitude resolution. The algorithm and list of GCMs used in this analysis are provided in the SI.

We use a 19-member multi-model GCM ensemble run under the 'business as usual' (Representative Concentration Pathway (RCP) 8.5) scenario. For each GCM, we calculate historical annual P/PET aridity values at each grid cell using the 1976–2000 time period. Note that while we consider annual values, P/PET will depend on the time unit of aggregation. We consider changes in aridity to be robust if at least 70% of models agree on the direction of change in aridity at any one grid cell and if the multi-model mean (MMM) aridity change exceeds 5% relative to historical values. If the MMM change in aridity is less than 5%, we consider there to be an insignificant change in aridity. For the moisture deficit index we consider absolute changes in P-PET to be significant if they exceed 84 mm yr⁻¹. This value was chosen to qualitatively match the mapped 5% changes in P/PET.

For observations, we use the CRU dataset [37] version TS4.00, which provides gridded monthly data over land spanning the 1901–2015 time period. This dataset provides precipitation, based on observations, as well as PET, estimated using the P–M equation. Trends in aridity using CRU data are calculated over the 1976–2015 time period using ordinary least squares regression.

For aridity zones, we adhere to the classification of the United Nations Convention to Combat Desertification [48], which uses P/PET values to subdivide drylands into aridity zones of hyper-arid (<0.05), arid (0.05–0.2), semi-arid (0.2–0.5), and dry sub-humid (0.5–0.65). We consider any P/PET values above 0.65 to be humid for the purposes of this analysis. Hyper-arid zones are excluded from our analysis because they occupy very little land area and have very few inhabitants.

For population density we use the Gridded Population of the World version 4 dataset [49], which provides population density estimates at a 30 arc second resolution for 2015. This dataset is compiled from the 2010 round of Population and Housing Census, which collected housing data from 2005–2014 and extrapolated for the year 2015. We average population density to a 2° latitude by 2° longitude grid resolution to match the GCM data.

For population projections we use country level UN population estimates (1950–2015) and population growth estimates (2016–2100) from the United Nations, Department of Economic and Social Affairs, Population Division [50]. These population growth estimates assume a medium fertility variant which assumes that global fertility rates decline from 2.5 children per woman in 2010–2015 to 2.0 children per woman by the end of the 21st century. Under this scenario, global population reaches 11 billion by 2100. While the most developed countries are projected to have smaller populations by mid-century than they do today, the least developed countries' populations are projected to continue to grow, though the rate of growth slows during the century. We assume that country level growth rates can be applied evenly to all population density grid cells found within the country boundaries.

Country level water consumption data come from Aquastat, a dataset from the United Nations Food and Agriculture Organization [51]. Total per capita water consumption is used in the analysis, which is an aggregate of municipal, agricultural, and industrial water consumption.

We begin by illustrating the emergence of drying trends in the CMIP5 RCP 8.5 MMM in figure 1. (For reference, calculated historical aridity distributions are presented in figure S1.) The panels on the left illustrate the percentage change in the aridity index whereas the panels on the right illustrate the absolute change in the moisture deficit. For both metrics, we see an emerging drying trend in Northern and Southern Africa, parts of Latin America, Southern Europe and Australia early in the 21st century. By mid-century, models project these regions, together with the southern portion of North America, to experience robust drying with aridity increases exceeding 10% and the moisture deficit exceeding 200 mm yr⁻¹.

Zoom In Zoom Out Reset image size

Upon inspection, figure 1 suggests that drier regions (as shown in S1) experience both an earlier emergence and a greater degree of drying. This is explored further in figure 2.

Zoom In Zoom Out Reset image size

One obvious difference between changes in the aridity index versus the moisture deficit index in figure 1 concerns the regions of uncertainty, as shown in white. For example, figures 1(a)–(c) suggests that models are uncertain about the direction of change in aridity over the Saharan desert, whereas figures 1(d)–(f) show a robust drying trend. This is a result of the historical aridity signal being nearly equal to zero in this region due to very small annual precipitation values. Small differences in precipitation between future and historical projections in any one GCM can have a large influence on the percentage change of the MMM, even if most GCM's aridity values stay very close to zero.

