1 2 Rainfall and streamflows in Greater Melbourne catchment area: variability and recent 3 anomalies 4 5 Authors: B. Timbal1, M. Griffiths1 and K.S. Tan2 6 7 8 9 10 Version 12: Final review in response to Climate Research reviewer's comments 11 12 13 14 15 Corresponding author: 16 Bertrand Timbal 17 [email protected] 18 19 1 20 Bureau of Meteorology 21 GPO Box 1289, Melbourne VIC3001, Australia 22 2 23 24 990, La Trobe Street, Docklands, VIC 3008, Australia : Centre for Australian Weather and Climate Research : Melbourne Water 1 25 Abstract 26 27 Observed rainfall and water availability is investigated across several catchments northeast of 28 Melbourne using gridded rainfall data over the last 113 years and reconstructed streamflow 29 observations for the last 100 years, focusing on the 1997-2009 record-breaking rainfall deficits, 30 associated record low streamflows and subsequent recovery from 2010 to 2012. The influence of 31 large-scale tropical modes of climate variability affecting rainfall, and subsequently reservoir 32 streamflows, are shown to be modulated by the orographic features marking this region. Since these 33 remote large-scale tropical climate forcings contributed strongly to the recovery, it helps explain the 34 uneven recovery since 2010. However, across these catchments these large-scale modes of natural 35 variability do not explain the long-term deficit in streamflows in the last 15 years. Annual streamflow 36 in these wet catchments can skilfully be reconstructed month by month using catchment-wide 37 observed rainfall. The year-to-year variability, decline during the last 30 years and magnitude of the 38 deficiency during the Millennium Drought are reasonably well captured but not fully accounted for by 39 the linear combination of the rainfall of the current month, the previous month and the previous 12 40 months. Maximum temperature does not have a sizeable additional impact when added to the 41 reconstruction, while the previous 12 months of rainfall is as critical information to capture many of 42 these statistics. 43 2 44 45 1. Introduction 46 South-eastern Australia (SEA), defined as continental Australia south of 33.5°S and east of 135.5°E 47 (Timbal, 2009), had its lowest 13-year rainfall between 1997 and 2009 compared to the entire 1900- 48 2012 period. The 1997-2009 annual average of 512.4 mm was 12% below the 1900-2012 average 49 annual of 582 mm (CSIRO, 2012). We will use the terminology Millennium Drought (MD) as used 50 by Leblanc et al. (2011) to name that extremely low rainfall period occurring between 1997 and 2009 51 at the time of the transition to the new millennium. Two other low rainfall periods were seen in the 52 instrumental record, which extends as far back as 1865 (Timbal and Fawcett, 2013), one around the 53 time of the federation of the Australian States (in 1901, Federation Drought) and the other during an 54 extended period preceding and encompassing World War II (WWII Drought). The MD was 55 unprecedented as it was more severe than either of these two previous droughts and, in contrast to the 56 previous droughts the rainfall deficit in SEA happened during the cooler part of the annual cycle from 57 April to October with the largest anomalies in autumn (Timbal, 2009). 58 59 Timbal and Hendon (2011) have shown that the MD could not be accounted for by naturally occurring 60 modes of variability generated within the tropics such as the El Niño-Southern Oscillation (ENSO) 61 and the Indian Ocean Dipole (IOD). This is consistent with other calculations (Smith and Timbal, 62 2012) and the fact that tropical modes of variability do not relate well with autumn rainfall in SEA 63 (Timbal and Murphy, 2007). Timbal et al. (2010) demonstrated that the rainfall deficit observed across 64 SEA was related to an observed strengthening of the belt of high pressure – a feature evident at every 65 longitude but for this study measured above the eastern part of the Australian continent following 66 Drosdowsky (2005). This belt of high pressure, known as the sub-tropical ridge (STR), is a key 67 controller of rainfall across southern Australia during the cool months of the year from April to 68 October. The 20th and early 21st century evolution of the ridge has been observed to evolve similarly to 69 global average temperatures and to account for about two-thirds of the observed deficit across SEA 70 during the MD (Timbal and Drosdowsky, 2013). 3 71 72 The MD concluded in 2009; the spring/summer of 2010/11 saw one of the strongest La Niña events on 73 record combined with a negative IOD event and a positive Southern Annular Mode (SAM) event – i.e. 74 all three key influences were in their wet phases in terms of expected rainfall impacts on south-eastern 75 Australia (CSIRO, 2012). These conditions, coupled with the warmest sea-surface temperatures on 76 record to the north of the Australian continent, contributed to making 2010 to 2011 Australia’s wettest 77 two-year period on record (Bureau of Meteorology, 2012). While the extensive flooding of 2010/11 78 was, by and large, consistent with natural variability (Hendon et al., 2013), it is possible that ongoing 79 global warming contributed to the magnitude of the event through its impact on ocean temperatures 80 (Evans and Boyer-Souchet, 2012). The spring and summer of 2011/12 saw the re-emergence of 81 another (weaker) La Niña event, combined with a positive SAM event during early summer, but this 82 time the Indian Ocean played a lesser role. Despite the La Niña events of 2010/11 and 2011/12, which 83 resulted in above average annual rainfall totals across the region, late autumn and winter rainfalls in 84 2011 continued to be below average, which is consistent with the trends associated with the observed 85 evolution of the STR associated with the expansion of the Hadley circulation (Nguyen et al., 2013); in 86 2010 to 2012, the intensity of the STR averaged 0.3, 1.1 and 1.2 hPa above the 20th century average 87 respectively in autumn, winter and spring. 88 89 The eastern area of Australia is dominated by the Great Dividing Range (GDR) – a range of 90 mountains, up to 2,000 m high, running approximately parallel to the coast from western Victoria in 91 the south to mid-Queensland in the north, 200-300km inland (Figure 1a). The southern part of the 92 GDR, within SEA, is where a large fraction of the run-offs and streamflows into dams is generated. In 93 this study we will focus on the central southern Victoria region (Figure 1b) where Melbourne and a 94 large part of the state of Victoria collect surface run-off into large dams North East of Melbourne. In 95 the area of interest the GDR runs basically west-east; on the northern side of the GDR, Lake Eildon 96 receives surface run-off from a large catchment (Fig. 1b); while on the south side lies the area 97 managed by Melbourne Water. There is a significant north-south orientated range connecting to the 4 98 GDR in a location named “the Triangle” separating the Thomson catchment area in the east from the 99 other major Melbourne Water catchments in the Yarra Basin to the west: Upper Yarra, O’Shannassy 100 and Maroondah (comprising streamflow from Watts River and diversion from Graceburn Creek). 101 These major water supply catchments all feed into the Yarra River, we will referred to them as “Yarra 102 catchments”, note we do not include some minor Melbourne water supply catchments and tributaries 103 within the Yarra Basin (e.g. Armstrong, Starvation and McMahon creeks). The various orientations of 104 these catchments in relation to the GDR may generate significant differences in term of how large- 105 scale climate modes of variability are impacting the rainfall across these catchments and how future 106 climate change may impact them. While the entire south-eastern Australia has been extensively 107 studied (CSIRO, 2012); a specific study on these important catchments is lacking. It is therefore of 108 interest to put the rainfall variability and in particular the magnitude of the MD across these major 109 water supply catchments in perspective and compare them to the entire SEA region and the broader 110 Murray-Darling Basin (MDB) which extend further North. Beside the importance of this region for 111 water management and the provision of drinking water to Melbourne, all these catchments are inter- 112 connected (water transfers are enabled from east to west and from north to south), it offers interesting 113 water supply operations and management opportunities and challenges for which improved 114 understanding of variability and changes are valuable. 