The Hydrologic Reference Stations is a network of 467 high-quality streamflow gauging stations across Australia that is developed and maintained by the Bureau of Meteorology as part of an ongoing responsibility under the Water Act 2007. The main objectives of the service are to observe and detect climate-driven changes in observed streamflow and to provide a quality-controlled dataset for research. We investigate trends and step changes in streamflow across Australia in data from all 467 streamflow gauging stations. Data from 30 to 69 years in duration ending in February 2019 were examined. We analysed data in terms of water-year totals and for the four seasons. The commencement of the water year varies across the country – mainly from February–March in the south to September–October in the north. We summarized our findings for each of the 12 drainage divisions defined by Australian Hydrological Geospatial Fabric (Geofabric) and for continental Australia as a whole. We used statistical tests to detect and analyse linear and step changes in seasonal and annual streamflow. Monotonic trends were detected using modified Mann–Kendall (MK) tests, including a variance correction approach (MK3), a block bootstrap approach (MK3bs) and a long-term persistence approach (MK4). A nonparametric Pettitt test was used for step-change detection and identification. The regional significance of these changes at the drainage division scale was analysed and synthesized using a Walker test. The Murray–Darling Basin, home to Australia's largest river system, showed statistically significant decreasing trends for the region with respect to the annual total and all four seasons. Drainage divisions in New South Wales, Victoria and Tasmania showed significant annual and seasonal decreasing trends. Similar results were found in south-western Western Australia, South Australia and north-eastern Queensland. There was no significant spatial pattern observed in central nor mid-west Western Australia, with one possible explanation for this being the sparse density of streamflow stations and/or the length of the datasets available. Only the Tanami–Timor Sea Coast drainage division in northern Australia showed increasing trends and step changes in annual and seasonal streamflow that were regionally significant. Most of the step changes occurred during 1970–1999. In the south-eastern part of Australia, the majority of the step changes occurred in the 1990s, before the onset of the “Millennium Drought”. Long-term monotonic trends in observed streamflow and its regional significance are consistent with observed changes in climate experienced across Australia. The findings of this study will assist water managers with long-term infrastructure planning and management of water resources under climate variability and change across Australia.
Australia is the driest inhabited continent on Earth, receiving only
450 mm yr
The “Millennium Drought” between 1997 and 2009 is described as the worst
drought on record for south-eastern Australia (Van Dijk et al., 2013).
During 2001–2009, south-eastern Australia suffered the driest period since 1900 – the longest uninterrupted series of years with below-median rainfall (Bureau of Meteorology data;
Global-scale investigation of trends in annual maximum streamflow has revealed that decreasing trends tend to occur in Asia, Australia, the Mediterranean, the western and north-eastern United States, and northern Brazil and that increasing trends appear mostly in central North America, southern Brazil and the northern part of western Europe (Do et al., 2017). Rainfall is one potential driver of these changes. Through analyses of streamflow trends, using data from more than 30 000 gauges across the world, Gudmundsson et al. (2019) found that, where trends are present in a region, the direction of the trend is often consistent across all trend analysis indicators, including high, average and low flows, for that specific region. Analyses of floods and extreme streamflow events across the world show little to suggest that increases in heavy-rainfall events at higher temperatures result in similar increases in streamflow (Wasko and Sharma, 2017). However, shifts in the timing of flood peaks and changes in streamflow timing were found to be consistent with rainfall change and in a similar direction (Wasko et al., 2020).
Australian streams showed the greatest influence of interannual variability in flow (Poff et al., 2006). Chiew and McMahon (1993) examined annual streamflow series of 30 unregulated Australian catchments to detect trends or changes in the means. They concluded that any changes were directly related to interannual variability, rather than to any changes in climate. Analysis of trends in Australian flood data from 491 stations (Ishak et al., 2010) indicated that about 30 % showed trends in the annual maximum flood, with downward and upward trends in southern and northern parts of Australia respectively. Analyses of 780 unregulated catchments (Gu et al., 2020) revealed a similar geographical distribution of trends. Other studies have investigated trends in selected streamflow components in particular regions, such as the south-western region of Western Australia (Durrant and Byleveld, 2009; Petrone et al., 2010) and south-eastern Victoria (Tran and Ng, 2009). In Victoria and the Australian Alps, the assessment and reconstruction of catchment variability and trends in streamflow showed a decline during the period from 1977 to 2012 (Fiddes and Timbal, 2016). A similar spatial pattern in streamflow reduction also occurred during the Millennium Drought between 1997 and 2009. Johnson et al. (2016) reviewed historical trends and variability in flood events across Australia and concluded that the link between trends in flood events and rainfall cannot be made due to the influence of climate processes, such as temperature and evapotranspiration, over different spatial and temporal scales. Similar results were also reported by Wasko and Nathan (2019) and Sharma et al. (2018): changes in rainfall and soil moisture did not always explain trends in flooding.
