An ice core derived 1013-year catchment-scale annual rainfall reconstruction in subtropical eastern Australia

Paleoclimate research indicates that the Australian instrumental climate record (∼ 100 years) does not cover the full range of hydroclimatic variability that is possible. To better understand the implications of this on catchmentscale water resources management, a 1013-year (1000– 2012 common era (CE)) annual rainfall reconstruction was produced for the Williams River catchment in coastal eastern Australia. No high-resolution paleoclimate proxies are located in the region and so a teleconnection between summer sea salt deposition recorded in ice cores from East Antarctica and rainfall variability in eastern Australia was exploited to reconstruct the catchment-scale rainfall record. The reconstruction shows that significantly longer and more frequent wet and dry periods were experienced in the preinstrumental compared to the instrumental period. This suggests that existing drought and flood risk assessments underestimate the true risks due to the reliance on data and statistics obtained from only the instrumental record. This raises questions about the robustness of existing water security and flood protection measures and has serious implications for water resources management, infrastructure design and catchment planning. The method used in this proof of concept study is transferable and enables similar insights into the true risk of flood/drought to be gained for other paleoclimate proxy poor regions for which suitable remote teleconnected proxies exist. This will lead to improved understanding and ability to deal with the impacts of multi-decadal to centennial hydroclimatic variability.