115 116 While average annual rainfall during the MD declined by an average of 12% across south-eastern 117 Australia, the impact on streamflows was even more extreme. Due to non-linear hydrologic processes, 118 a magnified response of streamflows relative to rainfall (the “elasticity factor”) is expected (Chiew, 119 2006) and are generally two to three times the magnitude of the respective decline seen in rainfall for 120 the MDB region (Post & Moran, 2013). However, non-stationarity was also observed during the MD 121 as the elasticity response was much larger than during previous dry periods (Potter et al., 2010). This 122 magnification of the streamflow response has been related to various factors including the absence of 123 wet years, increasing temperature and the change in seasonal rainfall patterns, which means that the 124 land surface is not well moistened in autumn, resulting in a substantial reduction in runoff generation 5 125 during subsequent seasons possibly leading to poorly understood hydrological processes (e.g. changes 126 in ground water levels and connectivity with surface runoff). Some hydrological studies have focused 127 on the Victorian and Melbourne Water catchments (Preston and Jones, 2008; Howe et al. 2006) and 128 many studies have focused on the hydrological implications of the MD (Kiem and Verdon-Kidd, 2009 129 and 2010; CSIRO 2010, 2012) but there is a lack of study focusing on the impact of the MD and 130 subsequent recovery for the catchments close to Melbourne. Furthermore, in light of the inability of 131 hydrological model to fully reproduce the impact of the MD on streamflow across the region and in 132 particular the increase in the elasticity factor (CSIRO, 2012), we decide to investigate an alternative 133 approach and evaluate how well simple statistical relationships between rainfall and streamflow could 134 capture the hydrological response to the MD in these important catchment east of Melbourne. 135 136 The aims of this research are as follow: 1) document the characteristics of the MD and subsequent 137 recovery from 2010 to 2012 across the catchments east of Melbourne with a focus on the importance 138 of the marked orography across these various catchments which have different orientation: north of the 139 GDR (Lake Eildon catchment), and south of the GDR (Melbourne Water major catchments) with a 140 separation between the south-west oriented catchment (the Yarra Basin catchments), and the south- 141 east oriented catchment (the Thomson catchment), 2) for the same area evaluate how the large-scale 142 drivers of the rainfall are being modulated by the orography, and 3) evaluate how simple relationships 143 between streamflows and rainfall can capture the specificities of these catchments and the marked 144 variability across the observation record and in particular in the recent decades. 145 146 This paper is structured as follows: first a description of the dataset used and their possible short 147 comings is provided, with additional details on the area being studied. Then a section presents the 148 results focusing on 3 directions: the rainfall and streamflow records over the region, investigating the 149 seasonal and year-to-year variability with a particular focus on the MD and the 2010 to 2011 recovery; 150 the relationship between observed local variability for rainfall and streamflows and large-scale climate 151 indices; and finally, the possibility to use observed rainfall and temperature data to reconstruct the 6 152 observed streamflows. A discussion section follows to synthesis the findings and draw some 153 conclusions from this investigation. 154 155 156 2. Data and Methods 157 158 We use the Australian Bureau of Meteorology’s current operational high-resolution monthly rainfall 159 analyses, generated as part of the Australian Water Availability Project (AWAP) (Jones et al. 2009). 160 These analyses are at 0.05°×0.05° resolution, or approximately 5 km × 5 km. The analysis 161 methodology employed in these analyses is a hybrid one, merging two-dimensional Barnes successive 162 correction analyses (Jones and Weymouth 1997) of fractions of monthly mean rainfall and three- 163 dimensional thin-plate smoothing spline analyses (Hutchinson 1995) of climatological monthly 164 rainfall. 165 166 The number of stations varies considerably during the period. For the purpose of understanding the 167 changes in real stations feeding the gridded rainfall in the vicinity of the area of interest, records from 168 stations have been combined, such that stations within 0.05 degree (approx. 5km) of each other are 169 grouped. Station record duration and location varies with time (Figure 2). A station is considered 170 active if there is data for at least 10 months of the year and for at least 75% of the years during that 171 time period. Focusing on three regions, covering Eildon, Thomson and the Yarra catchments; the time 172 evolution of the number of stations reporting for each of these three stations is likely to influence the 173 gridded rainfall as the numbers of stations has grown steadily until the 1970s before levelling off or 174 decreasing (Figure 2, lower left panel). Of particular concern is the coverage in the more mountainous 175 parts of the catchment. The average altitude of stations within each of these regions shows that it has 176 been increasing with time (Figure 2, lower right panel). This non-stationarity is likely to have an 177 effect on the time series of the AWAP catchment average as higher elevation stations tend to be wetter 7 178 than low elevation stations. It is therefore expected that the rainfall in the early record is likely to be 179 lower due to the station coverage. One would expect this effect to generate a spurious upward trend for 180 rainfall across these catchments (the relevance of this data issue will be discussed further in section 4). 181 182 The streamflow data at the Melbourne Water major water supply catchments (Yarra and Thomson 183 catchments) were provided by Melbourne Water, and the data for the Eildon catchment were provided 184 by the Department Environment and Primary Industries (DEPI). All streamflow data span the period 185 1913 to 2012. Monthly streamflow record at the Melbourne Water major harvesting reservoir sites are 186 based on reservoir mass balance approach calculated using hydrological models and water level 187 gauging records (taking into account of the streamflow, outflow and change in storage). However, the 188 earlier part of the monthly streamflow record prior to the construction and operation of the reservoirs 189 (pre-1984 for Thomson, pre-1957 for Upper Yarra, and around pre-1926 for O’Shannassy and 190 Maroondah) were derived based on flow correlation with nearby stream gauging record or catchment 191 area adjustment. For Graceburn Creek, the entire recorded monthly flow just upstream of the weir 192 (unimpacted flow) was used in this study as some of the flow was diverted at the weir into the 193 Maroondah Reservoir since 1941 depending on operational conditions. The monthly flow record at the 194 Melbourne Water major harvesting reservoir sites represents the main sources of streamflow from 195 Melbourne's water supply catchments that are not impacted by upstream diversions, but may be 196 impacted by changes to catchment (see further elements in the discussion section). 197 198 The long-term annual mean gridded rainfall as well as the mean for the four calendar seasons from 199 1900-2012 is shown in its native grid (approx. 5 km squares) across the area of interest (Figure 3). As 200 expected, rainfall is higher over the significant topography, and the highest rainfall is observed in 201 winter. This is consistent with the main characteristics of the mid-latitude rainfall systems (frontal and 202 cut-off lows) travelling from west to east across the region. In particular, in winter the maximum 203 rainfall is on the west facing slopes at high elevation – O’Shannassy and the southern part of the 8 204 Eildon catchment area. Spring is wetter than autumn. This picture is very relevant for the Murray part 205 of the MDB as most of the run-off is generated in these high elevations areas. 206 207 Factors known to influence the run-off are the rainfall, the soil moisture and the evaporation. In this 208 study we use the rainfall of the current month as the first predictors of the monthly streamflow data 209 continue with additional contribution to the pre-conditioning of the soil by using the rainfall of the 210 immediate past month as well as the rainfall over the previous 12 months. In addition, we also 211 consider the role that maximum temperature can play on streamflow via the evaporation by using the 212 average maximum temperature over the current month (based on the same operational gridded Bureau 213 of Meteorology dataset). While the role of temperature on evaporation is well established by the 214 Penman-Monteith equation, its impact on the complex evapo-transpiration aggregated over a 215 vegetation covered landscape is debateable (see Smith and Power, 2014). Some early studies 216 suggested that warmer temperatures made an important contribution to the anomalously dry conditions 217 experienced in the Murray Darling Basin region of eastern Australia during the prolonged drought of 218 1996 to 2010 (Nicholls, 2004; Cai et al., 2009). However, this explanation has been disputed by 219 Lockart et al. (2009). There is no evidence that temperature increases can explain this changed 220 relationship. A more likely explanation, as proposed by others, is that it reflects the combined effects 221 of changes in timing of rainfall events throughout the year, the absence of very heavy rainfall events 222 and, most likely, long-term changes in groundwater levels (Larsen and Nicholls, 2012). Studies 223 combining all these effects with temperature anomalies suggested a much more reduced contribution 224 from temperature than originally thought (CSIRO, 2012 and reference within) 225 226 Temperature data are available from 1910 onward, rainfall data from 1900 onward and streamflows 227 data from 1913 onward. A multi-linear regression approach was used based on the observed 228 relationship between the atmospheric variables and streamflows for the full 1913 to 2012 record. 