We investigate further the rate and degree of increases in aridity by aridity zone. Figures 2(a) and (d) illustrates the proportion of land area, by aridity zone, where there is a robust drying signal across models in this scenario. Figure 2(a) shows that 50% of land area in all aridity zones is projected to be at least 5% more arid by 2075. There is remarkable agreement across all aridity zones in terms of the proportion of land area that is projected to become more arid. In contrast, figure 2(d) indicates that the moisture deficit is increasing earlier and across more land area in all dryland classified zones as compared to the humid zone. For example, by 2035, ∼60% of arid land area will have a robust drying signal, while ∼45% of dry sub-humid land area and just over 20% of humid land area will have a robust drying signal. Figures 2(b) and (c) illustrate how the robustness of the aridity response changes as the threshold of drying increases. Figures 2(b) and (c) show that more severe drying emerges earlier in arid zones than in humid zones, and a greater fraction of the land area exhibits robust drying for arid regions. A similar trend is shown for the moisture deficit (see figures 2(e) and (f)). Figures 6 and S7 indicate that this difference in drying across aridity zones is driven mainly by precipitation change differentials across aridity zones.

Figures 3(a) and (b) show the trends from an ordinary least squares regression over 1976–2015 of the decadal percentage change in aridity relative to 1976–2000 for the CMIP5 multi-model median and CRU observations, respectively. The spatial pattern in trends shows remarkable similarity between the multi-model median and observations. A wetting signal is shown in the Sahel, Greenland, and Northwestern Australia, and a drying signal is shown throughout most of Sub-Saharan Africa, Europe, Saudi-Arabia, the Southwestern United States and Brazil. The observed trends show a larger magnitude signal than the multi-model median, though this is not surprising given that variability would be smoothed out in the ensemble median. Analogous maps to figure 3(b) for individual single runs of different GCM simulations (shown in figure S4) show similar magnitudes of trends, though sometimes in different locations.

Zoom In Zoom Out Reset image size

Figure S5 shows the proportion of land mass by aridity zone where CRU observations show a decrease in P/PET of 5% or greater relative to 1976–2000. The analogous plot for individual GCM simulations is shown in figure S6. Observations indicate that arid zones have already experienced a greater extent of drying than more humid zones. The same is projected to be true for many of the individual GCM simulations.

Figure 4 illustrates population density (number of people km⁻²) in locations projected to experience robust minimum aridity changes of various magnitudes relative to 1976–2000. These maps include projections of population growth. Figures 4(a) and (b) show that according GCMs, a robust drying trend of 5% or greater emerges in some regions in the 2001–2015 time period, coinciding with populations in parts of Northern Africa, the Middle East, and South America. Figure 4(c) is analogous to figure 4(b) but for CRU observations, which indicate that populations adjacent to the Mediterranean Sea have already experienced a drying of 5% or greater. By mid-century (figures 4(d)–(f)), the extent and degree of drying has increased such that aridity changes are expected to be 20% or greater in regions highlighted in figures 4(b) and (c). By mid-century, Southern Africa, most of Europe, and Central America are projected to experience a robust decrease in P/PET by 10% or greater, and China, most of the United States and parts of Canada are projected to experience a robust decrease in P/PET of 5% or greater. These projected drying patterns continue to spread in extent and become more severe by the end of the 21st century (see figures 4(g)–(i)).

Zoom In Zoom Out Reset image size

We could reframe this question by asking 'how many more people will live in arid zones by the end of the century'. Assuming population distributions remain unchanged, this analysis indicates that approximately 200 million more people are projected to live in arid and hyper-arid regions (P/PET <0.2) by the end of the century if global emissions follow a business as usual trajectory (see figure S8 showing population distribution by the CMIP5 MMM aridity index (P/PET) under historical climate and at the end of 21st century). If population growth estimates are accounted for then the number of people living in arid zones is projected to double. This is equivalent to approximately 700 million more inhabitants in arid zones by the end of this century. This suggests both that dry zones are projected to dry further and that dry zones are regions where we expect to see high population growth. The largest population increase in arid regions is projected to occur in Africa and the Middle East (see figures S9 and S10 for the locations of these populations).

One of the central aims of this paper is to probe the impacts of aridity change on global populations. To do this, we ask the question 'to what extent will the aridity index change in countries where per capita water consumption is presently very low'. To answer this question we consider aridity changes in countries where inhabitants are living under absolute water scarcity (shown in figure 5). Absolute water scarcity is defined by per capita water withdrawals at or below 500 m³ yr⁻¹ [51, 52]. As a reference point, the United States' per capita water withdrawal is just over 1500 m³ yr⁻¹, whereas most countries in Africa have a per capita water withdrawal of less than 100 m³ yr⁻¹. In figure 5, the most vulnerable regions are shown to be in Southern Africa, large parts of Europe and Brazil, where inhabitants presently live under absolute water scarcity, and drying is projected to emerge early in the 21st century. Note that some developed countries with low water consumption appear in parts of Europe (for example, France and Germany). This is in part due to low industrial and agricultural water consumption in these regions.