Introduction
Water and catchment management systems (e.g., drought and flood mitigation strategies) and water resources infrastructure have traditionally been designed based on the trends, patterns and statistics revealed in relatively short instrumental climate records (i.e., for Australia usually less than 100 years of data recorded post-1900) (Verdon-Kidd and Kiem, 2010;Ho et al., 2014;Cosgrove and Loucks, 2015;Razavi et al., 2015).This is a concern as Australian paleoclimate research suggests that instrumental climate records are not representative of the true range of hydroclimatic variability possible (Verdon-Kidd and Kiem, 2010;Gallant and Gergis, 2011;Kiem and Verdon-Kidd, 2011;Ho et al., 2014Ho et al., , 2015a, b;, b;Razavi et al., 2015;Vance et al., 2015).For example, paleoclimate archives show evidence of droughts of longer duration than the three major droughts that have affected eastern Australia over the instrumental period -the Federation drought (∼ 1895-1902), World War II drought (∼ 1937-1945) and Millennium or "Big Dry" drought (∼ 1997-2009) (Gergis et al., 2012;Vance et al., 2013;Allen et al., 2015;Vance et al., 2015).
Sources for paleoclimate proxy data include tree rings, coral skeletons, ice cores, speleothems (cave deposits), sediments and documentary evidence (Ho et al., 2014).Ideally, the climate proxy archives are located in the region of interest but in areas where proxy records are sparse or of low resolution, remote proxies are a viable alternative (Ho et al., 2014).Remote proxies exploit circulation teleconnections that link one region to another and are calibrated over Published by Copernicus Publications on behalf of the European Geosciences Union.the instrumental period, to develop paleoclimate reconstructions (e.g., rainfall, streamflow) for the target region (e.g., Verdon and Franks, 2007;McGowan et al., 2009;van Ommen and Morgan, 2010;Vance et al., 2013Vance et al., , 2015)).When using remote proxies the assumption is that large-scale climate processes driving climate variability at the location of the paleoclimate proxy also drive a high proportion of climate variability at the region of interest (Gallant and Gergis, 2011).For example, van Ommen and Morgan (2010) identified a relationship between precipitation (snowfall) recorded in ice cores from coastal Antarctica and rainfall in southwest Western Australia over the instrumental period, inferring rainfall variability in the region for the past 750 years.Similarly, Lough (2011) found significant correlations between coral luminescence intensity recorded in coral cores from the Great Barrier Reef and summer rainfall variability in northeast Queensland, which enabled the multi-century coral record to be used to reconstruct Queensland summer rainfall back to the 18th century.
Another option is to use the link between large-scale ocean-atmospheric climate processes and climate variability in the region of interest to develop a paleoclimate reconstruction based on a paleoclimate proxy of the climate process.For example, McGowan et al. (2009) reconstructed annual inflows in the Murray River back to 1474 common era (CE) from a reconstruction of the Pacific Decadal Oscillation (PDO) based on the previously identified relationship between the PDO and streamflow in southeastern Australia (e.g., Power et al., 1999a, b;Kiem et al., 2003;Kiem and Franks, 2004;Verdon et al., 2004).A similar approach was also followed by Verdon andFranks (2006, 2007) and Henley et al. (2011).Vance et al. (2013Vance et al. ( , 2015) ) used a hybrid of the approaches discussed above.Vance et al. (2013) developed a millennial length rainfall reconstruction for subtropical eastern Australia by exploiting a relationship between the region's annual rainfall and the summer sea salt record (see Sect. 3) from the Law Dome ice core, East Antarctica (Fig. 1).Of key importance is that the strength of the relationship during the instrumental period (in this case 1889-2009) varies synchronously with the Interdecadal Pacific Oscillation (IPO) (Power et al., 1999a, b), the basin-wide expression of the PDO, with increased correlations found during IPO positive phases (Vance et al., 2013(Vance et al., , 2015)).The IPO represents decadal sea surface temperature (SST) variability across the Pacific Ocean whereby a positive IPO phase is associated with warming across the tropical Pacific and cooling of the north and south Pacific; the opposite occurs during the negative phase (Power et al., 1999a).The most recent defined complete IPO phases are two positive phases (∼ 1924-1941, ∼ 1979-1997) and one negative phase (∼ 1947-1975) (Power et al., 1999a;Kiem et al., 2003;Kiem and Franks, 2004;Verdon et al., 2004).Vance et al. (2015) demonstrated that during the IPO negative phase there is a predominantly zonal pressure pattern across the high to mid-latitudes, which switches to a more meridional pattern in IPO positive.Folland et al. (2002) also found that during the IPO positive phase, the mean posi-tion of the South Pacific Convergence Zone (SPCZ) (usually bounded by Samoa and Fiji) is displaced to the northeast.This northeastern displacement is associated with a more meridional circulation pattern and enhances the link between eastern Australia and mid-to high-latitude climate variability and hence explains the stronger relationship between sea salt recorded at Law Dome and rainfall in eastern Australia during the IPO positive phase.Based on their reconstruction of the IPO, Vance et al. (2015) could therefore identify periods in time (i.e., positive IPO phases) where they had greater confidence in the rainfall reconstruction.A key finding from Vance et al. (2015) was the identification of a century of IPO positive aridity (1102( -1212 CE) CE), including evidence of a 39year drought in southeast Queensland, which is well outside the bounds of instrumental drought duration.This illustrates the importance of investigating climate variability over millennial timescales, particularly in the Southern Hemisphere where many paleoclimate records only span the last 200-500 years (Neukom and Gergis, 2012).Indeed, it is evident that (a) instrumental data are not long enough to allow for meaningful planning for climate variability; (b) paleodata, particularly at the millennial timescale, offers an important insight into the climate beyond the instrumental period; and (c) there is a need to incorporate insights from paleodata into water resources planning and management.
Further work is also required to assess the robustness of the relationship between climate variability in East Antarctica, large-scale climate processes and eastern Australia, a region with limited local paleoclimate proxy data (Vance et al., 2013;Ho et al., 2014).Practical usefulness of the insights provided by the paleoclimate reconstructions for water resources management at the catchment scale also requires investigation.Therefore, the links between the Law Dome sea salt record, eastern Australian rainfall and the IPO are further explored in this paper through the development of a millennial length, annual resolution and catchmentscale rainfall reconstruction for the Williams River (WR) catchment (Fig. 1).The WR catchment is located on the eastern seaboard of New South Wales, east of the Great Dividing Range (Fig. 1).The eastern seaboard contains about half of Australia's population, and a proportionate amount of economic infrastructure and activity.The region has hydroclimate features that are distinct from the rest of Australia (e.g., Verdon and Franks, 2005;Timbal, 2010) and lacks high-resolution paleoclimate proxies (Ho et al., 2014).This means there is significant vulnerability, uncertainty and knowledge gaps relating to flood and drought risk in eastern Australia.This recognition has recently motivated the development of the Eastern Seaboard Climate Change Initiative (ESCCI), to better understand the causes and impacts of current and future climate-related risk in the region (http://www.climatechange.environment.nsw.gov.au/Aboutclimate-change-in-NSW/Evidence-of-climatechange/Eastern-seaboard-climate-change-initiative).
The WR catchment is of particular regional importance because it forms part of the conjunctive-use headworks scheme for potable water supply to ∼ 600 000 people in Newcastle, the sixth largest residential region in Australia (Kiem and Franks, 2004;Mortazavi-Naeini et al., 2015).
In the following sections we present a description of the WR catchment location and relevant climate data, including a discussion of the link between Law Dome, East Antarctica and eastern Australia.We proceed with an investigation into the relationship between summer sea salts from Law Dome and rainfall in the WR catchment and follow with the development of a 1013-year catchment-scale rainfall reconstruction (based on the Law Dome sea salt record) and discussion of the insights and implications emerging from this reconstruction.