229 Several attempt were made to linearly reconstruct the monthly streamflow data using these predictors, 9 230 five are reported here in order to document the importance of several factors as well as the possibility 231 of spurious skill due to over fitting of the dataset: 232 (1) R 1 is the most complete reconstruction and combine current monthly rainfall with the rainfall of 233 the previous month, the rainfall of the previous 12 months (not including the current month) as 234 well as the temperature of the month, 235 (2) R 2 repeats R 1 but using a fully cross-validated approach, 236 (3) R 3 repeats R 1 but with the influence of the temperature of the month removed, 237 (4) R 4 repeats R 3 but with the influence of the rainfall of the previous 12 months removed; and 238 (5) R 5 repeats R 4 but with the influence of the rainfall of the previous month removed. 239 While these are very clear simplifications of complex hydrological processes occurring at small sub- 240 catchment scale, the focus here is to see how much of the catchment-wide streamflow can be captured 241 with this simple approach as an alternative to hydrological model of various complexity (e.g. storage 242 routing) which are used in many studies. 243 244 245 246 3. Results a. Rainfall and streamflow variability in the recent decades 247 In order to characterise the specificities of the MD across the catchments that were considered in this 248 study and to put it into the broader perspective of the SEA and MDB, we compare the MD to the 249 previous driest periods on record for each area. It was found that the earlier driest periods on record 250 were centred around 1940 for the MDB and SEA, and around 1910 for the Melbourne Water 251 catchments and for Eildon. Of particular interest are the rainfall deficit as a percentage of the long- 252 term mean, the magnitude of the wettest years during both dry periods and the seasonality of the 253 decline (Table 1). 254 10 255 In the Eildon catchment, the previous largest rainfall deficit was 8% (over the 13-year period 1902- 256 1914). This is a remarkably small number compare with the deficit of 18% observed during the MD. 257 For Melbourne Water catchments, previous driest 13-year periods were even less remarkable, never 258 exceeding 5% deficits. Another clear difference between the MD and previous historical droughts is 259 the absence of very wet years: e.g. in both Eildon and Yarra catchments, there was no single year 260 above the long-term average for the 13 years of the MD. In comparison, during previous long-term 261 drought, well above average rainfall years were observed (up to 15% above the long-term average). 262 The seasonality of the rainfall deficit is also markedly different. During the MD, the autumn deficit is 263 the largest in percentage terms, almost 30% whilst the other seasons have 7 to 16% decreases. For the 264 1902-1914 drought, the decline was predominantly in summer (about 10%), the other seasons all had 265 declines of less than 5%; winter being the key season for streamflow was close (within ±2%) to the 266 long-term average for Eildon and all Melbourne Water catchment areas. 267 268 The annual cycle of rainfall in each of the four catchment areas highlights the differences between the 269 broader MDB which has a broad summer maximum for rainfall and a smaller early winter peak in 270 June and the catchments across the GDR where rainfall is predominantly a cool-season (April to 271 October, peaking in July) phenomenon (Figure 4). The bi-modal nature of the annual cycle reflects the 272 large latitudinal extent of the MDB (from about 38oS to 24oS) with a northern basin dominated by 273 warm season rainfall and the southerly Murray Basin dominated by cool season rainfall. During the 274 MD, the dry conditions in the southern catchments were apparent over the whole year but worse for 275 the entire half of the year when the monthly temperature is below the annual mean (i.e. the cool 276 season), whilst the dry conditions in the MDB were limited to the cool season with a small increase 277 during the warm season. 278 streamflows in the three southerly catchments is very large. The largest reduction in streamflow in 279 absolute terms is in winter and spring which is the peak season for streamflow while the largest 280 rainfall decline occur in autumn. In percentage terms nevertheless, the annual cycle of the streamflow The consequence of the rainfall deficiency during the MD on the 11 281 reductions match the rainfall deficiency being largest in autumn, this illustrates the importance of 282 autumn rainfall to saturate the catchment prior to the winter rains 283 284 2010 when the MD ended together with 2011 were the wettest two years recorded across Australia 285 (BoM, 2012). For the two catchment areas north of the GDR, Murray Darling Basin and Eildon, 2010 286 was extremely wet ranking 1st and 7th respectively out of the 110-year record (Table 2). Further south 287 (Yarra and Thomson) the 2010 rainfall was above the long-term mean but only ranked 29 and 26 288 respectively for the 113-year record (decile 8), well short of the highest values on record established 289 during the 1950s or the 1970s depending on catchments. That north-south difference in ranking was 290 less apparent in 2011 with rainfall ranking between 11 and 36 for all catchments considered. These 291 annual totals hide very contrasting seasonal behaviours. Dividing the rainfall into the warm months of 292 the year (November to March), and the cold months of the year (April to October), the record-breaking 293 rain came during the warm months of the year whereas the cool month rainfall was close to or below 294 the long-term average in 2010 and 2011 across the Yarra and Thomson catchments. While in Eildon, 295 the cool month rainfall was above average in 2010 but below in 2011 similar to SEA and the entire 296 MDB. The rainfall patterns for the average of 2010 and 2011 when compared to the long-term mean 297 (Figure 5) underlines that across the catchments of interest the largest differences occurred in summer 298 and spring while the winter and autumn maps are similar to the long-term mean. During the 2-year 299 wet period, total rainfall during the warm season across the area was mostly higher than the total 300 rainfall during the cool season, and this contrast is significant in the entire MDB and the Eildon north 301 of the GDR in comparison with the Yarra and Thomson catchments south of the GDR. 302 303 The size of the recovery in comparison of the size of the MD deficit differ markedly between the 304 entire MDB and the Eildon north of the GDR and the Yarra and Thomson catchments south of the 305 GDR. These quantities are expected to oscillate around zero but for the 16 years (192 months) shown 306 (Figure 6) sizeable deficits are obvious. Across the southerly catchments, the deficit for the MD (1997 307 to 2009) is only partially offset by the subsequent recovery in 2010 to 2011 and at the end of the 1612 308 year period; the accumulated deficit is still larger than at any time in the past instrumental record prior 309 to the onset of the MD in 1997 (the previous lowest 16-year period drawn from the 1900 to 1996 310 record is considered in Figure 6 to provide that historical benchmark). The magnitude of the record- 311 breaking nature of the MD for the 3 southern catchments is clearly shown. In contrast, the recovery 312 from the long-term perspective is relatively modest in all three catchments except for the Eildon 313 catchment north of the divide where the recovery is sizeable. 314 315 316 317 b. Large-scale modes of natural variability on rainfall and reservoir streamflows 318 Australia is affected by many types of weather systems, the frequency and magnitude of these are 319 often influenced by some large-scale features of the atmospheric circulation (Risbey et al, 2009). Best 320 known and documented such natural large-scale modes emanate from tropical oceans: primarily in the 321 Pacific (the El Niño-Southern Oscillation: ENSO) and its response in the Indian Ocean, as measured 322 by the Indian Ocean Dipole (IOD) index proposed by Saji et al. (1999). To investigate these large- 323 scale influences on the high elevation catchments we are using two indices based on tropical sea 324 surface temperatures. Niño4 (N4) is the classic index of the ENSO variability based on central Pacific 325 SST anomalies, the tripole is an index developed by Timbal and Hendon (2011) which encompasses 326 the anomalies of all tropical SSTs with a strong relationship with the SEA rainfall and hence capture 327 the additional effect of the IOD on Australian rainfall. 328 329 In addition, regional indicators are used which in part respond to remote modes of variability but for 330 other part respond to broad scale change of the mean meridional circulation which influence the 331 climate of SEA through the role of mean sea level pressure (MSLP) in controlling local rainfall and 332 streamflow (CSIRO, 2012). 