Streamflow decline in southwestern Australia, 1950–2008

Authors


Abstract

[1] Southwest Western Australia (SWWA) has experienced a 15–20% reduction in rainfall since the 1970s with severe reductions in inflows to Perth drinking water reservoirs. To quantify rainfall and runoff patterns, we used trend and change point analyses for a 50 year record (1950–2008) and in the last two decades (1989–2008). From 1950–2008, trend tests showed significant declines in annual rainfall and runoff with corresponding change points for both rainfall and flow in the late 1960s or mid-1970s. In the more recent record (1989–2008), runoff declined in the majority of catchments, but rainfall did not show a significant downward trend. Rather, streamflow decline was observed as a step change in response to the occurrence of below-average rainfall years. A shift from perennial to ephemeral streams and a decline in the runoff coefficient (runoff/rainfall) in the last decade suggests a new hydrologic regime has developed with important implications for future surface water supply.

1. Introduction

[2] Surface water availability is projected to decrease with climate change in arid and semi-arid regions of the world and, at the same time, population growth will place an increasing demand on water supplies [Vorosmarty et al., 2000]. This scenario is particularly true in SWWA where Perth metropolitan drinking water catchments supplied up to 80% of the ∼125GL demand in the 1970s, but despite new dam construction now comprise less than 40% of the ∼250GL demand that is supplemented with groundwater and desalinated sea-water. Overall, runoff from Perth's drinking water catchments has declined up to 70% in the last 40 years that is associated with a 15–20% rainfall reduction [Bates et al., 2008]. Potential causes of rainfall decline have been well documented in this region, including changes in the frequency of synoptic weather troughs [Hope et al., 2006] and large-scale weather patterns that may be partly driven by anthropogenic climate change [Cai et al., 2005; van Ommen and Morgan, 2010]. However, relatively little is known about the relationship between rainfall and streamflow. The factors influencing streamflow in this region are important because Perth's population (1.5 M in 2009) is projected to double by 2050, and water demand is increasing more than 3% per year.

[3] Catchment runoff is largely influenced by the balance of precipitation and evapo-transpiration (ET), and the ratio of long-term average rainfall to potential evaporation (PE) has been used to infer runoff across regions [Zhang et al., 2001]. Within a specific region, climate-soil-vegetation interactions influence how rainfall becomes runoff, and changes in vegetation [Bosch and Hewlett, 1982] or soil water and groundwater storage [Latron and Gallart, 2008] may affect runoff response. In regions where PE exceeds rainfall, and runoff rates are low, subtle changes in the long-term water balance may have profound effects on catchment water yield. For example, in the Jarrah forest of SWWA a relatively small proportion of runoff (3–20%) is generated from rainfall [Ruprecht and Stoneman, 1993], and reduced rainfall may exacerbate low runoff efficiency. In this study, we use change point and trend analysis to examine an extensive dataset of rainfall, dam inflow to Perth metropolitan drinking water catchments, and flow in unregulated streams since the 1950s. Previous research has established the pattern and cause of recent rainfall decline in SWWA [Bates et al., 2008]. Here our goal is to build on these studies by analysing both precipitation and runoff records to better understand hydrologic consequences of a changing climate on the Darling Plateau that provides Perth's main surface water supply.

2. Study Site, Data, and Methods

[4] The Darling Plateau is a 300m elevation ridge that runs north-south 20 km inland from the coast, and serves as the surface water supply for the Perth metropolitan area (Figure 1). Catchments are native dry sclerophyllous forest dominated by jarrah (Eucalyptus marginata) and marri (Corymbia callophylla) with an understory of small trees, woody shrubs, palms and Melaleuca sp. that are common along watercourses. The climate is Mediterranean and 80% of rainfall occurs during the winter months (May–October) while summer months are hot and dry (December – March). Annual rainfall is greatest on the Darling Plateau (>1100mm) and decreases to the east and north (900–1100mm). Basement rock is predominantly Archean granite that has been strongly weathered resulting in deep (>30m) regolith.

Figure 1.

Darling Plateau water supply dams and individual catchments. Sites that correspond to map numbers are shown in Table 2.

