Today's 100 year droughts in Australia may become the norm by the end of the century

Climate projections of monthly precipitation and daily temperatures (minimum, maximum and mean) from 1981 to 2100 for Australia were obtained from nine GCMs (table S1 available online at stacks.iop.org/ERL/17/044034/mmedia) of the ScenarioMIP experiment [25] of CMIP6 [23] under four different scenarios of greenhouse gases emissions and socio economic development by the end of the century (SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5). The nine GCMs were chosen based on the availability of selected variables at ∼100 km resolution. The same variables were obtained from 1981 to 2019 from the ERA5-Land reanalysis dataset [26, 27] to serve as a comparison to GCMs historic period. Given that we are interested in the analysis of anomalies, rather than the actual magnitude, of climate variables, we did not apply any bias adjustment procedure to GCM outputs. Our decision on analyzing the original GCM outputs without any bias adjustment is based on the fundamental assumption that GCMs provide plausible projections of climate anomalies and consistent with recent studies on future droughts [2, 11, 28].

Potential evapotranspiration (PET) was estimated based on the Hargreaves–Samani (HS) method [29, 30]. Although previous research indicated the Penman–Monteith (PM) method as the most physically robust [31, 32], it requires significantly more inputs, which would reduce the number of available GCM's. A comparative analysis between PET methods is presented in the supplementary material (section S2.3) to demonstrate the level of uncertainty and consistency of findings of this work with respect to the PET method used.

Droughts were identified using two well established indices, the Standardized Precipitation Index (SPI) [33], and the Standardized Precipitation Evapotranspiration Index (SPEI) [34]. These indices are defined as anomalies in the monthly standardized time series of precipitation (SPI) and precipitation minus PET (SPEI) calculated over a predefined temporal scale (e.g. 3, 6, 12 months, usually termed SPI₃, SPI₆, SPI₁₂, etc) preceding the month of interest. Since we aim to quantify future, potentially unprecedented values, a parametric approach for the computation of the indices is preferred. Following previous studies [35], a Gamma distribution was used for standardizing SPI, and a three parameter generalized extreme value for SPEI (see section S2.4 for details). The parameters describing these distributions were estimated from the historic part of GCMs and were used to calculate historic and future indices.

Distinct drought events were identified as continuous negative SPEI/SPEI series with at least one value ⩽−1 (i.e. a negative anomaly exceeding 1 standard deviation) [33]. Drought severity was defined as the accumulated SPI/SPEI values within a drought event and a duration of interest [33]. Although the approach can be applied to any time scale, our analyses are based on SPI₃/SPEI₃; a temporal scale of drought that has been found to correlate well with fire severity [36] and plant productivity [37]. Drought indices were calculated at monthly resolution and drought severity was computed by integrating the indices over different durations within drought events.

Changes in the spatial extent of droughts were analyzed by investigating the drought area ratio (DAR) defined as the fraction of total area of Australia under drought conditions. DAR was analyzed for all drought conditions together with different severity classes: moderate (−1 ⩾ SPI/SPEI > −1.5), severe (−1.5 ⩾ SPI/SPEI > 2) and extreme (−2 ⩾ SPI/SPEI) [33].

By definition, only a few droughts may occur every year. Thus, the statistical analysis of extreme droughts challenges the asymptotic assumption of the extreme value theorem [38, 39]. Here, we address this issue by adopting a framework based on the concept of ordinary events, which are the independent realizations of the process of interest [40–42]. The approach is based on the idea that a fraction of data much larger than the extremes, the so-called ordinary events, can be used to (a) parametrize the tail of the distribution, and (b) explicitly consider the number of occurrences of the process of interest. By relying on an increased sample (ordinary, as opposed to extreme, events), this approach reduces stochastic uncertainties and systematic errors stemming from bias in the data and limited length of the data records [20–22]. Following the methodology detailed in [43], we focus our analysis on the largest portion of ordinary events whose statistical properties are shared with extremes. Analyses run over several drought temporal scales and locations across Australia suggest that the largest 30% of the ordinary events can be effectively used to describe drought extremes (see section S2.5 for details). The tail of the distribution is found to be characterized by a stretched-exponential decay, which can be parametrized by a two-parameter distribution in the form: $F\left( {x,\theta } \right) = 1 - {{\text{e}}^{ - {{(x/{\theta _1})}^{{\theta _2}}}{ }}}$. Extreme T-year drought severities $x$ are then computed explicitly considering the occurrence frequency of droughts by inverting the extreme value distribution: ${p_{\text{T}}}\left( x \right) = F{(x,\theta )^n}$, where ${p_{\text{T}}}$ is the annual exceedance probability, F is the tail model described by the parameters $\theta$, and n is the ratio between the total number of drought events and the number of years in the considered time period.

