The characteristics of seasonal-scale droughts in Australia, 1911–2009

Authors


A. J. E. Gallant, School of Earth Sciences, University of Melbourne, VIC 3010, Australia. E-mail: agallant@unimelb.edu.au

ABSTRACT

Climatologies and variations in seasonal-scale droughts in Australia are quantified using four indices representing the characteristics of the lower tails of rainfall and soil moisture distributions. These indices estimate variations in drought frequency, duration and intensity from 1911 to 2009 across Australia and for five large-scale regions. Since 1911, large interdecadal variations in the characteristics of seasonal-scale droughts have overlain trends towards less frequent, shorter and less severe droughts across much of Australia, with the strongest trends in northwest Australia. Regional exceptions include increases in seasonal-scale drought frequency, duration and intensity in areas of southwest and southeast Australia. In parts of the west and southeast of the Murray–Darling Basin, the average duration of seasonal-scale droughts, defined as successive seasons in drought, statistically significantly increased by between 10 and 69% during the second half of the 20th Century. Averaged across large-scale regions in southeast and northwest Australia, decades with longer-lasting and more intense soil moisture-based seasonal droughts had statistically significantly higher actual evaporation compared with other decades. These were combined with modest rainfall deficits, suggesting that evaporation may be an important process for regulating drought duration or intensity in these regions. However, other hydroclimatic processes that were not assessed here likely also influence soil moisture, making attribution difficult. Copyright © 2012 Royal Meteorological Society

1. Introduction

The large variations in Australia's hydroclimate lead to regular droughts (Chiew et al., 1998; Kirono et al., 2011). While moisture deficits occur on time scales from months to decades, it is the longer-term deficits that can be particularly detrimental to water resources and the agricultural sector (Nicholls, 2004; Hennessy et al., 2008; Mpelasoka et al., 2008; Verdon-Kidd and Kiem, 2009; Kiem and Verdon-Kidd, 2010).

Decadal-scale variability of seasonal and interannual drought frequency in regional Australia is well documented and previous studies have found large variations within small catchment basins and at individual stations (Chiew et al., 1998; Wooldridge et al., 2001; Kiem and Franks, 2004; Verdon et al., 2004; Hennessy et al., 2008; Mpelasoka et al., 2008; Verdon-Kidd and Kiem; 2010). The regulators of drought have been linked to antecedent land-surface conditions (Timbal et al., 2002) and the characteristics of daily rainfall before and during droughts (Verdon-Kidd and Kiem; 2009, Kiem and Verdon-Kidd, 2010).

These regulators have been further linked to changes in regional circulation patterns. Changes via the El Niño–Southern Oscillation (ENSO), fluctuating sea-surface temperatures (SSTs) of surrounding oceans, and large-scale circulation patterns such as the Southern Annular Mode and Madden–Julian Oscillation have all been described as modulators of seasonal and interannual variations in the Australian hydroclimate (Nicholls, 1988; Chiew et al., 1998; Wooldridge et al., 2001; Hendon et al., 2007; Ummenhofer et al., 2009b; Verdon-Kidd and Kiem, 2009, 2010; Wheeler et al., 2009; Smith and Timbal, 2010). On the decadal-scale, fluctuations in the state of the Pacific Ocean, commonly referred to as the Interdecadal Pacific Oscillation (IPO), are thought to modulate droughts by changing the regional impact of ENSO (Power et al., 1999a, 1999b; Kiem and Franks, 2004; Verdon et al., 2004; Verdon-Kidd and Kiem, 2010).

The socio-economic impacts of droughts are strongly dependent on their characteristic behaviour. For example, a moisture deficit recorded over multiple years will likely contain some months or seasons of average or above-average rainfall, which can temper the effects of the longer-term deficit in the agricultural sector (McIntosh et al., 2007). Thus, recurring seasonal-scale droughts interspersed with months of average rainfall may have a smaller negative impact on agriculture than a single uninterrupted drought. However, factors such as seasonality are also important.

The characteristics of seasonal-scale droughts effectively describe aspects of the lower tails of distributions of moisture-based variables. Studies to date that have examined these characteristics include Kiem and Franks (2004), who found that the frequency of critically low storage levels in reservoirs (30% or less) varied with the phase of ENSO and the IPO for the Williams River Catchment in eastern Australia. Mpelasoka et al. (2008) also defined a measure of drought frequency from the tails of observed rainfall and model soil moisture distributions, reporting increases in central and southeast Australia and decreases in the west from 1970 to 1999.

Although the above studies gave insights into seasonal-scale Australian drought frequency, many other important characteristics such as drought duration, intensity and recurrence interspersed with non-drought periods, remain poorly understood. Our study builds on the previous work above by examining decadal-scale variations in these additional characteristics and provides insight into how they are linked to the broader hydroclimate. Like Mpelasoka et al. (2008), we examine the characteristics of seasonal-scale droughts over the entire the Australian continent. However, the availability of two new datasets since that study was published means that an additional 64 years of data are now available. These additional data are particularly valuable for examining variability on the decadal-scale. Furthermore, this work provides new information on droughts in northwest and parts of central Australia that have been given little attention in previous work.