Rainfall variability in the Williams River catchment
For the calibration data in this study we used daily 5 km × 5 km gridded rainfall data obtained from the Australian Water Availability Project (AWAP) (Jones et al., 2009) for the period 1900-2010.The AWAP grid cells overlapping the WR catchment were extracted and used to calculate catchment average monthly rainfall totals for the WR catchment.Due to known biases and uncertainty associated with gridded climate data (e.g., Tozer et al., 2012), the AWAP-based information was ground-truthed with data from a high-quality (Lavery et al., 1997) rainfall gauge (61010) located within the WR catchment.The highest and most variable rainfall in the WR catchment is received from December to May (summer and autumn) (Fig. 2) and the hydrological water year for the WR catchment is therefore defined as October to September in order to encompass this high rainfall period (pers. comm., Brendan Berghout, Senior Water Resources Engineer, Hunter Water Commission).
Rainfall variability in the WR catchment is influenced by the Great Dividing Range to the west (Fig. 1), which provides orographic enhancement and the Tasman Sea to the east, which brings moisture to the region (Pepler et al., 2014).

C. R. Tozer et al.: An ice core derived 1013-year catchment-scale annual rainfall reconstruction
Synoptic scale influences known as East Coast Lows (ECLs), marine or continental low-pressure systems, which tend to develop in the Tasman Sea, are responsible for much of the extreme weather (e.g., heavy rainfall, high winds) recorded in eastern New South Wales (Speer et al., 2009;Browning and Goodwin, 2013;Pepler et al., 2014b;Ji et al., 2015;Kiem et al., 2015;Twomey and Kiem, 2015).Indeed, ECLs have been found to contribute 20-30 % of annual rainfall in the WR region (Pepler et al., 2014a).In addition to these local influences several large-scale ocean-atmospheric processes influence rainfall in the WR catchment (e.g., Kiem andFranks, 2001, 2004;Risbey et al., 2009).The El Niño-Southern Oscillation (ENSO) and IPO have been related to interannual to multi-decadal variability in both WR rainfall and runoff (Kiem andFranks, 2001, 2004).Drier (wetter) catchment conditions typically occur during El Niño (La Niña) events and the IPO modulates both the frequency and magnitude of ENSO impacts such that drought risk is increased during IPO positive phases and flood risk is increased during IPO negative phases (Kiem andFranks, 2001, 2004;Kiem et al., 2003;Kiem and Verdon-Kidd, 2013).Indian Ocean SSTs are also known to influence eastern Australian rainfall during winter and spring (Verdon and Franks, 2005;Risbey et al., 2009).
In addition, the subtropical ridge (STR) and Southern Annular Mode (SAM) impact rainfall variability in the eastern seaboard (e.g., Risbey et al., 2009;Ho et al., 2012;Whan et al., 2013).A positive SAM phase has been related to increased daily rainfall in summer and spring (Hendon et al., 2007;Risbey et al., 2009) while variability in the position of the STR is significantly correlated with rainfall in the eastern seaboard.That is, a shift south of the STR is associated with increased rainfall in the region (Timbal, 2010;Whan et al., 2013).Variability in the intensity of the STR is also related to rainfall variability in the eastern seaboard though to a lesser extent than variability in the STR position (Timbal, 2010).