333 Drosdowsky (2005) has been shown to be tight controller of SEA rainfall during the cool season 334 (Timbal and Drosdowsky, 2013) and is used in this study. The geographical location of these indices 335 are summarised in Figure 7. The sub-tropical ridge (STR) intensity and position computed by 13 336 337 Monthly correlation coefficients between both rainfall and streamflow (not shown) and the four 338 climate indices for each of the four catchments are very weak and not significant from November to 339 March but are statistical significant from April to October in most instances. This is explored further 340 for the key seasons of winter and spring (Figure 8) when the relationships are usually stronger and 341 statistically significant and when the bulk of the streamflow is generated. Higher correlation 342 coefficients with the tripole rather than N4 are noted everywhere and are consistent with the fact that 343 the tripole was constructed to maximise these correlations across SEA. The added value of the tripole 344 versus the classic N4 is obvious in winter where correlation coefficients for N4 fail to reach the 99% 345 significance level across the entire region. 346 347 However, by and large, both the STR position and intensity are more important in controlling year-to- 348 year variability across the Victorian catchments than tropical modes. Contrary to tropical influences, 349 the STR has a stronger relationship across the entire region in winter rather than in spring. South of the 350 GDR, the reduction from winter to spring is marginal and hence by spring the area where the strongest 351 control on the rainfall from the STR shift from being north of the GDR to being south of it. Spatial 352 features are very similar for STR intensity and position; the stronger relationship is always with the 353 intensity rather than with position, consistent with previous study (Timbal and Drosdowsky, 2013; 354 Whan et al., 2013). 355 356 The relationship between streamflow and the state of the tropical oceans was investigated by building 357 composites from both the long-term historical period and during the more recent MD and subsequent 358 recovery in 2010 and 2011. We compute composite based on the tripole annual values which by and 359 large are consistent with known ENSO and IOD years (Figure 9). Few noteworthy exceptions are 360 1928 and 1929 when, a negative tripole was associated with El Niño years, and 1933 when a positive 361 tripole was associated with La Niña years, and 1999 when a high positive tripole year was neither 362 associated with La Niña or IOD years and 1918, 1930, 1960 and 1961 when negative tripole years 14 363 were also not linked to ENSO or IOD variability. Composites were computed based on positive or 364 negative tripole years and compared with averages; the data were split in two periods: 1910 to 1996 365 and from 1997 to 2012 (Figure 10). This compositing demonstrates that tropical modes of variability 366 continue to have the same effect on a year-to-year basis i.e. positive tripole years are wetter than 367 average and negative tripole years are drier than average but it is happening with a reduced mean 368 streamflow post-1996. The shift is particularly marked for Melbourne Water catchments with almost 369 no overlaps between the two envelopes a clear indication of the role of the GDR in limiting the 370 tropical influence across Melbourne catchment and making the long-term drying trend since 1996 371 more prominent. Beside the overall reduction, it is also noted that the peak streamflow month is now 372 delayed from August for the pre-1997 period to September in the post-1996 period. This could 373 possibly be the effect of drier autumn with lower antecedent catchment condition at start of winter 374 months, and coupled with reduced rainfall, leading to reduced runoff generation efficiency and a delay 375 in peak monthly streamflow. 376 377 378 c. Using catchment rainfall to estimate reservoir streamflow 379 In this section we explore the relationship between rainfall and streamflow: using simple linear models 380 described in an earlier section we attempt to reconstruct the streamflow using the high resolution 381 gridded rainfall and temperature data. Key statistics for reconstructed streamflows are presented and 382 compared to observed rainfall and streamflow (Tables 3 and 4). The ability to reproduce the observed 383 streamflow decreases when the number of explanatory hydro-climate variables is reduced. For all 384 statistics, rainfalls at lag are important either from the previous month (R 4 versus R 5 ) and previous 12 385 months (R 3 versus R 4 ); the impact of temperature is very marginal (R 1 versus R 3 ). 386 387 Explained variances (i.e. squared correlation coefficients) are high and significant (Table 3) even for 388 the simplest of approach but grow to exceed 0.9 when multiple explanatory variables were used. This 389 is not reduced by a full cross-validation done for the most complete reconstruction (R 2 versus R 1 ). The 15 390 improvement in the amount of variance reconstructed is noteworthy rising to close to 80% for the 391 multiple variable reconstructions, however there it is worth noting that a full cross-validation reveals 392 that part of the gain may be spurious (compare R 2 and R 1 amount of observed variance reproduced in 393 Table 3). In addition to the year-to-year variability, it is interesting to investigate how well these 394 reconstructions capture trends (both long and short term). We have chosen to look at 3 different time 395 periods: the Millennium Drought (1997-2009), the recent period (1980-2012) and the entire period for 396 which we have data (1913-2012). Results from the fully cross-validated approach (R 2 ) are not 397 meaningful from the full historical period and are not reported. 398 399 There has been a net decrease in the streamflow for each of the time periods and for each catchment 400 (Table 4). Over the last 30 years the most complicated reconstruction captures the trend well (R 1 ). 401 Similarly the magnitude of the decline in streamflow during the MD which has been reported to be 402 larger than what is expected from the well-established rainfall-runoff elasticity (Potter et al., 2010; 403 CSIRO, 2012 and reference within) is well captured (in particular for Eildon and Yarra, less so for the 404 Thomson catchment). For these two results, the cross-validation using data prior to the last 30 years 405 (R 2 ) ensures that these results are not spurious. 406 407 Contrary to the recent trend and intensity of the MD, only half the secular trend is captured at best by 408 the more comprehensive multiple variable model. In light of the model’s ability to reproduce the more 409 pronounced trends observed since 1980, it does raise the issue that long-term trends are partly affected 410 by the non-stationarity in the number of actual observations used in the rainfall grid (see data section 411 and Figure 2 for details) thus preventing the model to match the trend in the longer-term streamflow 412 record. 413 414 415 4. Discussion and conclusions 16 416 417 a. Severity of the Millennium Drought 418 South Eastern Australia experienced, from January 1997 to December 2009, its worst long-term 419 drought on record: the Millennium Drought (MD) the driest 13-year period since reliable climate 420 records from the mid-18th century. It is different from previous droughts in several ways including 421 lower year-year variability and the seasonality of the rainfall decline. The MD was more severe than 422 previous historical drought across the entire SEA, but more remarkably so for the catchments dotted 423 across the Great Dividing Range (GDR), in the State of Victoria, east of Melbourne: by a sizeable 424 factor of 3 to 4 compare to the previous historical drought. The GDR acts as a physical barrier in terms 425 of natural variability and is the source of high run-off and streamflows into large dams which have 426 been constructed to provide surface water supply to Melbourne (Yarra and Thomson catchments) and 427 the rest of Victoria north of the GDR (Eildon catchment). The long-term rainfall deficiencies (and 428 streamflow reductions) have led to challenges in water management in Melbourne and across the SEA 429 (Post et al. 2014). The previous driest period on record was only a third (Thomson catchment) to less 430 than a quarter (Yarra catchments) of the severity of the rainfall deficit during the MD shading new 431 light on the extreme severity of the MD for these high elevation, high yield catchments. 432 433 This finding confirms earlier study about the very unusual nature of the MD. The question about the 434 stationarity of the rainfall record in the recent decades compared to the full historical record was raised 435 by Timbal and Fawcett (2013) whom demonstrated that to be very unlikely to be the case across the 436 broader south-east of Australia using a longer historical perspective (from 1865 to 2012). Furthermore, 437 using paleoclimate proxy records from the Australasian region, Gallant and Gergis (2011) developed a 438 reconstruction of streamflow for the Murray River from 1783 to 1988, and estimated that the 1998– 439 2008 streamflow deficit has an approximate 1 in 1,500 year return period. For Melbourne’s major 440 water supply catchments, Tan and Rhodes (2013) highlighted that the 10-year (1997-2006) drought 441 streamflow was estimated to have less than a 0.2% probability of occurrence (worse than 1 in 500- 442 year) based on historical streamflow record since 1913, but could become a commonly occurring event 17 443 in a changing and variable climate assuming a ‘return to dry scenario’ similar to that experienced 444 during the recent MD. 445 446 b. Recovery following the Millennium Drought 447 The MD concluded in early 2010. In 2010 and 2011, very high rainfall across the entire Australian 448 continent (highest 2 year rainfall total across the continent) brought relief to SEA and most reservoirs 449 located in the catchments considered in this study recovered. In 2012, at the end of the refilling season 450 (November), the MDB authority reported its storage capacity at 100%, Lake Eildon also reached full 451 capacity while Melbourne Water storages exceeded 80%. These are clear indications that from a water 452 management perspective, the abundant rainfall in 2010 and 2011 provided the required relief from the 453 MD and the recovery was complete. 