Meteorological and agricultural seasonal-scale droughts (hereafter referred to simply as ‘droughts’) are defined as extreme deficiencies in 3 month mean rainfall and soil moisture, respectively. Note that our definition of agricultural drought is narrow as it does not include seasonality, nor does it incorporate the effects on crops or livestock, e.g. wilting points. Sections '4.1. Climatologies' and '4.2. Historical variations in the characteristics of seasonal-scale droughts' provide background by presenting climatologies and quantifying historical variations in the characteristics of drought for Australia. Section '4.3. Drought and mean rainfall' examines how these seasonal-scale characteristics are connected to the broader-scale hydroclimate. Specifically, their behaviour with decadal-scale mean rainfall is examined, establishing whether decadal-scale dry periods are associated with more frequent, recurrent, longer and more severe drought in terms of moisture deficits and vice versa for wet periods.

Finally, the relative influence of evaporation on more recent droughts has been debated in the literature. Nicholls (2004), Mpelasoka et al. (2008) and Cai et al. (2009) reported that higher than normal actual and potential evaporation exacerbated recent droughts. These conclusions were disputed by Lockart et al. (2009) citing the importance of other land-surface processes. This debate has continued in the literature (Roderick et al., 2007; Cai et al., 2010; Franks et al., 2010). We contribute to this dialogue by examining the significance of the relative contributions of actual evaporation to the historical variations in the characteristics of agricultural droughts in Section '4.4. The influence of evaporation and rainfall on severe agricultural drought'

2. Data

Deficiencies in seasonal-scale (3 month running mean) rainfall and modelled soil moisture were calculated for the period 1911–2009. Two sources of rainfall data were employed. The first is a mostly spatially complete gridded monthly rainfall data set developed for the Australian Water Availability Project (AWAP) (Jones et al., 2009), which mapped meteorological and hydrological data across the Australian continent (http://www.csiro.au/ science/AWAP.html). The second is a spatially incomplete high-quality station data from the Australian Bureau of Meteorology (Lavery et al., 1997). However, the two data sets are not independent as those stations contained in the high-quality data set are also included in the AWAP gridded set.

Figure 1 shows the locations of the 293 high-quality monthly rainfall stations used to examine meteorological drought. These stations are taken from a larger set of 379 stations that have been rigorously quality controlled and cross-referenced with nearby stations to ensure they contained no artificial inhomogeneities or spurious data (Lavery et al., 1997). In this study, stations were discarded if more than 1 month of data were missing during 10% of years from 1911 to 2009, resulting in the final set of 293 stations. These stations represent the best available rainfall data for climate change analysis and are useful for comparison with the AWAP gridded rainfall data set, now described.

Figure 1.

The map of Australia shows the locations of the 293 high-quality monthly rainfall stations. The five large-scale river catchment regions are outlined in black and are labelled as northwest Australia (NWA), southwest Australia (SWA), the northeast coast (NEC), the Murray–Darling Basin (MDB) and the southeast coast (SEC). The grey shading denotes regions where the AWAP rainfall grids used to generate the meteorological drought indices and the AWAP rainfall input to the WaterDyn model are inconsistent (r < 0.9, see Section '2. Data'). These data have been omitted from the analyses

The AWAP gridded rainfall data set was produced from between 2967 and 7278 rainfall stations across the continent. Rainfall at each station was broken into a monthly climatological average and an anomaly component. The climatological averages and their corresponding anomalies were then each interpolated onto a 0.05° × 0.05° grid using three-dimensional smoothing splines and the Barnes successive-correction method, respectively (see Jones et al. (2009) for further details). Splitting the station data into a mean and anomaly component results in smoother interpolation and more accurate results around topographic features, which is particularly important for parts of eastern Australia (Jones et al., 2009).

Soil moisture is regulated by multiple hydroclimatic processes, such as evaporation, run-off and groundwater recharge, which are potentially important drivers of drought variability. A modelled soil moisture reanalysis data set produced for the AWAP was used as no long-term in situ measurements of soil moisture exist for Australia (Raupach et al., 2008). These reanalyses were produced by the CSIRO and are available at the same spatial resolution as the rainfall grids.

The magnitude of soil moisture variations and the potential maximum moisture capacity of the soil vary considerably across Australia and are dependent on parameters such as soil and vegetation type and soil depth. Soil moisture on the AWAP grids is represented as a proportion of maximum capacity. Zero represents completely dry soil and one represents saturated conditions. Soil moisture is calculated using the WaterDyn water-balance model, which determines the soil moisture in two spatially varying layer depths. The soil layer depths are from empirical measurements of soil properties (Raupach et al., 2008). The upper-layer soil depth is around 0.2 m across much of the continent, while the lower layer typically extends to about 1.5 m. Leaching between the soil layers and the deeper ground, deep drainage from the lower soil layer, surface run-off, evaporation and transpiration from vegetative and soil properties are incorporated in the model. Only the upper-layer of soil moisture is analysed here as this level responds readily to monthly and seasonal-scale variations in the climate (Raupach et al., 2008).