[5] We examined inflows for nine metropolitan reservoirs (combined catchment area ∼3800 km2) with data provided by the Water Corporation of Western Australia (www.watercorporation.com.au/), and long-term rainfall trends provided by the Australian Bureau of Meteorology (www.bom.gov.au/). Reservoir inflows and rainfall since 1950 were examined where continuous data were available. A shorter 20 year period (1989–2008) was used to establish more recent rainfall and runoff trends, and includes recently constructed dams as well as numerous catchments which commenced gauging in the 1980s. To quantify trends across various spatial scales, all stream gauging stations with continuous flow records (n = 18; with catchment areas from 0.9 to 147 km2) were compared with continuous rainfall stations which in most cases were located close to catchment weirs (Figure 1). Catchment rain and flow observations were monitored by the Western Australian Department of Water for the 1989–2008 period. Less than 1% of the streamflow and rainfall records were missing and linear interpolation was used to in-fill gaps in flow data while missing rainfall data were in-filled with interpolated data from SILO data drill to the nearest twentieth of a degree location (http://www.longpaddock.qld.gov.au/silo/). We focused on annual rainfall and runoff in this study, but early winter (May, June July) and full winter rainfall (May through October) were also evaluated and results are supplied in the auxiliary material.

[6] Statistical trends in annual rainfall and streamflow were examined using the non-parametric Mann-Kendall trend test [Kendall, 1970], with p values calculated using the algorithm of [Best and Gipps, 1974]. Calculations were performed in R software using the “Kendall” package (available at http://CRAN.R-project.org/package=Kendall). Change points in time series were assessed using the non-parametric change point method of [Pettitt, 1979], and a significant change year indicates a statistical shift between the data prior to the specified year and this year onwards. Significant auto-correlation was found in annual streamflow time series, but not in annual rainfall time series. Accordingly, streamflow time series were pre-whitened using the “trend-free” procedure of [Yue et al., 2002] prior to trend and change point analysis. The magnitude of the trend was calculated as the slope of the linear correlation between time and annual rainfall or pre-whitened streamflow.

3. Results

[7] All long-term reservoir inflow records showed highly significant negative trends with decline ranging from 0.4 to 1.6 mm yr−1 (Table 1). The trend in annual rainfall over the same period was also significantly negative at four out of five sites with a slope between −3.4 and −8.9 mm yr−1. There was good agreement between rainfall and inflow to dams with change points for annual rainfall and reservoir inflow occurring in either 1969 or 1975, except for the more southerly Stirling dam which showed no rainfall change point (Table 1). For example, a shift in rainfall and dam inflow was observed in 1975 for the Serpentine dam, with a 16% drop in rainfall (1301 to 1095 mm) and a 59% drop in flow (122 to 50 mm) between the 1961-1974 and 1975-2008 periods (Figure 2a).

Figure 2.

(a) Time series in annual rainfall (top) and Serpentine dam inflow (bottom). Solid lines and dotted lines represent average flow or rainfall for 1961–2008 and 1989–2008 periods, respectively. (b) Time series in annual rainfall (mean = solid line, min and max = gray shading) for all catchments (top) and the annual runoff for individual catchments (bottom).

Table 1. Annual Rainfall and Inflow Statistics Since the 1950s for Major Perth Water Supply Reservoirs
SiteChange Point YearChange Point PLinear Slope (mm yr−1)Mann Kendall Slope PPre-Change Point Flow (mm)Post-Change Point Flow (mm)Changea (%)
  • a

    The % change in average flow for the pre- and post-change point periods.

  • b

    Period of observation is 1961–2008 for Serpentine dam and 1950–2008 for all other dams.

Rainfall
Churchman (Armadale)1969<0.0001−8.9<0.000114021021−27
Helena1975<0.001−3.4<0.00011085876−19
Canning1975<0.01−4.9<0.0011188959−19
Serpentineb1975<0.01−4.2<0.00113011095−16
Stirling (Wokalup)NSNS−3.8<0.05---
Dam Inflow
Churchman1969<0.0001−1.6<0.0001310175−44
Helena1975<0.0001−0.4<0.00013211−65
Canning1975<0.0001−0.6<0.018432−63
Serpentineb1975<0.0001−0.8<0.000112250−59
Stirling1975<0.001−1.0<0.0001241179−26

[8] During 1989–2008, we observed no significant trend in annual rainfall at the 19 catchment stations or the six dam stations examined. However, the majority of reservoir inflow (7 of 9) and streamflow (13 of 18) records showed significant negative trends (Table 2). The rate of streamflow decline was greater in the last twenty year period than over the long-term record and ranged from −1.6 to −20.0 mm yr−1. Further, streamflow decline was observed across orders of magnitude in catchment area (from <1 km2 to 1470 km2), but there was no significant relationship between catchment area and runoff decline.