454 455 However, from a purely climatic perspective, cumulative monthly rainfall anomalies tell a different 456 story regarding the magnitude of the recovery in comparison with the severity of the MD rainfall 457 deficiency; the accumulated rainfall deficit has not recovered anywhere close to the previous historical 458 accumulated deficits. For all these high elevation catchments on both sides of the GDR, the recovery 459 of the reservoir levels hides some important facts. Despite the 2010 to 2011 recovery, across the 460 Eildon, Yarra and Thomson catchments, accumulated rainfall deficiencies are still the largest at any 461 time within the historical record prior to the onset of the MD. The rainfall during the recovery was 462 predominantly a warm season phenomena while the long-term deficit is primarily a cool season 463 phenomena. Finally, the recovery was un-even spatially with the largest recovery observed north of 464 the GDR and not as much south of the GDR in Melbourne’s water supply catchment areas. 465 466 c. Role of large-scale drivers 467 It is well established that rainfall in SE Australia decreases during El Niño years and during years with 468 positive IOD, and it has been suggested that the MD was due to an extreme number of these years 469 (Ummenhofer et al., 2009); although that idea has since been discarded by several studies (Nicholls, 18 470 2009, Smith and Timbal, 2012, Timbal and Hendon, 2011). The analysis of inter-annual variability of 471 rainfall and streamflow across these catchments highlights the importance of the GDR. Large-scale 472 modes on natural variability emanating from Tropical Pacific and Indian oceans are an important 473 source of the observed inter-annual variability of rainfall and streamflow but this importance is 474 reduced south of the GDR. The tropical modes show a significant north-south gradient of influence 475 with stronger correlations to the north. Using a tripole index (Timbal and Hendon, 2011), it was 476 shown that the SST to the north of Australia and the Indian Ocean response to ENSO play a key role at 477 the onset of the winter-spring period; by spring, the relationships with the tropical indices are stronger 478 throughout the region than in winter but the strong influence of the GDR in limiting the relationship 479 with tropical influences further south across Victoria remains obvious. There is a similar effect further 480 north along the GDR between coastal and inland SEA, which has been explained in terms of moisture 481 flux anomalies and opposing influences between the direct influence of the Pacific tropical ocean and 482 the remote influence of the Indian Ocean (Pepler et al., 2014). 483 484 On the contrary to teleconnections with tropical oceans, local control of the rainfall due to the belt of 485 high pressure systems hovering across SEA (the sub-tropical ridge) is particularly strong for the entire 486 winter and spring period in these high elevation catchments. Rainfall is primarily generated during the 487 cool time of the year by rain bearing systems travelling from west to east and their ability to penetrate 488 inland and deliver rainfall is being reduced if the STR strengthens or shifts poleward. From a seasonal 489 prediction perspective, the correlations with rainfall are consistently stronger than those with 490 streamflow in particular with the STR indices, consistent with the lack of predictability and hence 491 memory from month to month for the STR behaviour. The difference is not as pronounced for tropical 492 indices where correlations are very similar for rainfall and streamflows in spring thus confirming that 493 because of the predictable and persistent nature, tropical modes of variability can be used to predict 494 streamflows and complement the memory from month to month which exist in streamflow (Robertson 495 and Wang, 2012). 496 19 497 For the catchments east of Melbourne, since 1997 there is a definite shift to lower streamflow during 498 the June to November dam filling season. This is the period of greatest rainfall and is therefore critical 499 to water management. Since 1997, high streamflow years due to a positive influence from the tropics 500 (i.e. positive tripole) provide similar streamflow to low streamflow years due to a negative tropical 501 influence (i.e. negative tripole) prior to 1996. It is clear from that analysis that while tropical modes of 502 variability remain important to discriminate between high streamflow and low streamflow years, the 503 marked reduction of streamflow during the period including the MD and the recovery in 2010-12 is 504 not explained by these modes. Instead, we continue to observed year-to-year variability driven by 505 these modes (as were the case during the wet 2010 and 2011 La Niña episodes or during the very dry 506 2006 to 2009 repeated positive IOD episodes) but applied to a different and lower baseline for mean 507 streamflow. In the meantime, other studies have reported that the reduction of rainfall across SEA 508 which is driving the streamflow reduction in these southerly catchments is primarily due to a 509 strengthening of the STR potentially associated with global warming (Timbal and Drosdowsky, 2013). 510 511 d. Relationships between rainfall, temperature and streamflow 512 The ability to reproduce monthly streamflow variability using solely the rainfall observed across the 513 catchments over different time scales and combined with temperature was investigated. For this part of 514 the study, the focus was on Eildon, Thomson and Yarra catchments. These southern catchments are 515 smaller in scale and have higher rainfall than the Murray Darling Basin and thus it is more likely that 516 the streamflow can be successfully captured by using catchment wide rainfall averages. In addition, 517 because of the complex made-man influence across the Murray Darling Basin it is unlikely that a 518 simple relationship can be found between rainfall, temperature and streamflow. It was found that when 519 current rainfalls, as well as rainfall from the previous month and the previous 12 months are used as 520 explanatory variables in the multiple linear regression models, most of the annual streamflow 521 variability can be reasonably reproduced including the decline in the last 30 years and the severity of 522 the streamflow decline during the MD. On the contrary to rainfall, temperature does not appear to play 523 a significant role, as discussed previously by Potter and Chiew (2011) and Smith and Power (2014) 20 524 and reference within. This was tested several times using various combinations with and without 525 temperature (not all results were shown here for the sake of brevity). However, it is fair to say that in 526 all attempts made rainfall of the month was always used and hence it is possible that due to the 527 dependence between rainfall and temperature (i.e. wet days are cold days), once rainfall is used the 528 effect of temperature on streamflow is already captured. Overall for future use of this simple approach, 529 testing both the optimal combination (which includes temperature) and a reduced combination 530 (without temperature) will be worthwhile. An important point as well is that the ability of the most 531 complete linear combination to reproduce streamflow characteristics during the last 30 years was 532 confirmed not to be a statistical artefact using a fully cross validated approach. 533 534 The long-term memory for rainfall (previous 12 months), which represents the groundwater or deeper 535 antecedent soil moisture conditions, is important while short-term and current rainfalls are not 536 sufficient. It was found to be a useful variable to simulate the long-term persistence in streamflow 537 decline observed during the MD which is in supplement of the long-term rainfall-runoff elasticity 538 observed in the past. This idea could be pushed further (i.e. using the prior 10-years rainfall mean) to 539 capture decadal shift linked to the Inter-decadal Pacific oscillation (IPO) which has a signature of SEA 540 streamflow (Pui et al., 2011). The fact that antecedent catchment conditions ranging from a few 541 months to a year can have large impacts on streamflow is well known and has been shown previously 542 for Melbourne Water catchments. 