The rainfall data used to initialize the WaterDyn model is an older version of the AWAP gridded rainfall data set previously described and contains fewer stations back in time (Jones et al., 2007). The older (Jones et al., 2007) and the newer versions (Jones et al., 2009) were compared and some differences were found at grid points in western and central Australia where the data was very sparse at the beginning of the 20th Century (Jones et al., 2009; Ummenhofer et al., 2011). The grid points where the correlation between the two versions of the AWAP rainfall grids were < 0.9 represent areas where the precipitation data used to force the WaterDyn model is unreliable back in time and is consistent with a similar analysis by Ummenhofer et al. (2011). These grid points have been masked and are not used in subsequent analyses (Figure 1).

Hereafter, the term evaporation refers to actual evaporation and transpiration from the soil and vegetation rather than potential evaporation that assumes an unlimited supply of moisture (Raupach et al., 2008). The evaporation is calculated as an inverse function of the Priestley and Taylor (1972) potential evaporation rate and is dependent on the fraction of available moisture and the fraction of bare soil. Transpiration from vegetation was taken as the minimum of either water or energy limited transpiration. Energy limited transpiration was calculated as a function of the Priestley–Taylor evaporation rate attenuated by vegetation cover. Water limited transpiration is a function of water extraction by roots, the soil depth and relative soil water (see Appendix A in Raupach et al. (2008) for further details). The Priestley–Taylor evaporation rate is a modified version of Penman–Monteith evapotranspiration (Penman, 1948; Monteith, 1965) and requires observations of surface radiation only. All aerodynamic terms are replaced by an empirically derived constant. Priestley and Taylor (1972) provide further details on the derivation of the constant and Raupach et al. (2008) provides details on the sensitivity of the WaterDyn model output to this parameter.

Direct validation of the WaterDyn model is not possible because of the lack of in situ measurements. However, the model outputs a number of other hydrological variables that are dependent on soil moisture, including catchment outflow, which has been validated against six observed outflow records from small-scale river catchments in southeast Australia (Raupach et al., 2008).

3. Methods

To determine drought at a location, monthly rainfall at each grid point and station, and monthly soil moisture at each grid point were first smoothed using 3 month running periods. These 3 month periods are known, hereafter, as seasons. These seasonal time series were then converted to standardized anomalies, ensuring that droughts would not be biased towards climatologically drier or more variable times of the year. The time series were standardized by removing the climatological (1911–2009) mean and dividing by the standard deviation. From these time series, a drought was defined for season s when the seasonal rainfall or soil moisture anomalies fell in the lowest 10% of all seasons s from 1911 to 2009. Figure 2 shows that in the northwest and central parts of the continent, there were seasons when the 10th percentile of rainfall was 0 mm. These seasons were omitted from further analyses, as they do not represent an unusual deficiency in rainfall.

Figure 2.

The number of seasons (defined here as a 3 month period) for which the 10th percentile of rainfall and soil moisture is > 0 mm (rainfall) or proportion of soil moisture (soil moisture). Drought indices are calculated for seasons where the 10th percentile exceeds zero only

The characteristics of drought were represented by four indices. Multi-year and decadal-scale variations in these characteristics were calculated using running 5, 11 and 15 year sampling periods. The first index, drought proportion, which is a measure of frequency, was calculated as the fraction of all seasons in drought during each 5, 11 or 15 year period and is expressed as a percentage. The second index, drought events, is another measure of frequency that indicates the recurrence of drought interspersed with non-drought periods. Drought events were defined as the number of droughts in a 5, 11 and 15 year period, where an event is defined as 1 to n successive seasons in drought. Drought duration is the third index. It measures the average length, in seasons, of a drought event for each 5, 11 or 15 year period. Note that seasons overlap as they were computed using a 3 month running means. The duration can be converted to the number of independent months by adding two to the number of seasons. The final index, drought intensity, is the average magnitude of the rainfall or soil moisture anomaly during drought events. Although, drought was defined using standardized anomalies, the drought intensity index is expressed as an absolute magnitude of the raw rainfall and soil moisture anomalies, i.e. each value is remultiplied by its seasonal standard deviation. Drought intensity is expressed in millimetres for rainfall and as the proportion of maximum moisture capacity for soil moisture.

The drought indices were also computed using regionally averaged rainfall and soil moisture time series computed for five of Australia's major river catchment regions, which contain drought-sensitive socio-economic and natural environments (Figure 1). These regions encompass the northeast coast (NEC) of Australia, the southeast coast (SEC) of Australia, the Murray–Darling Basin (MDB), the southwest of Australia (SWA) and the northwest of Australia (NWA).

The examination of extremes can potentially be problematic in data sets that have not had the same rigorous testing as high-quality station data. So, to assess the usefulness of the AWAP data for examining drought, the grid points closest to the high-quality station locations only were sub-sampled and used to generate regional catchment time series. From these stations and proximate grid points, area-averaging was performed using Thiessen polygons (Thiessen, 1911). The station weights were time dependent and were recalculated when data were missing from the high-quality station series. From the regional time series the four drought indices were then recomputed as described previously. The catchment-scale drought indices were also computed from the full AWAP grid and there were small differences only with the sub-sampled grid, indicating the station locations are able to capture the catchment-averaged variations.