Law Dome ice core site details
Law Dome is a small, coastal icecap located in Wilkes Land, East Antarctica (Fig. 1), and the site of the Dome Summit South (DSS) ice core, which spans around 90 000 years (Roberts et al., 2015).DSS has high annual snowfall of around 0.63 m (water equivalent), which allows for a monthly resolution record in the upper portion of the core and highly accurate dating (van Ommen and Morgan, 1997;Vance et al., 2013;Roberts et al., 2015).The ice core was dated by counting annual layers with known volcanic horizons used to establish dating accuracy (Plummer et al., 2012).As a result, the Law Dome record was dated with absolute accuracy from 1807 to 2009 CE and with ±1-year error from 894 to 1807 CE.The sea salt record used here was produced using trace ion chromatography from 2.5 to 5 cm sub-samples of the ice cores (Curran et al., 1998;Palmer et al., 2001).The Law Dome summer (December-March) sea salts (LD SSS ) were extracted from the full record and used by Vance et al. (2013) and Vance et al. (2015) as a rainfall proxy for eastern Australia.Here we use a slightly extended LD SSS record to cover the epoch 1000-2012 CE using the improved composite record of Roberts et al. (2015).

The link between sea salt deposition at Law Dome and large-scale ocean-atmospheric processes
The climate signals recorded in the Law Dome ice core are driven by large-scale ocean-atmospheric processes rather than local factors (Bromwich, 1988;Delmotte et al., 2000;Masson-Delmotte et al., 2003;Vance et al., 2013).The southern Indian Ocean is the main source of moisture delivered to Law Dome (Delmotte et al., 2000;Masson-Delmotte et al., 2003) and sea salt deposition is related to the midlatitude westerly winds (associated with the SAM) in the Indian and Pacific sectors of the Southern Ocean (Goodwin et al., 2004;Vance et al., 2015).Seasonal to annual scale SST anomalies in the equatorial Pacific are known to propagate to high southern latitudes (Karoly, 1989;Mo and Higgins, 1998;Ding et al., 2012).The resulting circumpolar geopotential height and zonal wind anomalies influence the SAM (L' Heureux and Thompson, 2006), and ultimately deliver sea salt aerosols to coastal Antarctica (Vance et al., 2013).Indeed, Vance et al. (2013) found a significant correlation between ENSO-related SST variability in the centralwestern equatorial Pacific and LD SSS , with low summer sea salt years associated with El Niño events over the period 1889-2009.Furthermore, spectral analysis of the 1010-year LD SSS record found significant (p < 0.01) spectral features in the 2-7-year ENSO band.Similar to the LD SSS rainfall proxy discussed previously, the LD SSS ENSO proxy varies decadally, coherent with the IPO, with a stronger relationship during IPO positive phases (Vance et al., 2013(Vance et al., , 2015)).
It is thus clear that the ocean-atmospheric processes associated with sea salt deposition at Law Dome (e.g., IPO, ENSO, SAM and variability in the Indian Ocean) are the same as those that influence rainfall variability in the WR catchment (discussed in Sect.2).We can therefore expect LD SSS to explain some variability in the rainfall in the WR catchment.
4 Investigating the relationship between LD SSS and rainfall in the Williams River catchment Vance et al. (2013) found a relationship between LD SSS and the prior January-December rainfall west of the Great Dividing Range.The region of interest in this study is further south and east of the Great Dividing Range, so we needed to re-evaluate if this temporal offset was appropri- ate.This re-evaluation was via a damped least-squares regression between AWAP grid-cell data and the LD SSS record using the Marquardt-Levenberg method, a method capable of multi-variate and non-linear regression.Although it was only used for uni-variate linear regression here, the method was selected for compatibility with planned future work.For every AWAP grid cell in New South Wales, we performed linear least-squares regression between the LD SSS record and 12-month-averaged rainfall over a 24 month lead/lag range centered about the summer sea salt period (December-March).The regression coefficients for each lead/lag were used to generate an estimated spatial rainfall time series.The Pearson correlation coefficient between the estimated rainfall and AWAP rainfall for each grid cell was then assessed for each lead/lag.This process allowed us to determine the seasonal window for rainfall that optimized the WR rainfall-LD SSS relationship in order to optimize the utility of the LD SSS record.
From the lead/lag analysis, October-September and November-October annual rainfall in the region encompassing the WR catchment was found to have the highest and most spatially coherent relationship with LD SSS .We present the October-September rainfall / LD SSS correlations (Fig. 3) as this period also corresponds to the water year in the Newcastle region (discussed in Sect.2) and hence all further analysis is based on the 12-month rainfall totals calculated from October-September.
Figure 3 shows spatial maps of the magnitude of the correlations between October-September WR rainfall and LD SSS for the 1900-2010 (i.e., October 1900-September 2010) period as well as subsets for the different IPO phases.For comparison Fig. 3a-e are inset with maps for the January-December rainfall / LD SSS correlations, the analysis period Table 1.Pearson correlation values between LD SSS and 12-month average (October-September) rainfall recorded at gauge 61010 and the AWAP WR catchment average for the 1900-2010 period and IPO phases.Bootstrap 95 % confidence intervals are also indicated (Mudelsee, 2003).Bold values are significant at 95 %.