A first indication of the projected changes in droughts can be obtained by analyzing the differences in the empirical cumulative distribution function (CDF) of drought indices for different periods (figure 1). Figure 1(a) presents the CDF for each GCM in the historic period which, by definition of SPEI, is a standard normal distribution. The spread of CDFs in this panel, which is more pronounced on the left tail of the distribution, highlights the uncertainty in representation of extremes due to stochastic sampling uncertainty. The historic median distribution shows a frequency of roughly 2% and 12% for extreme and severe/moderate indices respectively. Under the SSP1-2.6 scenario (figures 1(b) and (c)), most of the models show a clear shift towards dryer conditions (higher probability of low SPEI values) in comparison to the historic period. Relatively small changes are observed in SPEI frequencies between mid- to end-century, which indicates a stabilization for this scenario, in agreement with previous studies [2, 5]. The frequency of extreme indices (−2 ⩾ SPEI) rises to ∼5% (multi-model median value and model range ∼4%–9%) during mid-century, and ∼6% (∼3%–8%) by end-century. The frequency of moderate to severe indices (−2 > SPEI ⩾ −1) is slightly different for both periods, with multi-model median of ∼25% (and range ∼16%–41%) for mid-century and ∼19% (∼17%–22%) for end-century. These frequencies are considerably lower during the historic period than those obtained from the model's projections, which indicates that under SSP1-2.6, conditions become drier, but with a tendency of stabilization from mid- to end-century. Under SSP2-4.5 (figures 1(d) and (e)), the changes in distribution are similar to the previous, but with a slight increase in the frequency of extreme indices, which is ∼5% (∼4%–7%) during mid-century and ∼8% (4%–12%).

Zoom In Zoom Out Reset image size

Changes are more pronounced under scenarios SSP3-7.0 (figures 1(f) and (g)) SSP5-8.5 (figures 1(h) and (i)). Although the mid-century responses are similar for all scenarios, in these last two scenarios, the CDFs revealed a steep increase in frequency of severe and extreme drought indices towards the end of the century. This behavior is unanimous among the GCMs and, even though the magnitudes of change vary, they are always larger than the stochastic uncertainty represented by the spread of the models around the normal distribution in the historical period. The median frequency of extreme indices during the mid-century is ∼5% (∼3%–7%) for SSP3-7.0, ∼6% (∼3%–7%) for SSP5-8.5 and increases towards the end of the century to ∼10% (∼6%–20%) and ∼15% (∼5%–25%) respectively.

A drying behavior was also observed for SPI (figure S1) indicating an increase in the number of events of abnormally low precipitation, although the magnitude of the changes in the indices is higher for the former, which highlights the effects of PET in the SPEI estimates.

Figure 2 shows the evolution of the decadal average of DAR for different drought severity classes based on SPEI and GCM outputs, and the average DAR from ERA5-Land data (dashed lines). The portion of Australia under drought is consistently increasing towards the end of the 21st century in descending order of magnitude under SSP5-8.5 (red color in figure 2), SSP3-7.0 (blue color in figure 2), and SSP2-4.5 (yellow color in figure 2) while it increases and stabilizes after mid-century under SSP1-2.6 (green lines and shaded area figures 2(a)–(d)). More specifically, under SSP1-2.6 the multi-model median values increase until near 2060, then remain almost steady around 0.50 towards the end of the century. On the other hand, all the other three scenarios present a continuous increase in DAR, especially towards the end of the century. SSP5-8.5 increases continuously for mid- to end-century, while SSP2-4.5 and SSP3-7.0 have a stabilization period between 2050 and 2070 and then start to increase again. This behavior is also observed in the different classes of drought severity (figures 2(b)–(d)), with more pronounced changes in extreme droughts (figure 2(d)). At mid-century, the values of multi-model median DAR for all drought classes are above the average of the historic period (figure 2(a), dashed line at 0.35) for all four scenarios. The inter-model variability ranges from values below to more than double the historic mean (∼0.25–0.75), illustrating a large variability in model's projections, which nevertheless agree with the drying trend in the studied scenarios. The models' response for the end of the century shows diverging trends between SSP1-2.6 and the other scenarios. Under SSP1-2.6, the multi-model median DAR stabilizes around 0.50 after 2060, while it steadily rises to almost 0.70, 0.65 and 0.55 under SSP5-8.5, SSP3-7.0 and SSP2-4.5 respectively. Inter model variability is larger at the end-century period, ranging from near historic conditions (0.35) to almost three times higher than historic average (0.88). The response of severe (figure 2(c)) and extreme (figure 2(d)) droughts presents similar trends to the total drought, although the magnitudes and inter model variability differ for each class. For the moderate class, the historic average is ∼0.28 (figure 2(b), dashed line), which is nearly the same as the SSP5-8.5 value at 2100, while it is smaller than the SSP1-2.6, SSP2-4.5 and SSP3-7.0 values (∼0.40, 0.40 and 0.35). The historic average DAR for the severe class is ∼0.04, while the DAR is ∼0.06, ∼0.09, ∼0.095 and ∼0.11 for end of century projections for scenario SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 respectively. The DAR of extreme droughts presents larger relative variations with respect to the historic average (figure 2(d), dashed line at 0.02). By the end of the century, the SSP5-8.5 scenario projects a tenfold increase of the area subject to extreme drought, a nearly fivefold increase for SSP2-4.5 and SSP3-7.0, and the SSP1-2.6 a twofold increase.