Local and regional climatologies of the four drought indices were computed using the AWAP rainfall and soil moisture grids and the regional area-averaged drought time series respectively. The climatologies of the characteristics of drought were defined as the average conditions (the mean) and the average magnitude of the variations (the standard deviation) of the running 5, 11 and 15 year time series of each drought index, and were computed using the base period of 1911–2009. The trends in drought indices were computed using linear regression from 1911 to 2009, and from 1960 to 2009 to represent more recent changes. The statistical significance of the trends was determined using the Mann–Whitney test, which assesses the significance of monotonic changes in a time series (von Storch and Zwiers, 1999) and is useful for assessing trends in highly variable data.

4. Results

The characteristics of droughts in Australia are described on the grid point/station and catchment-scale. Most of the results described pertain to the decadal-scale (11 year) indices. However, the sensitivities of the drought indices to the sample length are also described.

4.1. Climatologies

The climatologies for the characteristics of drought illustrate typical drought behaviour in Australia. The climatologies of the four meteorological and agricultural drought indices are shown in Figures 3 and 4. The climatologies for the major river catchment regions are presented for the high-quality data only in Tables 1 and 2. However, the differences compared with the regional AWAP rainfall data were small and did not exceed 0.05% for drought proportion, 0.11 events for drought events, 0.1 seasons for drought duration and 0.84 mm for drought intensity. These results suggest that the AWAP data are useful for examining rainfall-based extremes on seasonal scales.

Figure 3.

The climatologies of meteorological (a) and (b) drought proportion, (c) and (d) events, (e) and (f) duration and (g) and (h) intensity as represented by the mean (left column) and the standard deviation (right column). All climatologies were computed relative to 1911–2009

Figure 4.

The climatologies of agricultural (a) and (b) drought proportion, (c) and (d) events, (e) and (f) duration, and (g) and (h) intensity as represented by the mean (left column) and the standard deviation (right column). All climatologies were computed relative to 1911–2009

Table 1. The climatologies of the 11 year meteorological drought index time series computed from the high-quality rainfall stations. The means and standard deviations (shown in brackets) of the indices from 1911 to 2009 were computed for each of the five large-scale catchment regions defined in Figure 1
 Proportion (%)Events (no. of events)Duration (seasons)Intensity (mm)
NEC9.5 (4.0)6.4 (2.6)2.0 (0.4)41.8 (7.4)
SEC10.1 (4.0)6.3 (2.1)2.1 (0.6)29.7 (2.7)
MDB9.7 (3.8)5.8 (2.3)2.3 (0.7)22.5 (1.7)
SWA9.8 (3.7)7.0 (1.9)1.9 (0.61)15.5 (1.5)
NWA10.3 (3.3)8.9 (2.4)1.5 (0.3)30.8 (9.5)
Table 2. The climatologies of the 11 year agricultural drought index time series computed from the AWAP soil moisture grids. The means and standard deviations (shown in brackets) of the indices from 1911 to 2009 were computed for each of the five large-scale catchment regions defined in Figure 1
 Proportion (%)Events (no. of events)Duration (seasons)Intensity (proportion)
NEC9.3 (3.7)5.5 (2.2)2.3 (0.6)0.10 (0.01)
SEC9.3 (4.1)4.7 (1.7)2.6 (0.8)0.11 (0.01)
MDB9.7 (3.5)5.0 (2.1)2.8 (0.8)0.10 (0.01)
SWA9.7 (4.7)5.0 (1.9)2.5 (0.7)0.07 (0.01)
NWA10.2 (4.4)5.8 (2.2)2.4 (0.7)0.07 (0.02)

As droughts are defined by seasons with the lowest 10% of rainfall or soil moisture anomalies, the climatological mean of the drought proportion index should be around 10%, and this was the case across the country and for all regions. On average, there were fewer meteorological drought events per decade in the climatologically drier central and northwest areas of the continent compared with the wetter south and east. Agricultural droughts typically lasted longer than meteorological droughts by up to two seasons because soil moisture has temporal persistence (Wooldridge et al., 2001; Timbal et al., 2002). The average intensities of meteorological and agricultural drought events had similar spatial variations to the mean rainfall climatology of the continent, with larger deficits in climatologically wetter areas.

4.2. Historical variations in the characteristics of seasonal-scale droughts

The historical fluctuations in the drought indices establish past changes in the characteristics of Australian droughts. An assessment of their sensitivity to sampling choice is also provided. The variations averaged over the catchment regions are shown in Figures 5-9. The descriptions of meteorological drought pertain to the AWAP data only as the differences between this and the high-quality data set were small. The correlations between the two data sets always exceeded 0.82 and the root mean square error was < 2.1%, 1.8 events, 0.25 seasons and 5.5 mm for drought proportion, events, duration and intensity, respectively. The linear trends were computed to detect multi-decadal to centennial-scale non-stationary and are described as a change in the unit per decade over the periods 1911–2009 and 1960–2009. The trends in the catchment regions are listed in Tables 3 and 4, and Figures 10 and 11 show the trends across Australia from 1911 to 2009 only.

Figure 5.

Time series of the decadal (11 year) variations in drought proportion, events, duration and intensity for the northeast coast (NEC) region from 1911 to 2009. The variations are shown for meteorological drought indices computed from the AWAP rainfall (solid line) and for agricultural drought indices computed from soil moisture (dotted line)

Figure 6.