Table 2. Inflow and Streamflow Trend Analysis From 1989 to 2008 for Perth Water Supply and Darling Plateau Catchmentsa
 Area (km2)MapChange Point YearChange Point PLinear Slope (mm yr−1)Mann Kendall Slope PPre-Change Point Flow (mm)Post-Change Point Flow (mm)Changeb (%)No-Flow Daysc
1989–20002001–2008
  • a

    No significant rainfall trends were observed for the 1989–2008 period. See auxiliary material for rainfall and gauging stations.

  • b

    The % change in average flow for the pre- and post-change point periods.

  • c

    The average number of no flow days per year for the 1989–2000 and 2001–2008 periods.

Catchment
   Jane73C1--NS----135168
   Waterfall Gully9C22001<0.01−8.0<0.01277164−4100
   Seldom Seen7C31998<0.0001−9.8<0.001239108−5505
   Vardi Road81C42001<0.05−6.0<0.0515472−54025
   More Seldom Seen3.4C52001<0.01−10.8<0.00117754−705157
   Bates2.2C61998<0.01−11.4<0.0001243106−5600
   Lewis2.0C7--NS----05
   Warren0.9C81998<0.01−13.6<0.000124653−78014
   Cameron West2.1C9--NS----240328
   Bennetts0.9C101998<0.01−20.0<0.000132773−7864226
   Gordon2.1C11--NS----270297
   Del Park1.3C121998<0.01−10.5<0.000121884−6133155
   Waterous14C132001<0.05−10.7<0.05326205−3700
   Hillview Farm17C141997<0.01−11.5<0.05293147−50034
   Sandalwood116C15--−6.1<0.05---02
   Worsley32C161997<0.01−9.0<0.0001240100−5838112
   Coolingutup4C17--NS----211230
   Dingo Rd147C18--−3.0<0.05---012
Water supply dams
   Churchman16D12001<0.01−8.6<0.001194113−42--
   Victoria37D21998<0.01−1.6<0.0019948−52--
   Helena1470D3--NS------
   Canning789D4--NS------
   Serpentine664D52001<0.05−2.4<0.055832−45--
   Wungong130D62001<0.05−7.0<0.0115386−44--
   South Dandalup311D71997<0.05−2.7<0.057943−46--
   North Dandalup153D82001<0.01−5.8<0.0114079−44--
   Stirling251D92001<0.05−9.3<0.01215131−39--

[9] Despite the lack of rainfall trends over the 1989–2008 period, we observed significant streamflow trends (Figure 2b). Streamflow change points found in 1997 or 1998 occurred after below average rainfall in 1997, and the change point in 2001 is associated with one of the driest years on record. For those catchments with significant change points in the 1989–2008 period, the change in runoff coefficient was dramatic. Post-change point runoff was at least one-third less than the pre-change point period, and the majority of sites showed more than a 50% decline in runoff (Table 2). For example, the Serpentine reservoir inflow dropped nearly one-half (58 to 32 mm) from 1989–2000 to 2001–2008 (Figure 2a). Since 2001 a smaller proportion of rainfall was generated as runoff across catchments that vary widely in drainage area (from 2.2 to 664 km2), resulting in distinct and significant relationships between annual rainfall and runoff between 1989–2000 and 2001–2008 (Figure 3a). Total annual rainfall explained greater variation in runoff than rainfall in the early winter (May, June, July) and full winter (May through October) periods (see Figures S1 and S2).

Figure 3.

(a) Relationship between annual rainfall and runoff for the Bates, Waterfall Gully and Vardi Rd catchments and the Serpentine dam. Regression lines represent the 1989–2000 and 2001–2008 periods. (b) Flow duration curves for Bates, Waterfall Gully, Vardi Rd and Del Park catchments calculated in 4 year bins for each catchment.

[10] In addition to changes in total annual runoff, we also observed a change in flow characteristics in the catchments. The number of “no flow” days increased for 15 of 18 catchments in the last decade (Table 2). Only three catchments had perennial flow, and seven catchments switched from perennial to ephemeral flow after 2001. Flow duration curves calculated in 4 year bins highlight the major changes in seasonal flow distribution (Figure 3b). Daily flows have decreased in perennial streams (Figures 3b; Bates and Waterfall Gully), days without flow have been observed since 2004 (Figures 3b; Vardi Road), and the number of dry days has increased markedly in ephemeral streams (Figures 3b; Del Park).