The insets of Fig. 3a-e reveal low correlations in the WR catchment region.The highest correlations occur in inland New South Wales and into southeastern Queensland, the focus region of Vance et al. (2013) and Vance et al. (2015).However, when the correlation is aligned with the WR catchment water year (October-September) we see a shift in the region of significant correlation (Fig. 3a) to coastal New South Wales and, in particular, large parts of the eastern seaboard.Importantly, correlations significant at the 99 % level are seen over the WR catchment region.Rainfall at gauge 61010 and AWAP catchment average show significant correlations with LD SSS (Pearson correlation coefficients of 0.29 and 0.28 respectively) over the 1900-2010 period (Table 1).
As expected, based on the results of Vance et al. (2013Vance et al. ( , 2015) ) (discussed in Sect.3), the strength of the correlation between October-September rainfall and LD SSS varies decadally.Figure 3b-c indicate that the relationship between the variables is stronger during the IPO positive phases relative to the negative phase. Figure 3d-e and the results in Table 1, however, suggest that although the relationship between October-September rainfall and LD SSS is stronger in IPO positive phase, this increase in strength relative to IPO negative and the full record  is primarily due to the very high correlation in the second IPO positive phase (1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997).In fact, the correlation between rainfall recorded at gauge 61010 and LD SSS during the IPO negative phase is greater than the correlation in the first IPO positive phase (Table 1).Figure 3f further highlights that an increase (decrease) in the strength of the LD SSS -WR rainfall relation-ship is not always synchronous with IPO positive (negative) phases.
This result appears to be in contrast to Vance et al. (2013Vance et al. ( , 2015)), who found a clear link between IPO phase, LD SSS and January-December rainfall in southeast Queensland and northeastern New South Wales (west of the Great Dividing Range).That is, the correlation between these variables was poor during the IPO negative phase, yet was significant for both positive phases.Vance et al. (2013Vance et al. ( , 2015) ) used calendar year rainfall as opposed to a more catchment specific analysis period used in this study.Another key difference is that the focus region here is further south, on the coast and under the orographic influence of the Great Dividing Range.Furthermore, although interdecadal and interannual tropical Pacific Ocean variability (e.g., ENSO and IPO) has been found to impact the whole of Australia at various times of the year (e.g., Power et al., 1999a;Risbey et al., 2009), the amount of rainfall variability explained by these processes reduces to the south while climate mechanisms stemming from the mid-to high latitudes (e.g., SAM and the STR, discussed in Sect.2) increase their influence on rainfall variability (Risbey et al., 2009).
In addition, as mentioned previously, around one-quarter of annual rainfall received in the WR catchment results from ECLs (Pepler et al., 2014a).In 1950 and 1955, the Newcastle region experienced heavy rainfall and floods as a result of severe ECLs (Callaghan and Helman, 2008;Callaghan and Power, 2014) and indeed eastern Australia in general was subject to an increase in intense storm activity between the 1940s and 1970s (Power and Callaghan, 2015).The relationship between LD SSS and rainfall in the WR catchment could not be expected to hold during these short duration but intense local-scale weather events and remote proxy records in general are usually incapable of resolving events like these.As such, this period of elevated intense storm activity may explain the marked reduction (and change of sign) in the correlation between LD SSS and rainfall in the WR catchment in the early 1950s (Fig. 3f).ECL variability has been related to the IPO, with Speer (2008) finding that during the second IPO positive phase (i.e., 1979-1997) there was a decrease in ECLs relative to IPO negative.This would correspond to a reduction in ECL-related rainfall over New South Wales in the most recent IPO positive phase and is further evidence that these short duration, chaotic events affect the relationship between LD SSS and rainfall in the WR catchment.
Indeed, a better understanding of the role of ECLs and also the relative influence of ENSO, IOD, SAM, STR and other large-scale processes on rainfall in the WR catchment (as is currently being investigated as part of the ESCCI project) will undoubtedly improve our understanding of the variability in the strength of the LD SSS -WR rainfall teleconnection.It should also be noted that one of the key difficulties in understanding the non-stationarity in the climate of the Southern Hemisphere is the lack of quality atmospheric/oceanic data in the Southern Ocean in the pre-1979 satellite era, particularly in the Indian/West Pacific sector.It is likely that more high-resolution ice core records from the Indian Ocean sector of East Antarctica will assist in filling this data gap (Vance et al., 2016).Underpinning the above issue is that variability in the Australian climate record can be up to centennial scale, which cannot be resolved using relatively short instrumental data sets (Gallant et al., 2013).Ultimately, for the purposes of this initial reconstruction, we have assumed stationarity in the LD SSS -Williams River rainfall relationship.