Time series of the decadal (11 year) variations in drought proportion, events, duration and intensity for the southeast coast (SEC) region from 1911 to 2009. The variations are shown for meteorological drought indices computed from the AWAP rainfall (solid line) and for agricultural drought indices computed from soil moisture (dotted line)

Figure 7.

Time series of the decadal (11 year) variations in drought proportion, events, duration and intensity for the Murray–Darling Basin (MDB) region from 1911 to 2009. The variations are shown for meteorological drought indices computed from the AWAP rainfall (solid line) and for agricultural drought indices computed from soil moisture (dotted line)

Figure 8.

Time series of the decadal (11 year) variations in drought proportion, events, duration and intensity for the southwest Australia (SWA) region from 1911 to 2009. The variations are shown for meteorological drought indices computed from the AWAP rainfall (solid line) and for agricultural drought indices computed from soil moisture (dotted line)

Figure 9.

Time series of the decadal (11 year) variations in drought proportion, events, duration and intensity for the northwest Australia (NWA) region from 1911 to 2009. The variations are shown for meteorological drought indices computed from the AWAP rainfall (solid line) and for agricultural drought indices computed from soil moisture (dotted line)

Figure 10.

The linear trends in decadal meteorological drought proportion (% per decade), events (events per decade), duration (seasons per decade) and intensity (mm per decade) from 1911 to 2009 for Australia. The thin black contour lines delinate positive and negative trends. The thick black lines highlight the boundaries of the five large-scale catchment regions

Figure 11.

The linear trends in decadal agricultural drought proportion (% per decade), events (events per decade), duration (seasons per decade) and intensity (proportion per decade) from 1911 to 2009 for Australia. The thin black contour lines delinate positive and negative trends. The thick black lines highlight the boundaries of the five large-scale catchment regions

Table 3. The linear trends in meteorological drought indices from 1911 to 2009 and 1960 to 2009 (in brackets), computed from the AWAP rainfall grids sampled at the same locations as the high-quality station data. Trends have been computed for the five large-scale catchment regions of northeast coast (NEC), southeast coast (SEC), Murray–Darling Basin (MDB), southwest Australia (SWA) and northwest Australia (NWA), defined in Figure 1. Bold indicates the trends are statistically significant at the 5% level (two-tailed) according to the Mann–Whitney test
 Proportion (%(decade)−1)Events (events(decade)−1)Duration (seasons(decade)−1)Intensity (mm(decade)−1)
NEC0.23 (2.01)0.41 (0.92)0.08 (0.17)0.51 (−2.21)
SEC0.02 (1.10)0.27 (0.38)0.11 (0.11)0.04 (−0.63)
MDB0.28 (0.86)0.50 (0.44)0.13 (−0.05)0.02 (−0.92)
SWA0.10 (−0.03)0.23 (−0.32)0.07 (0.05)0.16 (0.37)
NWA0.73 (−2.23)0.55 (−1.85)− 0.02 (−0.06)0.27 (1.72)
Table 4. The linear trends in agricultural drought indices from 1911 to 2009 and 1960 to 2009 (in brackets), computed from the AWAP soil moisture grids sampled at the same locations as the high-quality station data. Trends have been computed for the five large-scale catchment regions of northeast coast (NEC), southeast coast (SEC), Murray–Darling Basin (MDB), southwest Australia (SWA) and northwest Australia (NWA), defined in Figure 1. Bold indicates the trends are statistically significant at the 5% level (two-tailed) according to the Mann–Whitney test
 Proportion (%(decade)−1)Events (events(decade)−1)Duration (seasons(decade)−1)Intensity (proportion(decade)−1)
NEC0.39 (2.01)0.13 (1.65)0.02 (−0.19)0.001 (0.002)
SEC0.06 (1.76)0.29 (0.17)0.17 (0.38)0.001 (0.003)
MDB0.79 (−0.12)0.54 (−0.26)0.13 (0.22)0.002 (0.002)
SWA0.53 (0.43)0.27 (0.11)0.03 (0.05)0.001 (0.004)
NWA1.19 (−1.14)0.57 (−0.81)0.02 (0.29)0.001 (0.004)

The historical variations in drought proportion were broadly similar across eastern Australia. From the late 1910s to the early 1920s and again from the late 1930s to early 1940s, meteorological and agricultural drought proportion was above-average. In the earlier period the largest increases were concentrated in the central and northern areas of eastern Australia, often within the NEC and MDB regions. Here values regularly exceeded 15%, or approximately one standard deviation above-average (Figures 5 and 7). The latter period coincides with a previously identified dry period from 1937 to 1945 (Ummenhofer et al., 2009a; Verdon-Kidd and Kiem, 2009, 2010). At this time, drought proportion was above normal across all eastern Australia. The 1950s were very wet (Hennessy et al., 1999; Kiem et al., 2003; Gallant et al., 2007) and there were also fewer seasonal droughts in eastern Australia. The drought proportion was above-average around 1980 in the far southeast only, and during the late 1990s and early 2000s across most parts of eastern Australia.