The studies above undertook trend analysis of Australian rivers with limited
spatial or temporal coverage of streamflow data. This study undertakes a
systematic appraisal of changes and trends in observed streamflow records in
a large number of catchments across the country which are largely unaffected
by human influence. Zhang et al. (2016) undertook the first comprehensive study employing the Hydrologic Reference Stations and analysed data until 2014 from 222 stations across Australia. That study investigated many components of streamflow (including annual total, high and low flows; seasonal totals; and baseflow components); these components were analysed and are presented at
Guidelines for the selection of the Hydrologic Reference Stations (HRS) were described in detail by Turner (2012). These include a minimum of 30 years of continuous data with less than 5 % of missing data in unimpaired catchments. Details of the station selection procedure are provided on the HRS website. In a recent update of the service, the Bureau of Meteorology implemented two additional criteria: (i) the percentage of flow volume included as infilled data and (ii) the percentage of flow volume above the maximum gauged discharge. The thresholds for these two criteria were (i) a maximum of 10 % for flow volume of infilled data and (ii) a restriction of extrapolated data to a maximum of 25 %. Details of the station selection guidelines are presented on the website. All 467 gauging stations that met these guidelines and were included in the updated service are included in this study.
Of the existing 222 gauging stations in the network prior to the service (Zhang et al., 2016), 12 have now been decommissioned by the data-provision agencies, including 5 from the Northern Territory, 2 from Victoria, 2 from Western Australia, 1 from New South Wales and 2 from South Australia. The revised and updated selection guidelines were applied to all 222 gauging stations, and 179 of these stations passed the new guidelines. More information about the rating curves of 43 existing stations became available and suggested that these stations did not pass the selection criteria. Therefore, these 43 stations were removed from the HRS service in 2020 (Table 1).
Total number of stations.
Australia's instrumental record is relatively sparse before 1940, and few locations have continuous rainfall measurement before 1900. At present, there are approximately 4800 streamflow gauging stations across Australia. Many of these stations previously had insufficient data in the Bureau's system to be considered for the HRS service. Over recent years, more data have become available, and a set of 780 stations identified by Zhang et al. (2013) were selected for further investigation. These 780 unregulated, unimpaired catchments are widely spread across Australia (Fig. 1) and have undergone strict quality assurance/quality control (QA/QC), including quality code checking for daily streamflow records (Zhang et al., 2013). The total number of streamflow gauging stations that met the new guidelines and are now included in the service is 467, which is an increase from 222 stations (Table 1).
Map of Australia showing climate zones, drainage divisions and the location of all gauges (new, existing and decommissioned), all scaled to length of record.
The HRS service was updated with streamflow data ending in February 2019. This month was chosen to capture the 2018 water year for all stations. Commencement of recording of streamflow data varies across the country. The longest record begins in the 1950s and the shortest begins in the 1980s. Figure 1 shows the location of stations and the duration of the records, including the eight decommissioned stations. Data recorded at stations prior to 1950 were excluded from analysis, as the missing data were more than 5 % before this time in most cases. Stations with the longest records are generally those in the high-value water resource catchments and populated areas along the coastal regions. Upstream catchment area also varied across the country (from 4.5 to 232 846 km
Metadata for drainage divisions and selected hydrologic reference stations.
Continental Australia has a wide range of climate zones, as defined by
Köppen climate classification (Stern et al., 2000), including a tropical region in the north, temperate regions in the south, and grassland and desert in the vast interior (Fig. 1). In this work, water year is defined in accordance with the Hydrologic Reference Stations' glossary
(
Catchment area plotted against
A QA/QC process was applied to observed time series of daily streamflow from each gauging station. This process identified and removed erroneous data values such as negative and extreme values. The process of detection and removal was automated and then checked manually, as detailed on the Hydrologic Reference Stations website (
Quality-controlled and infilled daily data were accumulated to annual totals (based on water year) and to the total for each of the four seasons. These five streamflow statistics formed the basis for trend analysis to capture annual and seasonal trends. Seasons are defined as winter (June–August), spring (September–November), summer (December–February) and autumn (March–May).