Zoom In Zoom Out Reset image size

In this final section, we investigate the drivers of aridity change and the relative contributions of temperature and precipitation to aridity change and impacts. Figure 6(a) shows projected changes in aridity by mid-century where temperature is held constant at historical values. This pattern holds for other time periods. Here we see a robust drying in southern Africa, the southern United States and Mexico, Europe, China, parts of Australia and throughout most of Latin America. We also see a robust wetting in Northern Canada, the horn of Africa, India, throughout the Tibetan plateau and the eastern portion of Russia. The individual contributions of precipitation, temperature, and RH to total aridity change are shown in S11. Here we see that temperature contributes to ubiquitous drying, with increased drying in northern latitudes. The contributions from RH are small in comparison to both temperature and precipitation although it should be noted that such an offline calculation is unable to account for the inherent coupling between relative humidity, temperature, and precipitation. By considering these relative changes, we see that precipitation is driving the most variability in aridity change, whereas temperature change is leading to increased PET globally.

Zoom In Zoom Out Reset image size

Interestingly, the most extreme drying due to precipitation change occurs in the subtropics and coincides with the most arid regions of the world. This point is illustrated in S7, which shows the relative contributions of T and P to aridity change, by aridity zone, and by intensity of aridity change. We notice that temperature increases tend to dominate the drying response in all types of arid zone for mild drying (ΔP/PET = −5%). However, for more extreme drying, precipitation plays an increasingly important role, more so for drier regions.

In figures 6(b) and (c), we explore this question further by considering the robustness of the proportional changes in P across aridity zones. Figure 6(b) illustrates that robust decreases in P are projected to occur early in the 21st century over arid regions extending over approximately 30% of land area. Humid regions see very little decreases in precipitation by comparison. In contrast, figure 6(c) shows that humid regions are projected to see a robust increase in P over more than 50% of humid land area by 2075. This is a far greater increase in P than is exhibited by drylands. This robust trend in P across aridity zones suggests that precipitation may be the driving mechanism leading the differential trend in the aridity index across aridity zones. RH was also considered and results are shown in S12; arid zones are projected to see the largest extent of both RH increases and RH decreases, suggesting that RH is not driving the trend in aridity change across aridity zones.

While the Held and Soden [25] 'wet-get-wetter, dry-get-drier' mechanism was shown to break down over land [26], we have shown here that a moisture mechanism that appears to hold over land may be better described as 'most-land-gets-more-arid, humid land-gets-more-rain'. Note, however, that a number of the humid regions receiving more rain are sparsely populated, and we address this point further below.

Figure 7(a) shows the total number of people who live in regions projected to see an increase in aridity of 25% or greater (i.e. P/PET decreases by 25% or more), assuming that the population distribution remains unchanged. Here, the solid line represents the MMM value as a function of global mean temperature change and the shaded region indicates +/−1 standard deviation. The green line provides the projected values under RCP 8.5, the purple assumes temperature is held constant at historical values, and the red line assumes that precipitation is held constant at historical values. With a 5 °C increase in global mean temperature, the MMM projects that 2 billion people will be subjected to aridity increases of 25% or more. If global temperatures remain at 2 °C, this value is instead roughly half a billion people. If, instead we calculate these values assuming the medium variant population growth (see figure 7(c)), then the MMM projects that 3 billion people will be living in a region that is 25% more arid than historical values, as shown in figure 7(b). Considering the relative impacts of the purple and red lines, we see that for both figures 7(a) and (b), precipitation seems to more important than temperature in driving the population level aridity change, especially for global mean temperature values <3 °C. Their relative contributions converge for higher global mean temperature values. Figure S13 shows that more than half of the world's population is projected to experience drying of 5% or more.

Zoom In Zoom Out Reset image size

Figure S14 shows analogous plots to figure 7 but disaggregated by continent. According to GCMs, Africa is projected to experience the largest population impacts with more than a billion inhabitants expected to experience a drying of 25% or more. The second largest population projected to see increases in aridity is in Asia, where ∼750 million people are projected to be impacted by aridity increases of 25% or greater. Surprisingly, even in Europe, where populations are decreasing over time, models still project that nearly half a billion people will live in places undergoing severe drying of 25% or greater under a global mean warming of 5 °C. S14 shows that for Europe and North America, aridity change is largely driven by warmer air surface temperatures (note the proximity of the red line to the green line in these figures). In Africa and Latin America, aridity change is driven by decreases in precipitation.

The analogous plot to S14, but weighted by land area instead of population size is shown in S15.