543 Melbourne’s major water supply catchments under two vastly different climate scenarios. The results 544 indicated that given the moderately wet antecedent catchment conditions at the start of 2012, the 545 simulated runoff assuming a repeat of the 2006 observed daily rainfall sequences (a very low rainfall 546 year) would remain higher for three to four months before returning to the very low runoff conditions 547 similar to a dry year like 2006. In comparison, the simulated runoff assuming a repeat of the 2011 548 observed daily rainfall sequences (a relatively high rainfall year) would remain lower for a longer 549 period of about six to eight months before returning to the higher runoff conditions similar to a wet 550 year like 2011. Tan et al. (2012) investigated the runoff responses for the 21 551 552 It was also noted that an observed secular decline in streamflows was not well captured by these 553 reconstructions but there is credible evidence that it is likely due to a data issue. In the data section, it 554 was noted that for rainfall, the coverage was sparse in the earlier part of the record with an absence in 555 high elevation stations. This is likely to result in a spurious positive trend and hence it is possible that 556 the actual long-term rainfall declining trends is more pronounced than the one deduced from the 557 gridded observed rainfall and used in the reconstruction, leading to an underestimation of the 558 streamflow reduction trends. However other possibilities are worth mentioning. For example changes 559 in the accuracy of the streamflow data overtime cannot be ruled out (given the flow data prior to dam 560 construction in the Thomson and Yarra catchments were mostly derived based on flow correlation 561 with nearby stream gauging record or catchment area adjustment). It is also possible that differences 562 in catchment soil depth, non-linear hydrologic processes and thresholds effects as well as external 563 changes such as changes in land use or vegetation cover due to bushfires impacts which cannot be 564 captured by a simple multiple linear regression model have played a part in recent history. Such 565 factors may be at play in the fact that while the streamflow reduction during the MD is fully captured 566 for the Yarra catchments (-35% reconstructed for -36% observed), it is less so for the Thomson 567 catchment (-30% reconstructed for -37% observed). 568 569 Overall, for these high rainfalls high yield catchments, simple linear combination of observed rainfall 570 across the catchment can reproduce streamflow at the monthly time-scale. This simple analysis opens 571 up the possibility of providing a simple estimate of the impact of climate change on these catchments 572 using downscaled climate change projections on similar grid. This can be done using currently 573 available statistical downscaling technique (Timbal et al., 2009) which provide projections for future 574 changes of rainfall on such a high resolution grid (Timbal et al., 2011). Such work is underway and 575 will provide a benchmark against which to evaluate more complex approaches relying on several 576 hydrological models of various complexities fed with these downscaled rainfalls (Teng et al., 2012). 22 577 578 e. Conclusions 579 The additional knowledge gain in this study on the magnitude of the MD and the size of the 2010-12 580 recovery and how it differs on different side of the GDR is also very valuable; although very little 581 differences were found south of the GDR between the Yarra and Thomson on either side of a north- 582 south ridge. The significant contrast in the rainfall patterns during the 2010 and 2011 recovery 583 between the entire MDB and the Eildon north of the GDR, in comparison with the Yarra and Thomson 584 catchments south of the GDR opens up water management opportunities as all these catchments are 585 inter-connected by the water supply system connections (e.g. north-south connection). Advancing our 586 understanding of how the rainfall varies on inter-annual to longer timescale can help better inform the 587 management of these systems and potentially enhance water security, such as through inter-basin 588 transfer strategy during times of droughts and floods, and in responding to climate change. 589 590 591 Acknowledgement 592 This work has been supported by the Victorian Climate Initiative (VicCI). Inflow data were provided 593 by the Department of Environment and Primary Industries (DEPI, Victoria) courtesy of Rae Moran 594 and by Melbourne Water courtesy of Bruce Rhodes. Ian Smith and Mohammed Bari (BoM) have 595 contributed to improve an earlier version of this manuscript. We thank three anonymous reviewers for 596 very useful criticisms of an earlier version of this manuscript. 597 23 598 References 599 600 Bureau of Meteorology (2012) Australia’s wettest two-year period on record; 2010-11, Special 601 Climate 602 www.bom.gov.au/climate/current/statements/scs38.pdf. Statement No. 38, National Climate Centre, Bureau of Meteorology, 603 604 Cai, W., Cowan, T., Briggs, P., Raupach, M., 2009: Rising temperature depletes soil moisture and 605 exacerbates severe drought conditions across southeast Australia, Geophys. Res. Lett. 36, L21709. 606 http://dx.doi.org/10.1029/2009GL040334. 607 608 Cai, W, van Rensch, P., Cowan, T., and Hendon, H. 2011. Teleconnection Pathways of ENSO and the 609 IOD and the Mechanisms for Impacts on Australian Rainfall. J. Climate, 24, 3910–3923. 610 611 Chiew, F.H.S., 2006: Estimation of rainfall elasticity of streamflow in Australia. Hydrol. Sci., 51(4), 612 613-625 613 614 CSIRO, 2010: Climate variability and change in south eastern Australia - A synthesis of findings from 615 Phase 1 of the South Eastern Australian Climate Initiative (SEACI), SEACI report, 36pp, available on- 616 line at: http://www.seaci.org 617 618 CSIRO (2012) Climate and water availability in south-eastern Australia: A synthesis of findings from 619 Phase 2 of the South Eastern Australian Climate Initiative (SEACI), CSIRO, Australia, September 620 2012, 41 pp. 621 622 Drosdowsky, W., 2005: The latitude of the subtropical ridge over eastern Australia: the L index 623 revisited, Int. J. Climatol., 25, 1291-1299. 24 624 625 Evans, J. P., and I. Boyer-Souchet (2012), Local sea surface temperatures add to extreme precipitation 626 in northeast Australia during La Niña, Geophys. Res. Lett., 39, L10803, doi:10.1029/2012GL052014. 627 628 Gallant, A.J.E. and Gergis, J., 2011: An experimental streamflow reconstruction for the River Murray, 629 Australia, 1783–1988, Water Resources Research, 47, doi:10.1029/2010WR009832. 630 631 Hendon H. H., Lim E.-P., Anderson D. L. T. and J. Arblaster, 2013: Causes and predictability of 632 record wet spring Australia 2010, Mon. Wea. Rev., submitted 633 634 Howe, C., R.N. Jones, S. Maheepala and B. Rhodes, 2005: Implications of Potential Climate Change 635 for Melbourne’s Water Resources. A collaborative Project between Melbourne Water and CSIRO 636 Urban Water and Climate Impact Groups 637 638 Hutchinson, M. F. 1995: Interpolating mean rainfall using thin plate smoothing splines. Int. J. Geog. 639 Inf. Sys., 9, 385-403. 640 641 Jones, D.A. and Weymouth, G. 1997: An Australian monthly rainfall dataset, Technical Report 70. 642 Bureau of Meteorology: Melbourne, Australia, 19pp. 643 644 Jones, D.A., W. Wang and R. Fawcett, 2009: High quality spatial climate data sets for Australia, Aust. 645 Met. Oceanogr. J., 58, 233-48. 646 647 Kiem, A.S., Verdon-Kidd, D.C., 2009. Climate drivers of Victorian streamflow: Is ENSO the 648 dominant influence. Aust. J. of Wat. Res., 13(1), 17-29 649 25 650 Kiem, A.S., Verdon-Kidd, D.C., 2010. Towards understanding hydroclimatic change in Victoria, 651 Australia – preliminary insight into the "Big Dry". Hydro. Earth. Syst. Sci., 14, 433-445 652 653 Larsen, S.H. and N. Nicholls, 2012: The nature of the recent rainfall decrease in the vicinity of 654 Melbourne, southeastern Australia, and its impact on soil water balance and groundwater recharge, 655 Australian Meteorological and Oceanographic Journal, 62(1), 11-18 656 657 Leblanc, M., S. Tweed, A. Van Dijk and B. Timbal, 2011: A review of historic and future hydrological 658 changes in the Murray-Darling Basin, Global Planetary Change, doi:10.1016/j.gloplacha.2011.10.012. 659 660 Lockart, N., D. Kavetski, and S. W. Franks, 2009: On the recent warming in the Murray-Darling 661 Basin: 662 doi:10.1029/2009GL040598. Land surface interactions misunderstood, Geophys. Res. Lett., 36, L24405, 663 664 Nguyen, H, B. Timbal, I. Smith and C. Lucas, 2013: The Hadley Circulation in Reanalyses: 665 climatology, variability and expansion, J. of Climate, 26(10), 3357-3376 666 667 Nicholls, N., 2004: The changing nature of Australian droughts, Climatic Change, 63(3), 323-336 668 669 Nicholls, N., 2009: Local and remote causes of the southern Australian autumn-winter rainfall decline, 670 1958–2007, Climate Dynamics, DOI 10.1007/s00382-009-0527-6 671 672 Pepler, A., C. Rakich, B. Timbal and A. Coutts-Smith, 2014: A mechanism to explain the absence of 673 relationship between tropical modes of variability and Australian rainfall along the Eastern seaboard, 674 Journal of Climate, in press. 675 26 676 Pui, A., A. Lal and A. Sharma, 2011: How does the Interdecadal Pacific Oscillation affect design 677 floods in Australia?, Water Resources Research, 47(5), doi:10.1029/2010wr009420 678 679 Post, D.A., Moran R.J. 2013: Provision of usable projections of future water availability for south- 680 eastern Australia: The South Eastern Australian Climate Initiative. Australian Journal of Water 681 Resources, 17.2, 135-142 682 683 Post, D., B. Timbal, F. Chew, H. Hendon, H. Nguyen and R. Moran, 2014: "Decrease in south-eastern 684 Australia water availability linked 685 DOI: 10.1002/2013EF000194 to ongoing Hadley cell expansion", Earth Future, 686 687 Potter N.J., Chiew, F.H.S., Frost, A.J., 2010: An assessment of the severity of recent reductions in 688 rainfall and runoff in the Murray-Darling Basin, J. of Hydro., 381, 52-64 689 690 Preston, B. and R. Jones, 2008: Evaluating sources of uncertainty in Australian runoff-projections. 