Development of the Williams River rainfall reconstruction
The linear regression coefficients determined for the full 1900-2010 instrumental calibration period (Sect.4) were applied to the 1000-2012 CE LD SSS data to produce 1013 years of rainfall data for each AWAP grid cell in the WR catchment.This grid-cell data were then spatially averaged to produce a WR catchment average rainfall reconstruction time series.

Comparing the catchment-average rainfall reconstruction with instrumental (AWAP) data
A comparison between the AWAP catchment average and reconstructed WR catchment rainfall over the instrumental period (1900-2010) is presented in Fig. 4. The instrumental mean and pattern of peaks and troughs in the recorded rainfall is well represented in the reconstruction but the range of variability is underestimated.While the magnitude of the extremes is important, the key focus is that the reconstruction captures the duration and timing of wet and dry periods.The thinking behind this is that a short, but extreme (in terms of rainfall deficit) drought, for example, will have less severe implications on water security in a catchment than a drought of long duration with consistently below average (but not necessarily extremely below average) rainfall.Encouragingly, periods post-1900 that are known to be as- sociated with droughts and flooding in the WR catchment are identified in the reconstruction, e.g., the World War II drought in the late 1930s, the Millennium drought in the 1990s to 2000s and the flood dominated 1950s and 1960s (e.g., Verdon-Kidd and Kiem, 2009;Gallant et al., 2012;Callaghan and Power, 2014).
The rainfall reconstruction captures around 10 % of the rainfall variability in the WR catchment for the full 1900-2010 instrumental period (Table 1).In terms of IPO phases, it is clear that the reconstruction is in better agreement with the instrumental record for the most recent IPO positive phase (1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997) relative to the first IPO positive phase and the IPO negative phase.This is no surprise given the higher correlation between LD SSS and Williams River rainfall in the recent IPO positive period, i.e., LD SSS variability captures around 40 % of the Williams River rainfall variability (Table 1).Influences on the stationarity of the LD SSS -WR rainfall relationship were discussed in Sect. 4.
Table 2 presents the root mean square error (RMSE) and reduction in error (RE) between the rainfall reconstruction and 12-month average (October-September) rainfall recorded at gauge 61010 and the AWAP WR catchment average for the 1900-2010 period and IPO phases.An RE value greater than zero indicates that the reconstruction is skillful and has better predictive skill than climatology (Cook, 1992).While improved RMSE and RE statistics were recorded for the most recent IPO positive (1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997) relative to the first IPO positive and IPO negative phases, it is clear that the reconstruction has skill across the 1900-2010 instrumental period.For the full instrumental record, the reconstruction has an RMSE of around 25 % of the annual instrumental rainfall with an RE value greater than zero.