There are generally two types of change often observed in streamflow datasets: long-term persistent trends (monotonic change) and sudden abrupt change (step change). Monotonic trends generally occur due to long-term changes in rainfall, temperature and/or evapotranspiration, whereas step changes generally result from sudden changes in flow generation thresholds. Statistical tests used to detect monotonic trends and step changes in streamflow generally fall into two categories: parametric and nonparametric tests.
We applied the Theil–Sen approach (Sen, 1968; Theil, 1992), a nonparametric technique, to detect the magnitude of the linear trend for each of the statistics (the annual total and the four seasons' totals). In addition, trends were determined using the nonparametric Mann–Kendall (MK) test (Kendall, 1975; Mann, 1945), as this technique is distribution-free, robust against outliers and has a higher power for non-normally distributed data (Önöz and Bayazit, 2003; Yue et al., 2002). It has also been commonly used in streamflow trend analyses (Abdul Aziz and Burn, 2006; Birsan et al., 2005; Dixon et al., 2006; Lins and Slack, 1999). The MK test requires input data to be serially uncorrelated; any serial correlation in the data structure can lead to an overestimation of the significance of trends (Hamed and Ramachandra Rao, 1998; von Storch, 1995; Yue et al., 2002). To overcome the effect of serial correlation of higher order (namely short-term persistence – STP), two techniques are used here: (i) a modified MK test, commonly known as the variance correction (MK3) method (as proposed by Hamed and Ramachandra Rao, 1998), and (ii) a block bootstrap (MK3bs) method (Kundzewicz and Robson, 2000). The MK3 and MK3bs approaches are more suitable when the time series shows higher-order serial dependencies. Apart from short-term persistence (STP), the presence of long-term persistence (LTP), or the Hurst phenomenon, has been identified as a major source of uncertainty when analysing hydroclimatic data series (Koutsoyiannis, 2003; Kumar et al., 2009). To incorporate LTP behaviour in the MK test, the technique proposed by Hamed (2008) is used in this study. Employing these three modified versions of the MK test allows for a comparison of differences and also allows the validity of results for Australia to be checked. A nonparametric Pettitt test (Pettitt, 1979) and distribution-free CUSUM (cumulative sum) test (Chiew and Siriwardena, 2005) are generally used to detect abrupt step changes in streamflow. A Pettitt test was used to detect step change for all five streamflow statistics, as it performs better than others with respect to detecting step change and identifying the change point (Villarini et al., 2009). Finally, the regional significance of any monotonic trends and step changes was tested using a Walker test (Wilks, 2006).
The Theil–Sen slope approach, the original Mann–Kendall (MK) test, three modified versions of the MK test, the Pettitt test and the Walker test are briefly described in this section. Trend detection analysis may lead to misleading results when serial correlation (STP) and long-term persistence (LTP) in the streamflow data are ignored. The modified versions of the Mann–Kendall test, MK3, MK3bs and MK4, which account for STP, STP and LTP respectively, are used to identify trends in streamflow data. A more detailed description can be found in Mann (1945), Kendall (1975), Hamed and Ramachandra Rao (1998), Koutsoyiannis (2003) Hamed (2008), Pettitt (1979), Wilks (2006), Kumar et al. (2009), Zamani et al. (2017), Su et al. (2018), and Kundzewicz and Robson (2000).
We used the Theil–Sen estimator and three different forms of the Mann–Kendall test, (i) a variance correction approach, (ii) a block bootstrap approach and (iii) long-term persistence, for monotonic trend analyses.
The magnitude of the trend is obtained using the Theil–Sen approach (Theil, 1992; Sen, 1968), where the magnitude of the slope of the trend is estimated
as follows:
The Mann–Kendall test statistic (
The modified Mann–Kendall variance correction approach, proposed by Hamed and Ramachandra Rao (1998), considers all of the significant autocorrelation structure in a time series. The series
The block bootstrap (BBS) approach by Önöz and Bayazit (2003) was used to mitigate the effects of serial correlation in datasets by performing bootstrapping (in blocks of data) so that the autocorrelation in the data is replicated. Data are resampled in blocks many times to estimate the significance of the observed Mann–Kendall test statistic
In block bootstrapping, the simulation size
Autocorrelation in time series varies across different sites, and Politis (2003) recommends that the block length
The optimal block length for the annual streamflow time series and the four seasonal streamflow time series for all 467 stations.