691 Advance Water Resources, 31(5), 758-775 692 693 Risbey, J.S., M.J. Pook, P.C. McIntosh, M.C. Wheeler and H.H. Hendon, 2009: On the Remote 694 Drivers of Rainfall Variability in Australia, Monthly Weather Review, 137, 3233-3253 695 696 Robertson D.E. and Q.J. Wang, 2012: A Bayesian approach to predictor selection for seasonal 697 streamflow forecasting, Journal of Hydrometeorology, 13, 155-171, doi: 10.1175/JHM-D-10-05009.1. 698 699 Saji N.H., Goswami, B.N., Vinayachandran P.N. and Yamagata, T. 1999: A dipole mode in the 700 tropical Indian Ocean, Nature, 401, 360–3. 701 27 702 Smith, I. and B. Timbal, 2012: Links between tropical indices and southern Australia rainfall, Int. J. of 703 Clim., 32(1), 33-40, DOI: 10.1002/joc.2251 704 705 Smith, I. and S. Power, 2014: Past and future declines in rainfall and inflow into Perth (Western 706 Australia) dams, Journal of hydrology: Regional studies, 2, 84-96 707 708 Tan K.S., Crawford-Wilson P. and Rhodes B., 2012: Assessing catchment runoff responses under dry 709 and wet year scenarios – a case study, Proceedings of 34th Hydrology and Water Resources 710 Symposium, 19-22 November, Sydney, Engineers Australia, p1157-1164. 711 712 Tan K.S. and Rhodes B.G., 2013: Drought severity estimation under a changing climate, Australian 713 Journal of Water Resources, 17(2), 143-151. 714 715 Teng, J., F.H.S. Chiew, B. Timbal, Y. Wang, J. Vaze, B. Wang, 2012: “Assessment of an analogue 716 downscaling method for modelling climate change impact on runoff”, Journal of Hydrology, 717 doi:10.1016/j.jhydrol.2012.09.024 718 719 Timbal B. and Murphy, B. 2007: Observed climate change in South-East of Australia and its relation 720 to large-scale modes of variability, BMRC Res. Lett., 6, 6-11. 721 722 Timbal, B., 2009: “The continuing decline in South-East Australian rainfall – Update to May 2009”, 723 CAWCR Res. Let., 2, 4-11 724 725 Timbal, B., Fernandez, E. and Li, Z, (2009): Generalization of a statistical downscaling model to 726 provide local climate change projections for Australia. Environmental Modelling & Software, 24 (3), 727 341-358 728 28 729 Timbal, B., J. Arblaster, K. Braganza, E. Fernandez, H. Hendon, B. Murphy, M. Raupach, C. Rakich, 730 I. Smith, K. Whan and M. Wheeler, 2010: Understanding the anthropogenic nature of the observed 731 rainfall decline across south-eastern Australia, CAWCR Technical Report 26, 180pp. 732 733 Timbal B. and H. Hendon, 2011: The role of tropical modes of variability in the current rainfall deficit 734 across the Murray-Darling basin, Water Resources Research, doi:10.1029/2010WR009834. 735 736 Timbal, B., Y. Wang and A. Evans, 2011: “Downscaling climate change information: an essential 737 ingredient to incorporate uncertainties into adaptation policies”, In Chan, F., Marinova, D. and 738 Anderssen, R.S. (eds) MODSIM2011, 19th International Congress on Modelling and Simulation. 739 Modelling and Simulation Society of Australia and New Zealand, December 2011, pp. 1652-1658. 740 ISBN: 978-0-9872143-1-7 741 742 Timbal, B. and W. Drosdowsky, 2013: “The relationship between the decline of South Eastern 743 Australia rainfall and the strengthening of the sub-tropical ridge”, Int. J. of Climatol., 33, 1021-1034, 744 DOI: 10.1002/joc.3492 745 746 Timbal, B. and R. Fawcett, 2013: “An historical perspective on South Eastern Australia rainfall since 747 1865”, J. of Climate., 26, 1112-1129 748 749 Ummenhofer, C. C., M. H. England, P. C. McIntosh, G. A. Meyers, M. J. Pook, J. S. Risbey, A. S. 750 Gupta, and A. S. Taschetto, 2009: What causes southeast Australia’s worst droughts?, Geophys. Res. 751 Let., 36, L04706, doi:10.1029/2008GL036801 752 753 Whan, K. B. Timbal and J. Lindsay, 2013: “Linear and nonlinear statistical analysis of the impact of 754 the sub-tropical ridge intensity and position on south-east Australian rainfall”, Int. J. of Climatol., 755 DOI: 10.1002/joc.3689 29 756 757 758 30 759 List of Figures 760 761 762 763 764 Figure 1: Map of South Eastern part of the Australian continent (upper panel) detailing the catchment boundaries for the Murray Darling basin (red line), the Eildon catchment and the main Melbourne Water catchment areas (enlarged in the lower panel). Colours show the height above sea level in metres. Dark blue shows the location of the reservoirs. The gridded observation resolution is shown by the white crosses. The location of "The triangle" is shown with a red symbol. ............................... 37 765 766 767 768 769 770 Figure 2: Map of the area of interest (upper panel) showing the location of weather stations contributing to the data set and how they have changed over time. Red circles show 1900-1970, green circles 1970-1980, blue circles 1980-2010, red triangles 1900-1980, green triangles 1970-2010 and blue triangles 1900-2010. The rectangles identify the three regions shown in the bottom panels: time series of the number of stations in the each of the three regions (left) and the average altitude of stations in each of the three regions (right)............................................................................................ 38 771 772 773 Figure 3: 1900 to 2012 mean rainfall based on the Bureau of Meteorology operational 0.05 degree gridded rainfall for the annual total (top panel) and the four seasonal means: summer (middle left), autumn (middle right), winter (bottom left) and spring (bottom right). ................................................ 39 774 775 776 777 Figure 4: Monthly 1900-2012 rainfall climatology (left column) compared with the 1997-2009 values and anomalies for each of the four catchment areas (MDB: top, and then Eildon, Yarra and Thomson). Same number but for the streamflows in three of the catchments; the monthly streamflow anomalies are provided in absolute term (green bars) and percentage (orange bars). ............................................ 40 778 Figure 5: As per Figure 3 but for an average of 2010 and 2011. ........................................................... 41 779 780 781 782 Figure 6: Accumulated rainfall anomalies for each of the catchment areas for three different time periods. The red curve shows the deficit for the 1997 to 2012 period. The blue curve shows the 19942009 period, encompassing the MD. The black curve shows the 16 year period with the lowest deficit prior to 1994........................................................................................................................................... 42 783 784 785 786 787 Figure 7: Location of indices of large-scale modes of variability used in this study: Niño4 SST region (the red rectangular), Saji et al. (1999) Dipole Mode Index (DMI, yellow boxes), Timbal and Hendon (2011) tropical tripole (blue shapes), Drosdowsky (2005) subtropical ridge position (green lines): the mean annual position is shown as a solid line, with northern (winter) and southern (summer) extent shown as dashed lines. ........................................................................................................................... 43 788 789 790 791 Figure 8: Correlation coefficients in winter (left) and spring (right) between the rainfall and climate indices (N4, top row, the tripole, second row, the STR position, third row and intensity, fourth row). Correlations coefficients larger than +/-0.25 are significant at the 99% level. Correlations coefficients are computed on time-series from 1913 to 2012. .................................................................................. 44 792 793 794 Figure 9: Time series of the tripole index. The bars show the value of the tripole index for each year. The colours of the bars show significant El Nino (red) and La Nina (blue) years, with yellow being neutral, the coloured stars show significant positive (red) and negative (blue) IOD years. The 31 795 796 horizontal dashed lines show 0.8 standard deviation of the tripole index. The vertical dashed line marks the start of 1997........................................................................................................................... 45 797 798 799 800 Figure 10: Eildon, Thomson and Yarra composite streamflows based on the tripole index values. Blue lines are computed using years prior to and including 1996; red lines show the composites computed since and including 1997. Lines with stars are the mean over the period; lines with diamonds, are years with a positive tripole, and with triangles, years with a negative tripole. ............. 47 801 802 List of Tables: 803 804 805 806 807 808 Table 1 Characteristics of the Millennium Drought (1997-2009) compared to the next most intense lowest 13-years period on record not overlapping with the MD. Anomalies are given as percentage from the long-term mean. The annual rainfalls of the wettest year embedded within the 13-year droughts are indicated as well as the ranking of that year and the anomalies form the long-term mean in percentage term. For both the MD and previous lowest 13 years on record, seasonal anomalies are also indicated in percentage terms. All statistics are based upon the complete 1900 to 2012 dataset. . 33 809 810 811 812 813 814 Table 2: Statistics for 2010 and 2011 rainfall across 5 areas of interest: The Murray-Darling Basin, the Eildon catchment, the Melbourne Water major catchments (Thomson and Yarra) and the SouthEastern Australia as a whole. Highest rainfall on record prior to 2010 (1900 - 2009) are indicated as a bench mark. Annual rainfall and ranking are provided for 2010 and 2011 (ranking are based on the full 1900 to 2011 database) cool months (April to October) and warm months (January to March and November to December). Bold font indicates record breaking statistics .............................................. 