Implications for water resources management
While Fig. 5 gives insights into periods of above and below average rainfall, of particular interest for hydrological studies and water resources management is not only whether a year or sequence of years is above or below the long-term average but also whether a multi-year or multi-decadal epoch is gen-erally wet or dry even though some years within that epoch may be slightly below or above the long-term average.For example, a year that is only 0.1 standard deviations above the average is unlikely to provide enough rainfall to break a drought or fill reservoirs.To account for this, we define "wet" and "dry" years as (Eq.1): wet = years where rainfall > mean − x × SD, (1) dry = years where rainfall < mean + x × SD.
For example, for a standard deviation (SD) of 0.1 (and annual average of 1100 mm), the range is 1092.6-1107.4mm.That is, a wet year will be defined as any year with annual rainfall greater than 1092.6 mm and a dry year as any year with annual rainfall less than 1107.4mm.Some years will be defined as both "wet" and "dry" but this methodology avoids a situation where a consistently wet (or dry) period is broken by a single year that is slightly below (or above) the mean.
Table 3 compares the persistence of the longest above and below average rainfall periods (x = 0 in Eq. 1), and "wet/dry" periods (x = 0.1, 0.2, 0.3, 0.4, 0.5 in Eq. 1) in the AWAP catchment average rainfall and the reconstruction.Note that the time periods identified in Table 3 should be considered in light of the ±1 year dating uncertainty of the LD SSS record discussed in Sect.3.1.As shown in Table 3a and b, the reconstruction captures the dry periods, in terms of duration and timing, of the AWAP instrumental record well and also the duration of the longest wet periods.However, the timing of the wettest periods detected by the reconstruction is different to that seen in the AWAP record.As previously discussed this is likely due to the inability of the LD SSS reconstruction to characterize extreme local-scale synoptic activity in the WR region (i.e., ECLs).Importantly, this also implies that the wettest epochs in the reconstruction may be an underestimation, as the reconstruction is least accurate during wet periods caused predominantly by local-scale influences (e.g., ECLs).In other words, wet periods associated with increased ECL activity (e.g., similar to the 1950s) are possible and the magnitude of rainfall associated with these events would be  2).over and above the preinstrumental wet epochs suggested by the LD SSS reconstruction.
Figure 6 shows the duration of above and below average rainfall periods during each century since 1000 CE (and also for the whole 1013-year reconstruction period, panel k).To easily visualize the results, Fig. 6 combines all durations > 15 years (information on all durations is included in Table S1 in the Supplement).Figure 6 clearly shows that (a) some centuries are drier (more pink) than others (more blue) and (b) the most recent complete century (1900-1999, panel j), where the majority of our instrumental record comes from, is not representative of either the duration or frequency of periods of above or below average rainfall experienced pre-1900.
While the results in Fig. 6 are important, of greater interest is the identification of the persistence of wet or dry periods even though some years within the otherwise dry (wet) regime were slightly wetter (drier) than average.Table 3a and b show that for varying x values in Eq. ( 1) the duration of the longest wet or dry periods in the instrumental period does not markedly change.If dry and wet epochs, defined relative to the 1100.0mm instrumental mean are extracted from the preinstrumental reconstruction, a different story emerges (Table 3c).For a standard deviation threshold of 0.3, for example, the results show that the longest dry epochs persist for up to 12 years instead of a maximum of 8 years post-1900, while wet epochs have lasted almost 5 times as long (maximum of 39 years preinstrumental compared to a maximum of 8 years in the instrumental period).A similar result is observed if  3d), with both the dry and wet epochs persisting up to twice as long in the preinstrumental compared to the instrumental period.Figure 7 (and the associated Tables S2 and S3) further illustrates this point (and the points made in relation to Fig. 6) by clearly showing that the proportion, frequency and duration of wet/dry epochs in the instrumental period  is not representative of either the overall situation throughout the last 1000 years or the situation in any century pre-1900.Also of interest is that some centuries tend to have short dry periods compared with long dry periods and vice versa -e.g., the 15th and 16th century (Fig. 7e and f) compared to the 12th and 13th century (Fig. 7b  and c).The same can be said for wet periods.The variation in the distribution of dry/wet-period duration between centuries further suggests that water resources management and planning based on the statistics of 100 years of data (or less) is problematic.