We estimated the significance of the original Mann–Kendall test statistic
The long-term persistence (LTP) version of the Mann–Kendall method was proposed by Hamed (2008) and described by Kumar et al. (2009), and it considers the Hurst coefficient (Hurst, 1951),
In this study, the following steps were carried out to apply the MK4 test:
We used the nonparametric Pettitt test (PT; Pettitt, 1979) to detect step
changes in annual and seasonal streamflow. This procedure is least sensitive
to outliers and to skewed distributions of the streamflow datasets, in
comparison with other methods used for detecting step changes, which makes it most
suitable for this analysis (Sagarika et al., 2014). This test can identify anomalies in the mean and/or median streamflow when the time of step change is unclear. It uses a version of the Mann–Whitney statistics to quantify the significance of probabilities by testing two samples from the same population. In the Pettitt test, the
Using Pettitt (1979), let us assume
In this study, we adopted a significance level of
We assessed the regional significance of trends detected at a local point
scale to examine if similar trends were also detected at neighbouring locations. The main objective is to assess whether a certain minimum number
of locations with significant trends occur at a regional scale, thereby indicating
regional significance, or not. The test for regional significance is rather
conservative, as it is based on a null hypothesis of independent trends
across stations, when the trends within a region are positively correlated.
We applied a Walker test (Wilks, 2006; Sagarika et al., 2014) to detect
regional significance of monotonic and step changes in streamflow time
series – for the annual total and across the four seasons at a 90 % confidence level (
The Walker test considers a set of
Box plot of the Theil–Sen slope in millimetres per year per year for annual and seasonal streamflow for drainage divisions in
For the water year and the four seasons, the monotonic trends, considering short-term persistence (STP) and long-term persistence (LTP), and the step changes were evaluated at a significance level of
Box plots of trend slopes estimated by the Theil–Sen approach for annual and
seasonal streamflow in the southern and northern divisions that were significant are shown in Fig. 4. For example, the trend slope for a site in the South Esk River in Tasmania is shown in Fig. 5a, where the decrease in annual streamflow is 5.91 GL yr
Typical examples of
Medians of slopes for annual and seasonal streamflow were located close to
or above zero for drainage divisions in the northern part of Australia
(Fig. 4b). The highest (
Streamflow data from stations that are significantly autocorrelated (lag 1
or more) at
Spatial distribution of stations where streamflow shows persistence, short-term persistence (STP) and long-term persistence (LTP), in
Data from stations with a significant correlation at
Summary of stations with short-term persistence (STP) and long-term
persistence (LTP) of streamflow and of stations that showed trends using the MK3 and MK4 tests across drainage divisions for the water year and the four seasons at
Table 4 summarizes the MK3, MK3bs and MK4 test results. Figure 7 shows the
distribution of trends for each drainage division for the water year,
autumn, winter, spring and summer for all three MK tests. The MK3 test
results are similar to the MK3bs test results for water years and all four
seasons, as they both consider the full autocorrelation structure (STP) of
the streamflow series. The spatial distribution of trends using all three MK tests in the water year suggests that the annual mean streamflow has increased in the Tanami–Timor Sea Coast division in northern Australia, whereas it has decreased in the south-western and south-eastern parts of the country (Fig. 7). The magnitude trends in streamflow volumes expressed by the Theil–Sen slope show a maximum increase of 2.4 % yr
Results of the three Mann–Kendall (MK) tests, MK3, MK3bs and MK4, for the water year and all four seasons. The numbers in each column represent the number of stations with increasing trends (denoted using “
Entries in bold indicate results that are regionally significant at
Maps showing seasonal trends using the
Although most of the 467 streamflow stations showed trends, only 14 stations had data with increasing trends across water years, and 212 stations showed decreasing trends that were statistically significant (Table 4). With respect to water year, more than 50 % of stations with increasing trends were located in the Tanami–Timor Sea Coast (VIII) drainage division. Streamflows that showed decreasing trends were within the South East Coast (II), Tasmania (III), Murray–Darling Basin (IV), South Australian Gulf (V) and South West Coast (IV) drainage divisions. Most stations in other divisions showed no significant trend (Fig. 7)
For autumn (March–May), streamflows at more than 50 % of stations showed
significant trends in the South East Coast (II), Tasmania (III), Murray–Darling Basin (IV), South Australian Gulf (V), South West Coast (IV) and North East Coast (I) divisions. Other divisions showed autumn
decreasing trends that were very similar to those found across water years,