From 1911 to 1960 the number of drought events in southeast Australia mostly fluctuated in phase with drought proportion, and average drought duration had only small variations. As drought proportion is effectively the product of drought events and duration, the increases in drought proportion during the first half of the record were mostly due to more recurrent, rather than longer droughts. However, after 1960 the variations in southeast Australian drought events and proportion often diverged, illustrated by the differences in the correlations between the two indices before and after 1960. From 1911 to 1960 the correlations were > 0.94 in the SEC and above 0.80 in the MDB. However, post-1960 these correlations weakened to around 0.60 in the MDB and in the SEC for agricultural drought only.

The change in the relationship between drought proportion and events in southeast Australia over the instrumental record is consistent with the linear trends computed from 1911 to 2009 and from 1960 to 2009. Areas in the southeast and west of the MDB have had statistically significant reductions in the number of drought events and increases in their average duration over both periods. In these areas, comprising approximately one fifth of the total MDB, agricultural droughts were 10–69% longer during the second half of the record compared with the first half. There were similar changes in parts of the southern SEC for agricultural drought only.

The variations in drought intensity were small in southeast Australia and large in northeast Australia, but were mostly unrelated to drought frequency and average duration. Thus, more frequent and/or longer droughts are generally unrelated to their severity in terms of moisture deficits here. There were small but significant decreases in meteorological drought intensity across most of eastern Australia from 1911 to 2009 and 1960 to 2009. The exceptions were increases in agricultural drought intensity in the southeast including in the SEC and eastern MDB (Figure 11).

The variations in the characteristics of drought in west Australia differed from north to south. Drought proportion in SWA was much above normal by at least one standard deviation during the early 1940s, the 1950s, the mid 1970s and the late 1990s and early 2000s (Figure 8). In contrast, these indices were much above normal during the 1930s and 1940s only in NWA (Figure 9). In both regions, average meteorological drought duration was the longest on record during the 1930s/1940s and was up to four seasons, or 6 months, in SWA.

In NWA, the lack of periods with high meteorological and agricultural drought proportion post-1940s was reflected in the long-term trends from 1911 to 2009 and 1960 to 2009. These reductions exceeded 2.0% per decade at some locations. The catchment-wide trends in average drought duration and intensity in NWA reflect that the trends were consistent in sign throughout the region (Tables 3 and 4). The catchment-scale increases in SWA drought proportion and intensity from 1911 to 2009 are dominated by strong local increases in the west and southwest of the region of up to 2.0% per decade and 1.7 mm per decade, respectively. These results suggest that the previously reported drying in that area since the 1970s includes changes to seasonal drought frequency and intensity (Hope et al., 2006).

Although, most of the regional trends in the drought indices were statistically significant, extracting a signal from a highly variable time series is potentially problematic. Regression is sensitive to outliers and consequently large trends can hinge on a single point. To investigate this sensitivity, the linear trends were recomputed over all possible periods from 1911 to 2009. Figure 12 shows the results for MDB drought proportion. In this example, there has been consistency in the sign of the trend for at least the past 80 years, illustrating that the long-term (1911–2009) non-stationarity is robust despite the large interdecadal variations in the time series.

Figure 12.

The linear trends in Murray–Darling Basin drought proportion (% per decade) for (a) meteorological drought computed using the high-quality stations, (b) meteorological drought computed using the AWAP rainfall and (c) agricultural drought computed using AWAP soil moisture. Stippling indicates negative trends. Linear trends were calculated over all possible periods during 1911–2009. The abscissa denotes the year from which the trend was computed, and the ordinate, the length of time over which the trend was computed

The sensitivity of the variations in drought indices to the length of the sampling period, i.e. 5 year, 11 year or 15 year blocks was also examined (Figure 13). In all cases, the variations were consistent between the 11 and 15 year sampling periods. However, there were sometimes differences between the 5 year and the longer samples and the 5 year samples often had short periods with large variations. Typically, one to five drought events only occurred during a 5 year period. So, if just one or two events contained a highly abnormal value of the index, they substantially skewed the 5 year average. The 11 and 15 year sampling blocks were less affected by individual droughts.

Figure 13.

The drought proportion index for the MDB computed using 5 year (light grey), 11 year (middle grey) and 15 year (black) sampling periods

Despite some differences in the variations of the drought indices when computed using different sampling lengths, there were common traits in their multi-decadal variations. For example, average drought duration in southeast Australia was longer in the latter half of the record regardless of the length of the sampling period. The robustness of these multi-decadal variations was further confirmed by the lack of sensitivity of the long-term trends to the sampling interval. Of the 180 regional indices computed (5 regions, 4 indices, 3 sampling intervals, 3 data sets), only 6 (3.3%) had a different sign in their trend between different sampling intervals.

4.3. Drought and mean rainfall

Here, we investigate the strength of covariations between mean rainfall and the characteristics of agricultural and meteorological drought. This analysis determines how the characteristics of drought vary with a process important to the broader hydroclimate. Past studies have identified that soil moisture and mean rainfall do not necessarily co-vary because of the influence of other processes, such as land-surface interactions. Also the regulation of the characteristics of meteorological drought by land-surface factors may be weak but potentially non-negligible (Wooldridge et al., 2001; Timbal et al., 2002). The covariations between mean rainfall and the drought indices were computed as correlations, with results described as the percentage of the variance in drought index that can be explained by mean rainfall (i.e. the square of the correlation).

