Nature: Historical records of precipitation, streamflow and drought indices all show increased aridity since 1950 over many land areas1, 2. Analyses of model-simulated soil moisture3, 4, drought indices1, 5, 6 and precipitation-minus-evaporation7 suggest increased risk of drought in the twenty-first century. There are, however, large differences in the observed and model-simulated drying patterns1, 2, 6. Reconciling these differences is necessary before the model predictions can be trusted. Previous studies8, 9, 10, 11, 12 show that changes in sea surface temperatures have large influences on land precipitation and the inability of the coupled models to reproduce many observed regional precipitation changes is linked to the lack of the observed, largely natural change patterns in sea surface temperatures in coupled model simulations13. Here I show that the models reproduce not only the influence of El Niño-Southern Oscillation on drought over land, but also the observed global mean aridity trend from 1923 to 2010. Regional differences in observed and model-simulated aridity changes result mainly from natural variations in tropical sea surface temperatures that are often not captured by the coupled models. The unforced natural variations vary among model runs owing to different initial conditions and thus are irreproducible. I conclude that the observed global aridity changes up to 2010 are consistent with model predictions, which suggest severe and widespread droughts in the next 30–90 years over many land areas resulting from either decreased precipitation and/or increased evaporation.
Although the historical and future aridity changes have been discussed in previous studies1, 2, 3, 4, 5, 6, 7, there still is a need to validate the historical changes and reconcile them with model projections. Here I focus on synthesizing the observed aridity changes and comparing and reconciling them with model-simulated changes, thereby improving our understanding of global-warming-induced drought changes.
Different drought indices can result in somewhat different change patterns, especially on small scales14. Here I focus on the large-scale drying trends in precipitation, streamflow and soil moisture fields, which are commonly used to quantify, respectively, meteorological, hydrologic and agricultural drought1. Because historical records of soil moisture are sparse, I also used the self-calibrated Palmer drought severity index (PDSI) with potential evapotranspiration estimated using the Penman–Monteith equation (sc_PDSI_pm; ref. 2). The PDSI is calculated from a water-balance model forced with observed precipitation and temperature and has been widely used in monitoring drought development over the USA, palaeoclimate reconstruction15 and studying aridity changes2, 5, 6, 16. The revised sc_PDSI_pm has improved spatial comparability and uses a more realistic estimate of potential evapotranspiration, thus improving its applicability to global warming scenarios (see ref. 2 for more details).
Figure 1: Trend maps for precipitation and sc_PDSI_pm and time series of percentage dry areas. 
Figure 1a,b shows that the broad patterns of the linear trends from 1950 to 2010 in observed annual precipitation and calculated sc_PDSI_pm using observation-based forcing2 are comparable. These patterns are also broadly comparable to those seen in observed streamflow trends since 1948 in the world’s main river basins1, 17. Some regional and quantitative differences are expected among them as they are different variables, albeit closely related physically. The patterns are characterized by drying over most of Africa, southeast Asia, eastern Australia and southern Europe, and increased wetness over the central US, Argentina and northern high-latitude areas. As the precipitation and streamflow data are from independent measurements, the broad consistency among their change patterns suggests that these trends are real. This also suggests that the sc_PDSI_pm is a useful measure of aridity changes. One advantage of the sc_PDSI_pm is that it can be used to examine the impact of individual forcing on the aridity trend by comparing the cases with and without this forcing in calculations of the sc_PDSI_pm.
Figure 1c shows that the warming since the 1980s (note the jump around the early 1980s is due to the 1982/1983 El Niño) has contributed considerably to the upward trend in global drought areas, increasing the areas under drought by about 8% by the first decade of this century. This warming-induced drying results from increased evaporation and is largest over northern mid-high latitudes2. In contrast, precipitation decreases over Africa, southeast Asia, eastern Australia and southern Europe are the primary cause for the drying trend over there, and the long-term trends and decadal tomultidecadal variations in sea surface temperature (SST) are a major driver for many of the precipitation changes8, 9, 10, 11, 12. The long-term SST trend is part of the global warming; however, many of the observed decadal to multidecadal SST variations are absent in greenhouse-gas- (GHG) and aerosol-forced coupled model simulations13, implying that these SST variations are unforced, natural variations whose phase or timing and spatial patterns may depend on the initial conditions of the models and thus they are generally irreproducible.
[Figure description:
Long-term trends from 1950 to 2010 in annual mean a, observed precipitation2 and b, calculated sc_PDSI_pm using observation-based forcing2. The stippling indicates the trend is statistically significant at the 5% level, with the effective degree of freedom computed using the method of ref. 30. Note a change of 0.5 in the sc_PDSI_pm is significant in the sense that a value of PDSI between −0.5 to −1.0, −1.0 to −2.0, −2.0 to −3.0 and −3.0 to −4.0 indicates, respectively, a dry spell, mild drought, moderate drought and severe drought2. c, Smoothed time series of the drought area as a percentage of global land areas based on the sc_PDSI_pm computed with (red line) and without (green line) the observed surface warming. The drought areas are defined locally as the cases when sc_PDSI_pm is below the value of the twentieth percentile of the 1950–1979 period (results are similar for drought ]defined as PDSI<−2.0 and for using a longer base period from 1948 to 2010).