4. Discussion

[11] Our results establish a connection between a downward shift in rainfall and resulting reservoir inflows in the 1960s and mid-1970s for Perth's major water supply catchments. Previous studies have referred to the decline in reservoir inflow as evidence of runoff regime change e.g., [Bates et al., 2008] but we provide the first statistically significant connection between individual rainfall and reservoir inflow records. Our analysis quantifies the coincidence of change in rainfall and reservoir inflow in 1969 (Churchman dam) and in 1975 (Helena, Canning and Serpentine dams), and confirms previous studies that report significant rainfall break points in 1969 and 1975 for SWWA [Hope et al., 2006]. In addition to the 1970s climate and runoff shift, we found a more recent streamflow decline in the last decade that was subsequent to individual dry years rather than an overall rainfall trend. The lack of rainfall trend may be partly due to the short (20 year) period of observation, but our findings demonstrate that catchments are producing less runoff despite receiving average rainfall in the last decade.

[12] Although quantification of the specific variables (e.g. catchment storage and actual ET) that influence catchment runoff are beyond the scope of the present study, the persistence of runoff decline across catchments that vary widely in drainage area suggests that the mechanisms driving runoff change are regional rather than catchment-specific. In particular, the change in runoff response to rainfall in the last decade may reflect the non-linear nature of infiltration and soil moisture dynamics that drives runoff [Jakeman and Hornberger, 1993]. Highly weathered soil profiles of the Darling Plateau provide large soil water stores that can be accessed by deeply rooted eucalypt vegetation (>35 m) [Silberstein et al., 2001], and actual ET in excess of precipitation during a low rainfall year would create a deficit in soil moisture storage that is carried into the following year. This assertion is supported by consistently declining groundwater levels (up to 30cm yr−1) since 1975 [Croton and Reed, 2007], indicating that rainfall infiltration is no longer contributing to groundwater recharge in some areas.

[13] Numerous experimental studies in the Darling Plateau catchments have examined how a reduction in native vegetation influences groundwater levels and runoff [Bari et al., 1996; Ruprecht and Schofield, 1989; Stoneman, 1993]. After thinning or clearing, a rise in water table and an increase in the near-stream saturated area has been observed in experimental catchments, resulting in a positive shift in the annual relationship between rainfall and runoff. In our study, a decline in soil moisture and groundwater levels in the last decade may be driving the decline in the proportion of rainfall that becomes runoff. Furthermore, groundwater-surface water connectivity that is crucial in maintaining baseflow has likely uncoupled in several catchments that now cease to flow in the summer months. We found that perennial streams are now less common in drinking water catchments, and further stream drying may modify the assemblage of stream biota [Boulton, 2003] and influence the succession of riparian vegetation [Naiman and Decamps, 1997].

[14] Recent runoff reductions for SWWA reported in this study are of a similar magnitude to flow decline in the South-eastern Australia (SEA) catchments of the Murray Darling Basin [Potter et al., 2010], and it was recently shown that declining winter rainfall patterns for both regions are related and associated with increasing mean sea level pressure [Hope et al., 2009]. [Cai and Cowan, 2008] have argued that runoff decline in SEA is related not only to declining rainfall but also to increasing temperature that presumably increases evaporation. However, [Donohue et al, 2010] showed that, despite increasing air temperatures, changes in vapour pressure, net radiation and wind speed result in declining PE for much of Australia. Alternatively, [Potter et al., 2010] suggested that interannual rainfall variability and reduced autumn/winter rainfall may be driving disproportionate runoff decline in SEA. While autumn/winter rainfall did not explain more variability in annual runoff in this study (see supplementary material), our results suggest that low rainfall years create a distinct hydrologic shift that will not be reversed without above-average rainfall in the future.

5. Conclusions

[15] Rainfall variability superimposed on falling water tables may be an important cause of streamflow decline in SWWA observed as a threshold response in a changing climate. Shifts in flow distribution for native forest streams are likely related to falling groundwater levels and loss of groundwater-surface water connectivity, contributing to lower annual runoff. Current declines in catchment runoff and reservoir inflows have important implications for future water supply as well as the ecological function of aquatic ecosystems.

Acknowledgments

[16] We thank the Western Australia Department of Water for providing rainfall and streamflow data, the Western Australia Water Corporation for providing dam inflow data, and Geoff Hodgson (CSIRO) for creating the catchment map. This research was funded by a Western Australia Water Foundation grant and the CSIRO Water for a Healthy Country Flagship.

Ancillary