except for the Tanami–Timor Sea Coast (VIII) in the north, where only one
station showed an increasing trend. Data from 173 stations showed significant trends across water years, 4 of which were increasing and 169 of which were decreasing (Table 4). In autumn, the maximum increase in streamflow was 1.3 % yr
Across winter (June–August), streamflows at over half of the stations showed
significant decreasing streamflow trends in the North East Coast (I),
Tasmania (III), Murray–Darling Basin (IV), and South West Coast (IV) divisions, whereas they showed
significant increasing trends in the Tanami–Timor Sea Coast (VIII) division. However,
in the South East Coast (II), Pilbara–Gascoyne (VII), Carpentaria Coast (IX)
and the Lake Eyre Basin (X) divisions, fewer than 50 % of stations had flows with
significant trends. A total of 189 stations showed significant streamflow trends (for MK3) across Australia, 9 of which were increasing trends and 180 of which were decreasing trends (Table 4). The maximum increase in winter flows was 2.3 % yr
Spring (September–November) saw more stations with decreasing flow trends compared with water years or the other seasons. Trends in flow were mostly detected at stations in all divisions except for Lake Eyre Basin (X) and North Western Plateau (XI), for which there is very limited flow data. The South East Coast (II) and Murray–Darling Basin (IV) divisions had the highest number of stations with decreasing trends in streamflow, compared with trends in the water year and other seasons (Table 4). There were stations in the South East Coast (II), Tasmania (III), Murray–Darling Basin (IV) and South
Australian Gulf (V) that showed significant decreasing trends. For spring
(September–November), 235 stations showed significant trends: 8 increasing and 227 decreasing (Table 4). The maximum increase in spring flows was 3.1 % yr
For summer (December–February), stations in the Tanami–Timor Sea Coast (VIII) division showed significant increasing flow trends, as with trends in other seasons. Stations in the South East Coast (II), Tasmania (III), Murray–Darling Basin (IV), South Australian Gulf (V) and South West Coast (VI) divisions showed significant decreasing trends in flow. In summer, 175 stations showed significant flow trends: 26 increasing and 149 decreasing (Table 4). The maximum increase in summer flows was
While results from the MK3 and MK3bs tests are nearly identical for most cases, MK3 and MK4 do show differences for most annual and seasonal flow statistics. A typical example of the MK3bs statistic obtained from 2000 samples from two locations with (a) decreasing and (b) increasing trends is shown in Fig. 8. It illustrates how the observed values are located below or above the 5th and 95th percentiles respectively. A small number of stations have significant streamflow trends when LTP behaviour (MK4) is considered (Fig. 7, Table 4). This is particularly evident in the South East Coast (II) division for water years and in the Murray–Darling Basin (IV) division for autumn (Fig. 7, Table 4). The MK4 test resulted in more stations with trends in winter and spring, compared with autumn and summer. Across Australia for water years, 14 stations showed significantly increasing trends in flow, and 196 stations showed significantly decreasing trends, slightly less than the MK3 and MK3bs test results (Table 4). Similar results are also evident for the four seasons across Australia (Table 4).
Examples of histograms for two stations representing the frequency distribution of the MK statistic (
The nonparametric Pettitt test (Pettitt, 1979) was used to test for step changes for water years as well as for all seasons. This test is biased towards finding step changes in the centre of a time series (Mallakpour and Villarini, 2016). As the annual streamflow has a skewed distribution and possibly has outliers, this nonparametric test, which is least sensitive to these characteristics, is well suited to change-point detection. A typical example of step change in northern Australia is shown in Fig. 5b. Step-change maps (Fig. 9a) clearly reveal a spatial pattern in the location of stations that exhibited a significant step change in flow. The direction and significance of step changes are consistent with results from the MK tests (Fig. 7) for most stations. Years when step changes occurred show spatial groupings within several drainage divisions. Significant step changes or shifts across water years as well as for all four seasons for each drainage division are summarized in Table 5. For water years, the South East Coast (II), Murray–Darling Basin (IV), South Australian Gulf (V), South West Coast (VI) and Tanami–Timor Sea Coast (VIII) drainage divisions all showed significant step changes for more than 60 % of station flows (Fig. 9a, Table 5). Increasing shifts were seen in northern Australia, including the Indian Ocean (VII), Tanami–Timor Sea Coast (VIII) and Carpentaria Coast (IX) divisions, whereas decreasing shifts were seen in all drainage divisions except for the Tanami–Timor Sea Coast (VIII), Carpentaria Coast (IX) and North Western Plateau (XI). The South East Coast (II), Murray–Darling Basin (IV), South West Coast (VI) and Tanami–Timor Sea Coast (VIII) divisions had significant step changes across water years (Fig. 9).
