Comment on ‘Drought variability in the eastern Australia and New Zealand summer drought atlas (ANZDA, CE 1500–2012) modulated by the Interdecadal Pacific Oscillation’

The study of (Palmer et al Environ. Res. Lett. ) details a spatial reconstruction of drought across eastern Australia and New Zealand over the last 500 years. The authors used a global 0.5° by 0.5° gridded network of the self-calibrating Palmer drought severity index (scPDSI) spanning 1901–2012 as the basis for a nested point-by-point regression to reconstruct austral summer (DJF) scPDSI for this region. Their study used 176 tree rings from New Zealand, Indonesia and Australia, and one coral record from the Great Barrier Reef. In their paper Palmer et al () compared three publically available proxy records and reconstructions derived from the Law Dome ice core (East Antarctica) to their reconstructed scPDSI. These were the LD summer sea salt (LDsss) series, which is a proxy for Western Pacific sea surface temperature and subtropical eastern Australian rainfall (Vance et al J. Clim. , Geophys. Res. Lett. , Tozer et al Hydrol. Earth Syst. Sci. ), and two Interdecadal Pacific Oscillation (IPO) reconstructions produced using two independent methods, namely the Piece-wise Linear Fit (PLF) and Decision Tree (DT) series (Vance et al Geophys. Res. Lett. , Clim. Past ). We show that the treatment of the Law Dome LDsss record and the PLF and DT IPO reconstructions mis-characterizes both the utility and targets of the three records.


Introduction
We argue that the analysis of Palmer et al (2015) mischaracterizes the skill of our Interdecadal Pacific Oscillation (IPO) reconstructions. It does this by correlating austral summer (dJF and dJFM) and annual (May-June) instrumental indices of the IPO (Power et al 1999 andHenley et al 2015) with our 13 years smoothed PLF and DT-median reconstructions (see table 1(b) in Palmer et al 2015). Clearly, the different temporal resolutions of the correlated series mean our 13 years smoothed reconstructions will be unable to capture variance in seasonal or annual indices; correspondingly, the squared Pearson correlation coefficient (RSQ) values presented are low, leading to the conclusion that the ANZDA EOF1 is more strongly related to the IPO than the Vance et al (2015) PLF and DT reconstructions. Unfortunately, readers of the Palmer et al (2015) paper are likely to be unaware that the RSQ values have been calculated using smoothed reconstructions against seasonal/ annual instrumental data. Our IPO reconstructions were developed with the decadal-scale band as the target and are in actual fact highly skillful in the decadal scale band that defines the IPO (Vance et al 2015).
Additionally, Palmer et al (2015) compute correlations between the LDsss series and indices that do not particularly make sense-for example, against IPO indices. We have not used or published the LDsss series as an IPO proxy, and suggest that the resulting low RSQ values presented in Palmer et al (2015) are unsurprising and have no bearing on the utility of the LDsss series as a Western Pacific sea surface temperature (SST)/eastern Australian rainfall proxy. Our published work about the LDsss series details significant relationships with SST in the Western Pacific and with eastern subtropical Australian rainfall (Cai et al 2010, Vance et al 2013, Tozer et al 2016, not with the IPO. Western Pacific SSTs are particularly associated with rainfall variability in eastern Australia during winter/spring (not summer) (www. bom.gov.au/climate/updates/articles/a008-el-ninoand-australia.shtml, Gallant et al 2012). The relationship between the Western Pacific and Law Dome also occurs during austral winter/spring, as would be expected with a developing ENSO-related wavetrain in the SW Pacific, which is then transmitted to higher latitudes by austral summer (Karoly 1989, Mo andPaegle 2001) and appears as a signal in the summer sea salt concentrations (LDsss series) in snowfall at Law Dome. This stronger relationship between LDsss and the Western Pacific and Australian rainfall should have been mentioned, rather than just presenting a low RSQ value with Niño 3.4 SST from outside of the SST season described as being important for mechanistic reasons in Vance et al (2013).
To illustrate these issues, we provide two tables.

Methods
IPO Indices-HADISST monthly data was obtained directly from Chris Folland (UK Met Office) in 2014. This data was used to calculate dJF/dJFM and annual (May-June) records. TPI IPO (HADISST) monthly data obtained from Ben Henley, as part of Henley et al (2015), and also calculated to seasonal and annual as above. Niño 3.4 seasonal (dJF and dJFM) produced from monthly data downloaded from KNMI Climate Explorer.
Calculation of correlation values with bootstrap confidence intervals (CIs)-the method of Olafsdottir and Mudelsee (2014) was used for the calculation of correlation values and associated bootstrap CIs. This method accounts for autocorrelation in the time series when computing correlations. The correlation result is significant at 95% if the 95% CI does not span zero.