Zoom In Zoom Out Reset image size

The dynamics of drought trends found for each emission scenario follow the overall projected trends in precipitation and temperature for each scenario respectively [44]. Future projections for the SPI (figure S2 in the supporting information) exhibit also higher values of DAR when compared to the historic period, however the magnitudes of changes are smaller than the SPEI results. This indicates that the overall projected change in precipitation is less significant than the combined changes in precipitation and PET combined, highlighting the important control of temperature on future drought characteristics.

SDF curves synthesize three of the most important drought characteristics. Figure 3 shows SDF curves for fixed 12 month duration droughts in the state of South Australia, derived from SPEI based on spatially averaged precipitation and temperatures. In this region, extreme droughts are becoming more frequent under all studied scenarios, considering the multi-model median. Figures 3(a), (c), (e) and (g) show respectively the SDF for SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5. There is a clear increasing trend in severity towards the end of the century, illustrated by the vertical shift between curves, with a more pronounced increase in SSP3-7.0 and SSP5-8.5. Analysis of the SDF curves allows us to directly quantify changes in drought risk. For example, considering SSP5-8.5, the present 100 year drought—that is, a drought severity which in current climate is expected to occur on average once in 100 years—is expected to be exceeded on average once every 30 years by the mid of the century, and once every 3 years by the end of the century. This corresponds to an increase in drought risk from 1% to 3.3% in the mid-century and to approximately 33% towards 2100. Scenarios SSP3-7.0 and SSP2-4.5 exhibit the same response for mid-century (30 years exceedance period, risk increase from 1% to 3.3%), but slightly smaller return periods at the end of century, respectively eight and ten years (risk increase to 12.5% and 10%). A smaller but still alarming increase is projected under the SSP1-2.6 scenario, in which a historic 100 year severity is expected to be exceeded on average once every 40 years by mid-century and once every 20 years by the end of the century. This corresponds to an increase in drought risk from 1% to 2.5% and 5% in the mid- and end-century, respectively. Figures 3(b), (d), (f) and (h) display the relative change in drought severity at mid- and end-century compared to the historic period for three return periods: 10, 50 and 100 years. Figure 3(b) shows a positive range of values for the three return periods and both mid- and end-century, indicating an increase in severity compared with current conditions. There is no significant difference, based on the Wilcoxon rank sum test at the 5% significance level, in the median for mid- and end-century under SSP2-4.5 and SSP1-2.6, confirming the previous indication of a stabilization behavior. On the other hand, figures 3(f) and (h) show a pronounced increase in severity from mid- to end-century as well as in comparison to the historic period for SSP3-7.0 and SSP5-8.5. Despite considerable variability among GCM's the median response in the end-century relative change is ranging from 100% to 150%, while the midcentury values range from 0% to 50%.

Zoom In Zoom Out Reset image size

Overall, both SPEI and SPI (see figure S3 and relevant discussion in section S1 in supplementary material) present drying trends towards the end of the century for high emission scenarios, with results more pronounced for SPEI. For South Australia, all scenarios lead to an increasing risk of extreme droughts towards the end of the century. The ability to quantify, through SDF analysis, the expected increase in risk, provides important information for climate adaptation strategies. For example, exposure to future climate risks can be incorporated in monitoring and management plans that identify compounding hazards for at-risk species across Australia, which can be used to prioritize government funding [45, 46].

The analysis of changes in the current 100 year drought towards mid- and end-century revealed significant influence of greenhouse gases emissions. Figure 4 shows the return periods of severity associated with the historic 100 year drought of 12 month duration (multi-model median). Figure 4(a) shows the spatial pattern of SPEI₃ drought severity over a 12 month duration corresponding to a current return period of 100 years. It is possible to observe multiple hotspots of high severity along the Australian territory, e.g. darker areas on central-east and northwestern portions. Figures 4(b) and (c) show respectively the mid- and end-century return periods of today's 100 year droughts under SSP1-2.6. Red colors indicate that the occurrence probability is increasing, while blue colors indicate decreases. Large areas in Australia will experience an increase in extreme drought frequency, with a few exceptions in the Northwest and Central-East regions. There is no significant change in the spatial patterns of mid- and end-century of SSP1-2.6, however a slightly stronger increase in frequency in western and southern regions is observed. Figures 4(d)–(i) present the mid- and end-century return periods of today's 100 year droughts under SSP2-4.5, SSP3-7.0 and SSP5-8.5. Although the pattern for mid-century is very similar for all scenarios, the end-century presents a significant hazard increase in the three scenarios with higher emissions, with SSP5-8.5 standing out for its pronounced increase in frequency. Significantly larger areas in the western and southern Australia present a corresponding return period of almost one year, i.e. under SSP5-8.5, what is now an extreme drought would correspond to normal conditions by the end of the century. Despite the difference in magnitude, all scenarios agree on the drying pattern. Results for SPI (see figure S4 and relevant discussion in section S.1.) also present drying patterns for some regions of Australia, although, as stated previously, changes are significantly larger considering SPEI. An important finding from our analysis (see section S.2.3) is that despite the differences between the PET method used (HS) and the more physically robust method (PM), the increase in drought risk remains in the same order of magnitude.

Zoom In Zoom Out Reset image size