Annual and seasonal shifts in different drainage divisions using a Pettitt test at
Entries in bold indicate results that are regionally significant at
For summer (December–February), the lowest proportion of stations had significant step changes in flow compared with the other seasons or with water years (Table 5). The South East Coast (II), Murray–Darling Basin (IV) and South West Coast (VI) divisions had significant step changes at
The timing of step changes and the number of stations with step changes for different drainage divisions with
Figure 9 shows the number of stations with step changes in flow, each year
between 1950 and 2018, for water years and for each season. The first step change was detected in 1964, and most changes occurred between 1970 and 1999. Of 467 stations, a total of 253 stations showed step changes in flows for water years (
Autumn (March–June) had a total of 210 stations with step changes (
In winter (June–August) 246 stations showed a step change (
Spring (September–November) had a total of 219 stations with step changes
(
Summer had a smaller number of stations (160 out of 467) with step changes
(
Figure 10 shows the timing of step changes for several drainage divisions where the water years mostly range from 1976 to 2009. The period from 1990 to 2000 showed decreasing step changes for 199 stations, mostly in the South East Coast (II), Murray–Darling Basin (IV) and South West Coast (VI) divisions. Most increasing step changes are in the Tanami–Timor Sea Coast (VIII) division during 1992–1998. The South East Coast (II), Murray–Darling Basin (IV), South West Coast (VI) and Tanami–Timor Sea Coast (VIII) divisions showed a greater number of stations with step changes distributed across the study period, indicating a higher association with natural changes and climate variability than for other regions (Table 5).
After the step changes, the distribution of mean monthly streamflow within a water year changed across all drainage divisions. The distribution of mean monthly streamflow at four selected gauging stations from four drainage divisions is shown in Fig. 11. In the northern part of Australia, the increase in streamflow is well distributed over most of the months. However, in south-western Queensland and the southern part of Australia, there seems to be a phase shift as well.
Distribution of mean monthly streamflow before and after step changes at four selected gauging stations representative of the
Table 6 summarizes the regional significance of monotonic trends and step changes within each drainage division (at the drainage division scale) for water year and each of the four seasons. The Murray–Darling Basin (IV), South East Coast (I) and South West Coast (VI) divisions experienced decreasing trends for water year as well as for all seasons across all statistical tests. Similarly decreasing trends were also evident in the North East Coast (I), Tasmania (III) and South Australian Gulf (V) divisions with the exception of only a few statistical tests (Figs. 7, 9). In the North East Coast (I) division, only the MK3 test detected no significant annual trends. In the Tasmania (III) division, no step changes (PT) nor monotonic trends (MK4) were detected for spring and summer respectively. In northern Australia, the Tanami–Timor Sea Coast is the only division where most of the statistical tests show increasing trends for water years and for all seasons (Table 6). At annual scales, the MK3bs, MK4 and PT tests showed a significant increasing trend. With the exception of autumn, the MK3 and PT tests detected increasing monotonic trends for the other three seasons. There were no noticeable patterns in divisions in central Australia, including the North Western Plateau (XI) and Lake Eyre Basin (X) divisions. Trends in the Carpentaria Coast (IX) division were varied, and only summer saw a significant decreasing trend in flows (MK3). The lack of streamflow observations and relatively low number of stations, the length of records, or the number of non-zero-flow days in these central Australian divisions may play a role in detecting the trend.
Summary of regional significance across different drainage divisions for water year and the four respective seasons.
MK3, MK3bs and MK4 correspond to MK tests. PT corresponds to a Pettitt test. Entries in bold and roman font indicate respective positive upward and negative downward trends at
To maintain the quality of streamflow data, only gauging stations with less than 25 % of flow volume above the recorded maximum gauged discharge and less than 10 % of infilled flow volume were included in this study and in the Hydrologic Reference Stations service. A quantitative evaluation of the accuracy of the gap-filling procedure showed that gap filling by interpolation or simple rainfall–runoff modelling is most accurate and effective if the missing record is less than 10 % (Zhang and Post, 2018), which was followed in this study. However, when the percentage of gap-filled data represent the relatively high end of the flow volume, the uncertainty becomes higher. McMahon and Peel (2019) analysed uncertainties associated with streamflow by considering the gauged and extended range of the stage–discharge relationship from 622 rating curves for 171 of the HRS gauging stations. They found that estimated flow volumes beyond the gauged range of the rating curve occurred at many of these gauging stations and caused measurement uncertainty. As our analysis is limited to seasonal and annual totals, implications of the estimated volumes beyond the gauged range may be minimal.