Mean rainfall explained between 9% (SWA) and 72% (SEC) of the variations in meteorological and agricultural drought proportion, and up to 40 and 70% of the variations in meteorological and agricultural drought events, respectively. The significant amount of the variation in drought proportion and events in some regions (≥70%) makes mean rainfall a reasonable indicator of drought frequency, in particular recurrence, in these areas. Although, it is clearly not the only factor (Wooldridge et al., 2001; Kiem and Verdon-Kidd, 2010). For meteorological drought duration, mean rainfall explained between 0% (SWA) and 25% (SEC and NWA) of the variations, and between 2% (MDB) and 12% (SEC) for agricultural drought duration. For both types of drought, interdecadal variations in mean rainfall explained < 25% of the variations in drought intensity in all regions. Thus, decadal-scale mean rainfall has not provided an indication of the typical length of seasonal droughts or severity in terms of moisture deficits in the historical record.

The relationships between mean rainfall variations and agricultural drought were often stronger than for meteorological drought, which may be associated with the coupled interactions between soil moisture and rainfall (e.g. Timbal et al., 2002; Wooldridge et al., 2001). Resampling each time series with 10% of the data removed and recomputing the correlations tested the significance of these differences. The relationships were deemed significantly different if the 90% confidence intervals of the resampled correlations did not overlap. The results showed few significant differences, suggesting that the apparent differences in the strengths of the relationships between mean rainfall and meteorological and agricultural drought may simply be due to sampling. However, agricultural drought events in the SEC and NWA, drought proportion in the MDB and drought intensity in the SEC showed significantly stronger correlations with mean rainfall variations than meteorological drought. The fact that most of the significant differences occurred in southeast Australia perhaps suggests that the hydroclimate-soil moisture interactions described by previous papers are regionally dependent.

4.4. The influence of evaporation and rainfall on severe agricultural drought

There are numerous hydrological and meteorological regulators of soil moisture, which in turn, regulate agricultural droughts (Wooldridge et al., 2001; Timbal et al., 2002). While not the most important regulator, the debate surrounding the relative influence of evaporation on recent droughts has continued in recent literature (see Section '1. Introduction'). As a contribution to this body of knowledge, we examine the relative importance of evaporation and rainfall during periods of severe drought in the instrumental record. Here, severe drought periods were defined when regional drought indices exceeded their 90th percentile.

The strength of the covariations between the characteristics of meteorological and agricultural drought reflects the influence of rainfall on agricultural drought relative to other hydroclimatic processes driving soil moisture variability. Calculating correlations between the 11 year meteorological and agricultural drought indices using data spanning 1911–2009 determined the strengths of these covariations.

As expected, rainfall was the strongest modulator of seasonal-scale agricultural drought frequency compared with other factors. The correlations between meteorological and agricultural drought proportion and drought events indices were above 0.75 in all regions. The relationships between meteorological and agricultural average drought duration were regionally dependent. In the southern regions there were moderate correlations between 0.53 and 0.69. Conversely, in the tropical NEC and NWA, the correlations were < 0.17. The correlations between meteorological and agricultural drought intensity ranged from 0.07 in the MDB to 0.83 in NWA. The disparity between the variations in meteorological and agricultural drought duration and intensity in some regions might suggest that processes other than rainfall are driving soil moisture-based droughts here.

To test the robustness of the above correlations, they were recomputed using sub-samples of the 1911–2009 time series. Ten percent of time series was randomly removed as a block to retain any temporal dependence and the correlations were recomputed. This was done 1000 times and a 95% confidence interval calculated. All the correlations stated above fell within this confidence interval and, therefore, accurately reflect the strength of the relationships between meteorological and agricultural drought indices.

Soil moisture is computed in the WaterDyn model as a function of multiple hydroclimatic processes of which rainfall is only one (see Section '2. Data'). The influence of these other hydroclimatic processes would cause differences between the meteorological and agricultural drought indices. Some of these processes have been previously described as important regulators of soil moisture and possibly more important than evaporation (Chiew et al., 1998; Wooldridge et al., 2001; Timbal et al., 2002; Kiem and Verdon-Kidd, 2010; Verdon-Kidd and Kiem, 2010). However, although small, evaporation may still play a significant role in drought regulation.

Individual agricultural drought seasons (i.e. 3 month periods with soil moisture in the lowest 10% of all seasons from 1911 to 2009) were extracted and rainfall or evaporation was compared during severe and non-severe agricultural drought periods. Rainfall and evaporation were calculated for every drought season during a severe drought period and the median of each was computed. These median values were then compared with a 95% confidence interval of bootstrapped medians calculated from n random drought seasons occurring during non-severe drought periods, where n was the number of severe drought seasons. A median value outside this confidence interval defined significantly different rainfall or evaporation anomalies during severe drought periods compared with non-severe drought periods.

Unsurprisingly, median rainfall during periods of severe drought intensity was statistically significantly lower than median rainfall during non-severe periods in most regions. The exception was the MDB, where median rainfall was not statistically significantly different. Instead, the evaporation anomalies were 30% larger and statistically significantly different than during non-severe droughts (Figure 14). Compared with non-severe droughts, the evaporation anomalies were also statistically significantly larger for periods of severe average drought duration in the SEC and NWA by over 12 and 23%, respectively. Using the 95th and 85th percentile thresholds to define severe drought further tested our results. In both these cases, median evaporation remained near or above the upper confidence limit.