Conclusions
This study produced a 1013-year rainfall reconstruction for the WR catchment, a location without any local paleoclimate proxies.The strength of the relationship between LD SSS and annual WR rainfall was found to vary decadally but, unlike Vance et al. (2013Vance et al. ( , 2015)), was not always coherent with the IPO.This is likely due to the different climate regime that the coastal WR catchment is subject to (e.g., likely more influence from mid-latitude processes) compared to the previous studies, which were located further north and predominantly west of the Great Dividing Range.The WR catchment is also strongly influenced by local-scale coastal storms such as ECLs, which may provide an explanation for the different relationship to the IPO, as well as the breakdown in the East Antarctic-WR teleconnection in periods associated with increased ECL activity (e.g., the 1950s).
Despite the acknowledged non-stationarity in the relationship (which is being further investigated in ongoing research) the relationship between LD SSS and rainfall in the WR catchment is significant over the full 1900-2010 calibration period and indeed the reconstruction shows skill across this period.The reconstruction was found to agree well with identified dry/wet periods in other rainfall reconstructions in the eastern Australia region providing further validation.Ultimately, the LD SSS -based reconstruction shows that the instrumental period (∼ 1900-2010) is not representative of the proportion, frequency or duration of wet/dry epochs in any century in the preinstrumental era.This is consistent with recent independent studies focussed on Tasmania (Allen et al., 2015) and the Murray-Darling Basin (Ho et al., 2015a, b).
These findings provide compelling evidence that existing hydroclimatic risk assessment and associated water resources management, infrastructure design, and catchment planning in the WR catchment is flawed given the reliance on drought and flood statistics derived from post-1900 information.Figure 3 (and Fig. 4a in Vance et al., 2015) suggests that the same is true for most of eastern Australia and indeed may also be the case for other regions in Australia that are identified as (or yet to be identified as) having similar teleconnections with East Antarctica, e.g., southwest Western Australia (van Ommen and Morgan, 2010).Therefore, the robustness of existing flood and drought risk quantification and management in eastern Australia is questionable and the insights from paleoclimate data need to be incorporated into catchment planning and management frameworks, especially given the multi-decadal and centennial hydroclimatic variability demonstrated in this study.

Figure 1 .
Figure 1.Location of Law Dome in relation to Australia with insets indicating the Great Dividing Range, WR catchment boundary and the location of the 61010 high-quality rainfall gauge, Newcastle and Sydney.

Figure 2 .
Figure 2. Climatology of WR catchment rainfall.Shown is the mean and standard deviation of monthly rainfall recorded at the 61010 gauge and for the AWAP catchment average.

Figure 6 .
Figure 6.Histograms of duration of above (blue) and below (pink) average rainfall periods in each century since CE 1000.(a)-(j) are centennial subsets and (k) is the CE 1000-2012 period (note different axis scaling).Above/below average are defined using x = 0 in Eq. (1) (as per Table2).

Figure 7 .
Figure 7. Histograms of duration of WET (blue) and DRY (pink) average periods during each century since CE 1000.(a)-(j) are centennial subsets and (k) is the CE 1000-2012 period (note different axis scaling).WET/DRY are defined using x = 0.3 in Eq. (1) (as per Table2).

C
. R. Tozer et al.: An ice core derived 1013-year catchment-scale annual rainfall reconstruction the full reconstruction mean (1126.1 mm) is used to indicate wet or dry (Table Vance et al. (2015)13)andVance et al. (2015).Figure3findicates the 13-year moving window correlations between LD SSS and October-September for rainfall recorded at gauge 61010 and the AWAP WR catchment average in order to identify low frequency variability associated with the IPO.The Pearson correlation coefficients between LD SSS and October-September annual rainfall recorded at gauge 61010 and the AWAP WR catchment average for the full record and IPO phases are presented in Table used

Table 2 .
Root mean square error in mm year −1 (%) and reduction in error between the rainfall reconstruction and 12-month average (October-September) rainfall recorded at gauge 61010 and the AWAP WR catchment average for the 1900-2010 period and IPO phases.

Table 3 .
Duration of longest dry and wet periods for the AWAP and reconstructed rainfall.