The MK3 and MK4 tests show that there are strong autocorrelations or short-term persistence (STP) and significant long-term persistence (LTP) respectively in Australian streamflow in most drainage divisions (Fig. 6, Table 3). About 85 % of stations had data with a significant autocorrelation structure (Table 3), and about 40 % of these stations had more than lag-1 autocorrelation (Fig. 3). As such, the MK2 test (Kumar et al., 2009; Su et al., 2018) was not considered in the analysis, as it considers only the lag-1 autocorrelation structure. Around 30 % of stations had significant LTP only (Table 3). Therefore, two variations of MK3, which considers the total autocorrelation structure (STP), and MK4, which considers LTP behaviour, were used in the analysis. Our calculated trends are strongly affected by STP and LTP factors. For water years, there was a clear reduction in the number of stations with significant streamflow trends when the full autocorrelation structure (MK3) and LTP behaviour (MK4) were considered in the analysis: 263 stations with MK1, 226 with MK3 and 210 with MK4. A similar reduction was observed for most of the seasons (Table 4). Around half of stations with STP and only about 12 % of stations with LTP had significant trends for water year as well for all four seasons. The application of these different MK tests helped to differentiate trends that exist under the assumption of serial independence and short- and long-term persistence.
The Pettitt test showed that 54 % of stations experienced an abrupt shift in streamflows across water year. Most of these step changes (51 %) were decreases and were observed at more stations than for monotonic decreasing trends (38 %). Most of the increasing step changes took place in spring (September–November) and summer (December–February), whereas most of the decreasing step changes occurred in winter (June–August). The period from 1970 to 1999 had the largest number of step changes for water-year flows. The greatest number of stations (21 %) with a decreasing step change was found in 1996, whereas most of the increasing shifts (2 %) occurred in 1996. Seasonally, a greater number of stations with significant decreasing step changes in flow occurred in winter (Figs. 9b, 10).
For annual timescales, monotonic trends and step changes were similar in
direction across the country; however, slight variation in direction was
observed between seasons and different river basins. These results are consistent with previous findings from other studies on Australian streamflows (Zhang et al., 2016), despite having a larger number of gauge locations in all drainage divisions except the South West Plateau, different analyses methods, and different lengths and coverage periods of data (Durrant and Byleveld, 2009; Petrone et al., 2010; Tran and Ng, 2009). These findings are also consistent with the following observed changes in rainfall in different parts of Australia (BoM and CSIRO, 2020): (i) northern Australia has become wetter, particularly in the north-west, with rainfall generally above average in the dry seasons; (ii) Australia's climate has warmed on average by about 1.5
Monotonic changes in streamflow may generally occur in response to long-term
changes in the flow generation processes, including increases or decreases in
rainfall, evapotranspiration, temperature, humidity and changes in vegetation dynamics. Note that the catchments considered here are unimpaired, so the impacts of anthropogenic activities are not significant. For some of the locations, the decreasing step changes coincide with the Millennium Drought between 1997 and 2009, but some catchments underwent changes a few years before that period (Fig. 10). Streamflow deficits in the Murray–Darling Basin were very large during the drought, with an estimated return period of 1 in 1500 years (Gergis et al., 2012), which may be one of the reasons for the largest proportion of stations (96 out of 133) depicting step changes. Sudden and abrupt changes, including step changes in the streamflow generation process, happen due to changes in thresholds, which are most prominent when the filling season begins, which is in autumn (March–June) in the south-west of Australia (Silberstein et al., 2012) including Tasmania (Fig. 9, Table 5). Also evident is a phase shift in the streamflow distribution in south-western Western Australia (Fig. 11d) due to a reduction in rainfall and lateral shift within the year (Silberstein et al., 2012). A reduction in winter rainfall, an increase in temperature and extreme rainfall are projected to continue in the future due to climate change (BoM and CSIRO, 2020). Climate change could make extreme events more frequent, causing an increasing streamflow trend for the wet season, a decreasing trend for the dry season and no trend for the annual basis. An increasing atmospheric
The frequency and magnitude of extreme rainfall events have increased across Australia (Sharma et al., 2018), and further investigation is required to understand how this translates to maximum flow and flood extremes in different parts of the country. Changes in the frequency and duration of low flows are also of interest to water managers, particularly in maintaining environmental water flow requirements. Questions remain in relation to phase shifts in streamflow generation within the year (Fig. 11), in particular in the southern part of Australia. For example, are these shifts related to rainfall distribution alone or are there other contributing factors, such as changes in temperature or secondary changes in vegetation dynamics? The influence of step changes on gradual trends was not considered in this study. Further investigation to identify the conditions and processes that result in these step changes would be of great value. Research into these questions is important to guide better management of water resources, infrastructure development and environmental water allocation. The statistical analysis undertaken in this study depends on the length of the time series available, which varies from one station to the next (Fig. 1). In future analyses, streamflow data obtained before the 1950s should be considered in order to identify decadal persistence, variability and trends. A better understanding of the nature of trends and step changes in seasonal streamflow can support better regional water management to regulate the flows and maintain adequate levels in reservoirs during dry and wet periods.