Figure 14.

The box plots represent the distributions of median seasonal-scale rainfall (center box plots and ordinate) and evaporation (right box plots and ordinate) anomalies during droughts occurring in non-severe drought decades (defined when an index is below its 90th percentile) for the MDB. The whiskers of the plots represent the 95% confidence interval of the distribution, the top and bottom of the box the 75th and 25th percentile respectively and the central line, the median. The black dots represent the median rainfall and evaporation anomalies during droughts occurring in severe drought decades (defined when an index exceeds its 90th percentile)

5. Discussion

The results in Section '4.4. The influence of evaporation and rainfall on severe agricultural drought' suggest that evaporation might be an important contributor to drought duration and intensity in parts of southeast Australia. Our results are consistent with the prescribed Penman–Monteith relation employed in the WaterDyn model. This is, when relative evaporative losses were statistically significantly higher during more severe droughts the negative rainfall anomalies were also lower in magnitude. This is because evaporation is directly related to moisture availability (Priestley and Taylor, 1972; Morton, 1983). The combination of higher relative evaporation and a modest rainfall deficit would not necessarily cause severe soil moisture depletion in isolation, implying other processes must be involved.

Previous studies highlight the importance of antecedent conditions, including hydrological processes and the characteristics of daily rainfall (Wooldridge et al., 2001; Verdon-Kidd and Kiem, 2009; Kiem and Verdon-Kidd, 2010). Such processes may explain the role of combined relatively higher evaporative losses and modest rainfall deficits associated with severe drought intensity and/or duration in the southeast Australia. For example, soil moisture may have already been depleted before the drought. If this was followed by modest rainfall deficits that pushed soil moisture deficits below the threshold for drought, the accompanying relatively higher evaporative losses might make that drought more intense or last longer. However, this concept is speculative and requires further examination that is outside the scope of this study and should form the basis of future work. Also, various regulators of evaporation, such as temperature, radiation, atmospheric moisture or wind were not examined here (Granger and Gray, 1990; Roderick and Farquhar, 2004; Roderick et al., 2007; Kirono et al., 2009). The significant changes to evaporation with severe drought may also be symptomatic of changes to these processes, causing hydrological change. The complex interactions between the climate and hydrology make untangling the causes of droughts stemming from soil moisture deficiencies inherently complex and attributing a single cause difficult.

Seasonal variations in droughts probably show differences compared with annual variations given that the drivers of drought (see Section '1. Introduction') have seasonal signals (Chiew et al., 1998; Kiem and Franks, 2004; Verdon-Kidd and Kiem, 2010). However, the analysis here was not stratified by seasons, primarily because the drought events and duration indices are best calculated with continuous data. For example, drought duration calculated using a traditional 3 month season is limited to a maximum of 3 months and does not record whether the drought continues beyond that season. Recall that the agricultural sector can benefit from a single season of average or above-average rainfall (McIntosh et al., 2007). So, the drought event and drought duration indices are arguably the two most important characteristics to examine for agriculture. However, seasonal analysis can discern variations in drought at agriculturally important times of the year, e.g. the sowing or growing seasons. For this reason, seasonal analyses of some aspects of drought are recommended for future studies.

6. Conclusions

Meteorological and agricultural droughts were identified at the seasonal-scale and were defined using rainfall and soil moisture, respectively. Four indices representing decadal-scale variations in drought frequency, duration and intensity were computed across Australia and for five large-scale regions from 1911 to 2009.

Climatologies, variations and trends in these characteristics of drought were examined. In all areas of Australia, drought frequency and average duration varied considerably from 1911 to 2009. Drought intensity showed large variations in tropical regions only. These large variations have overlain some statistically significant non-stationarities on multi-decade to centennial time scales. Across much of the continent, droughts became less frequent, shorter and less intense from 1911 to 2009 and 1960 to 2009. However, there were several exceptions including far southwest Western Australia, which has had statistically significant increases in drought intensity. The average length of droughts in parts of southeast Australia statistically significantly increased since 1911. During the second half of the 20th Century droughts in these areas were between 10 and 69% longer than droughts during the first half of the 20th Century.

The characteristics of drought vary with other important aspects of the hydroclimate. Decadal-scale mean rainfall variations are a good indicator of meteorological and agricultural drought recurrence across most of Australia. However, mean rainfall variations do not reflect average meteorological or agricultural drought duration or intensity. In parts of southeast and northwest Australia, there were statistically significantly higher relative evaporation rates during severe agricultural droughts compared with non-severe agricultural droughts. These were accompanied by only modest rainfall deficits. Thus, evaporation might play a significant role in regulating these characteristics of drought in these regions. However, attribution is difficult as the influence, interactions and feedbacks between soil moisture and other hydroclimatic processes were not assessed here.

Acknowledgements

Parts of this research are drawn from A.J.E.G.'s Doctoral thesis, which was supported by a Monash University Graduate Scholarship and a CSIRO Top-Up Scholarship. Some further postdoctoral candidacy support for A.J.E.G. was provided by the Australian Research Council through the Discovery Projects scheme (Project FF0668679).

Ancillary