34 815 816 817 818 819 Table 3: Inter-annual variance explained (squared correlation coefficients) and percentage of the observed variance reproduced by the reconstructions. All statistics are computed using the entire series from 1913 to 2012 and for the 5 linear reconstructions: R 1 to 5 : predictors are rainfall of the month (R), rainfall from the previous month (R -1 ), rainfall from the previous 12 month (R -12 ) and temperature of the month (T). .................................................................................................................................... 35 820 821 822 Table 4: Statistics for observed rainfall and streamflow (in bold) as well as for the 5 linear streamflow reconstructions (see main text for details) for the three catchment areas: linear trends from the 1913 to 2012, from 1980 to 2012 and for the magnitude of the deficit during the MD. .................................... 36 823 32 824 Tables 825 826 Previous 13-year dry (PD) period Rainfall deficit Wettest year Summer anomaly Autumn anomaly Winter anomaly Spring anomaly MD PD MD Total (rank) Anomaly PD Total (rank) Anomaly MD PD MD PD MD PD MD PD MDB 19341946 -15 % -13 % 577 (15) 23% 543 (25) 16% -6% 11% -24% -16% -4% -7% +3% -26% Eildon 19021914 - 18 % -8 % 1090 (62) -2% 1284 (23) 15% -8% -10% -29% -1% -16% +1% -12% -5% Yarra 19021914 - 17 % -4% 1476 (61) -2% 1735 (19) 15% -8% -8% -25% +1% -13% 2% -14% -5% Thomson 19021914 - 15 % - 5% 1420 (49) 2% 1566 (22) 13% -8% -9% -24% -4% -12% +1% -13% -3% SEA 19371949 -12% -8% 611 (44) 5% 697 (13) 20% -6% +8% -26% -5% -9% -11% -6% -13% 827 828 Table 1 Characteristics of the MD (1997-2009) compared to the next most intense lowest 13-years 829 period on record not overlapping with the MD across 5 areas of interest: the Murray-Darling Basin, the 830 Eildon catchment, the Melbourne Water major catchments (Thomson and Yarra) and the South- 831 Eastern Australia as a whole. Anomalies are given as percentage from the long-term mean. The annual 832 rainfalls of the wettest year embedded within the 13-year droughts are indicated as well as the ranking 833 of that year and the anomalies form the long-term mean in percentage term. For both the MD and 834 previous lowest 13 years on record, seasonal anomalies are also indicated in percentage terms. All 835 statistics are based upon the complete 1900 to 2012 dataset. 836 33 837 838 MDB Highest annual rainfall between 1900 and 2009 2010 annual rainfall (ranking) 2011 annual rainfall (ranking) 784 (1956) Eildon Yarra Thomson 1724 (1956) 2275 (1952) 2245 (1952) SEA 878 (1956) 807 (1) 600 (13) 1516 (7) 1226 (36) 1702 (29) 1721 (26) 1557 (26) 1602 (18) 811 (4) 718 (11) Longterm mean Warm month rainfall 259 439 626 591 261 Cool month rainfall 212 676 879 799 323 1997 2009 Warm month rainfall 255 363 520 501 236 Cool month rainfall 191 556 730 675 279 Warm month rainfall Cool month rainfall Warm month rainfall Cool month rainfall 538 (1) 269 (21) 432 (5) 168 (86) 800 (1) 717 (48) 664 (7) 564 (87) 818 (11) 884 (57) 886 (6) 833 (69) 86 ( 781 (11) 773 (65) 816 (6) 785 (60) 469 (1) 342 (42) 450 (3) 267 (84) 2010 2011 839 840 Table 2: Statistics for 2010 and 2011 rainfall and compare with highest rainfall for 1900 to 2009 as a 841 bench mark for the five areas of interest as per table 1. Annual rainfall and ranking are provided for 842 2010 and 2011 (ranking are based on the full 1900 to 2011 database) cool months (April to October) 843 and warm months (January to March and November to December). Bold font indicates record- 844 breaking statistics 845 34 846 847 Reconstruction Predictors Eildon Yarra Thomson R1 R, R -1 , R -12 , T 0.90 / 84% 0.85 / 78% 0.77 / 74% R2 R, R -1 , R -12 , T 0.90 / 80% 0.85 / 72% 0.81 / 51% R3 R, R -1 , R -12 0.90 / 84% 0.85 / 78% 0.77 / 74% R4 R, R -1 0.85 / 64% 0.74 / 45% 0.74 / 53% R5 R 0.79 / 34% 0.69 / 18% 0.69 / 27% 848 849 Table 3: Inter-annual variance explained (squared correlation coefficients) and percentage of the 850 observed variance ranges reproduced by the reconstructions. All statistics are computed using the 851 entire series from 1913 to 2012 and for the 5 linear reconstructions (see main text for details): 852 predictors are rainfall of the month (R), rainfall from the previous month (R -1 ), rainfall from the 853 previous 12 month (R -12 ) and temperature of the month (T). 854 35 855 856 1997 – 2009 anomaly (% from longterm mean) 1980 to 2012 linear trend As percentage of 1980-2012 mean 1913 to 2012 linear trend (as a percentage of mean) (mm/decade) Rainfall Inflow R1 R2 R3 R4 R5 Rainfall Inflow R1 R2 R3 R4 R5 Rainfall Inflow R1 R2 R3 R4 R5 Eildon -18 -42 -40 -41 -40 -30 -20 -13 -38 -50 -49 -49 -42 -29 -6 -34 -17 N/A -17 -14 -9 Yarra -17 -36 -35 -34 -34 -21 -13 -11 -23 -29 -29 -29 -21 -12 -8 -34 -16 N/A -16 -10 -6 Thomson -16 -37 -30 -28 -30 -22 -15 -7 -18 -19 -15 -17 -14 -10 -8 -36 -15 N/A -15 -11 -8 857 858 Table 4: Statistics for observed rainfall and streamflow (in bold) as well as for the 5 linear streamflow 859 reconstructions (see main text for details) for the three catchment areas: magnitude of the deficit 860 during the MD, linear trends from 1980 to 2012 and from 1913 to 2012. 861 36 862 863 864 Figure 1: Map of South Eastern part of the Australian continent (upper panel) detailing the catchment 865 boundaries for the Murray Darling basin (red line), the Eildon catchment and the main Melbourne 866 Water catchment areas (enlarged in the lower panel). Colours show the height above sea level in 867 metres. Dark blue shows the location of the reservoirs. The gridded observation resolution is shown 868 by the white crosses. The location of hill top called "The triangle" is shown with a red symbol. 869 37 870 871 872 873 874 875 Figure 2: Map of the area of interest (upper panel) showing the location of weather stations 876 contributing to the gridded data set. Red circles show stations in existence from 1900 to 1970, green 877 circles from 1970 to 1980, blue circles from 1980 to 2010, red triangles from 1900 to 1980, green 878 triangles from 1970 to 2010 and blue triangles from 1900 to 2010. The coloured rectangles on the top 879 panel identify the three regions for which two statistics are provided in the bottom panels: time series 880 of the number of stations (left panel) and the average altitude of stations (right panel). 881 38 882 883 884 885 886 887 888 889 Figure 3: 1900 to 2012 mean rainfall based on the Bureau of Meteorology operational 0.05 degree 890 gridded rainfall for the annual total (top panel) and the four seasonal means: summer (middle left), 891 autumn (middle right), winter (bottom left) and spring (bottom right). 892 39 893 894 895 Figure 4: Monthly 1900-2012 rainfall climatology (left column) compared with the 1997-2009 values 896 and anomalies for each of the four catchment areas (MDB: top, and then Eildon, Yarra and Thomson). 897 Same number but for the streamflows in three of the catchments; the monthly streamflow anomalies 898 are provided in absolute term (green bars) and percentage (orange bars). 40 899 900 901 902 903 904 905 906 907 Figure 5: As per Figure 3 but for 2010 and 2011 averaged. 908 41 909 910 911 912 Figure 6: Accumulated rainfall anomalies for each of the catchment areas for three different time 913 periods. The red curve shows the deficit for the 1997 to 2012 period. The blue curve shows the 1994- 914 2009 period encompassing the MD which in all cases is the lowest 16-year period on record. The 915 black curve shows the 16 year period with the lowest accumulated deficit prior to 1994. 42 916 917 918 919 920 921 922 923 924 Figure 7: Location of indices of large-scale modes of variability used in this study: Niño4 SST region 925 (the red rectangular), Saji et al. (1999) Indian Ocean Dipole (IOD, yellow boxes), Timbal and Hendon 926 (2011) tropical tripole (blue shapes), Drosdowsky (2005) subtropical ridge position (green lines): the 927 mean annual position is shown as a solid line, with northern (winter) and southern (summer) extent 928 shown as dashed lines. 929 930 931 43 932 933 934 935 936 937 938 939 940 Figure 8: Correlation coefficients in winter (left) and spring (right) between the rainfall and climate 941 indices (N4, top row, the tripole, second row, the STR position, third row and intensity, fourth row). 942 Correlations coefficients larger than +/-0.25 are significant at the 99% level. Correlations coefficients 943 are computed on time-series from 1913 to 2012. 44 944 945 946 947 948 Figure 9: Time series (coloured bars) of the annual tripole index. The colours of the bars show 949 significant El Niño (red) and La Niña (blue) years, with yellow being neutral, the coloured stars show 950 significant positive (red) and negative (blue) IOD years. The horizontal dashed lines show ±0.8 951 standard deviation of the tripole index separating either positive or negative tripole years from neutral 952 years. The vertical dashed line marks the start of 1997. 953 954 45 955 956 957 958 46 959 960 961 Figure 10: Eildon, Thomson and Yarra composite streamflows based on the tripole index values. 962 Blue lines are computed using years prior to and including 1996; red lines show the composites 963 computed since and including 1997. Lines with stars are the mean over the period; lines with 964 diamonds, are years with a positive tripole, and with triangles, years with a negative tripole. 965 47
© Copyright 2024