Trends in annual and seasonal streamflow over Australia at 467 high-quality gauging stations listed in the Bureau of Meteorology's Hydrologic Reference Stations service were analysed and presented. The length of the daily streamflow data record ranged from a minimum of 30 years to a maximum of 69 years. Monotonic trend analyses were performed using three different forms of the Mann–Kendall test: a variance correction approach (MK3), a block bootstrap approach (MK3bs) and a long-term persistence approach (MK4). Identification and detection of step changes for seasonal and annual streamflow were accomplished using a nonparametric Pettitt test. The regional significance of these changes was analysed at the drainage division scale for different divisions and synthesized using a Walker test.
Monotonic decreasing trends in annual and seasonal streamflow at most of the gauging stations were detected in the Murray–Darling River basin and other drainage divisions in New South Wales, Victoria and Tasmania. Similar results were observed in south-western Western Australia, South Australia and in north-eastern Queensland. Decreasing trends in annual totals and in totals for all four seasons were also regionally significant at the drainage division scale. Only the Tanami–Timor Sea Coast drainage division in northern Australia showed increasing trends and step changes in annual and seasonal streamflow, and these were regionally significant. There were no significant spatial nor temporal patterns observed in central and mid-west Western Australia. One possible reason for this is the sparse density of streamflow stations and the length of the data record available for analysis.
In general, step changes were similar to the direction of monotonic trends across Australia. Only a handful of stations in northern Australia showed significant step-change increases in streamflow. At regional scales, that increasing trend was only statistically significant in the Tanami–Timor Sea Coast drainage division. Across southern Australia, most step changes occurred during the period from 1970 to the 1990s, with the majority of them being in 1990s, before the onset of the Millennium Drought in 1997. Most of the step changes occurred just at the start of winter when the rainy season begins.
Further investigation and research would assist in understanding the processes that govern the detected changes in flow generation, catchment memory and its interaction with rainfall change, increase in temperature and vegetation growth, and evapotranspiration.
The code and software used in this research are not available to the public. However, the output results can be provided upon request via the “Feedback” page of the Hydrologic Reference Stations website:
Data for each station can be obtained from
GEA undertook data curation, formal analyses, investigation, methodology, validation, visualization and writing. MAB contributed to conceptualization, investigation, methodology, project administration, resources, supervision, validation and writing. FW contributed to data curation, formal analyses, investigation, methodology, validation and visualization. PMF provided project administration, resource allocation, supervision, validation, and manuscript review and editing.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Streamflow data were initially provided by national, state and territory water agencies across Australia. The Hydrologic Reference Stations website was developed in consultation with The University of Melbourne, CSIRO Land and Water, the Department of Climate Change and Energy Efficiency (DCCE), and approximately 70 other stakeholders. We thank Tom A. McMahon for his ongoing contribution to the HRS technical review. We express our sincere thanks to the editor, Julien Lerat, Margot Turner and the Bureau of Meteorology's internal reviewers for their careful review and valuable comments and suggestions on the submitted version of this paper. We also thank the three reviewers, whose comments and suggestions greatly improved this paper.
This paper was edited by Yi He and reviewed by Nir Krakauer and one anonymous referee.