Pacific decadal variability over the last 2000 years and implications for climatic risk

Pacific decadal variability (PDV), commonly quantified by the Interdecadal Pacific Oscillation (IPO), influences drought, flood, wildfire, tropical cyclone risk, and temperature across the Pacific Basin and Antarctica. However, high-latitude observational records are short and may poorly characterize long-term variability. It is often assumed that IPO positive and negative phases are symmetric and last 20–30 years, yet the mechanisms (internal ocean-atmosphere processes vs. external forcing such as volcanic or anthropogenic aerosols) remain debated. To better understand PDV behaviour, phase lengths, prevalence, and associated climatic risks over multi-centennial to millennial timescales, the authors extend an existing Law Dome (East Antarctica) ice-core-based IPO reconstruction to cover the Common Era at annual resolution and compare it with an independent pan-Pacific tree-ring IPO reconstruction. A case study evaluates hydroclimatic implications for eastern Australia, where PDV strongly modulates rainfall and water security.

Prior work has linked PDV/IPO to regional hydroclimate impacts in Australia and beyond, and to recent Antarctic climate changes, but inference is limited by the short observational record. The relative roles of internal variability, stochastic atmospheric forcing, volcanic forcing, and anthropogenic aerosols in driving decadal variability (including IPO and AMO) are contested. Recent modelling and observational analyses (e.g., Mann et al.) suggest volcanic and anthropogenic aerosol forcing can pace multidecadal variability and may be necessary to reproduce such modes in simulations, raising questions about the degree of internal oscillatory behaviour. Meta-analyses of reconstructions indicate observed IPO phase lengths resemble those of the last few centuries, but a scarcity of multi-centennial reconstructions has limited assessment prior to ~500 years ago. Disagreement among reconstructions is common, complicating understanding of past PDV. The study builds on these findings using expanded Antarctic ice-core data and compares with a pan-Pacific tree-ring IPO reconstruction (Buckley et al., 2019).

Data sources and study design: The IPO is reconstructed using annually resolved proxies from Law Dome (East Antarctica) ice cores, extending a previous reconstruction to span CE −10 to 2017 with annual resolution. The reconstruction targets an annually averaged (July–June) Parker et al. IPO index smoothed with a 13-year (half power) Gaussian filter (sigma = 3.9) to match observational treatment. Ice core dating and measurements: The Dome Summit South (DSS) ice core composite incorporates DSS1617 (1990–2016), DSS97 (1888–1989), DSS99 (1841–1887), and DSS Main (to 1841 and deeper). Annual layer counting across multiple seasonal proxies (δ18O, δD, H2O2 maxima timing, winter sea salts Cl−, Na+, Mg2+, summer non-sea-salt SO4^2−, and SO4^2−/Cl−) provides a 2026-year layer-counted chronology (CE −10 to 2017). Reference volcanic horizons (e.g., Pinatubo 1992, Agung 1963, Tambora 1815, Samalas 1257, Unknown 422 CE) indicate dating errors that increase with depth but remain constrained (e.g., +6/−11 years between 422 CE and Samalas). Sampling resolution is increased with depth to counter layer thinning (chemistry: 5 cm near surface to 2.5–3 cm at depth; isotopes to 1 cm), maintaining mean seasonal sample density (≈8–21 samples yr−1 depending on period). Proxy inputs: Input time series include log-transformed warm (Dec–May) and cool (Jun–Nov) season sea salt (from chloride; sodium substituted via seawater ratios when needed) and annual snowfall accumulation rate (m ice equivalent) corrected for site characteristics. Years with insufficient seasonal data (>2 of 6 months missing/compromised) are excluded from seasonal means. None of the inputs display significant trends. Pre-processing and reconstruction methods: Warm and cool seasonal sea-salt series are combined using a 13-year sliding correlation window to maximize reconstruction skill. A Gaussian low-pass filter (sigma = 3.9; ~13-year FWHM) is applied. Two reconstruction approaches are used: (1) Piece-wise Linear Fit (PLF; multivariate adaptive regression splines) with basis functions chosen by minimizing RMSE and GCV on the calibration period (9 initial basis functions, pruned to 8); and (2) a Gaussian kernel correlation multiproxy reconstruction robust to missing/uneven sampling, generating an ensemble of 2000 reconstructions and producing mean and quartile bounds annually. Both methods agree on timing of major negative IPO events. Spectral and coherence analyses use the MTM toolkit. Observational Parker IPO index (monthly, 1870–2020) is processed to annual July–June means and smoothed identically (13-year Gaussian). Rainfall data for Australian domains and long stations are sourced from the Bureau of Meteorology High Quality network and regional/state averages. Rainfall distributions by IPO phase are compared using Welch’s t-tests. Phase definitions and analyses: IPO phases are defined as positive (>0.5), neutral (−0.5 to 0.5), and negative (<−0.5). The study also introduces and analyzes a ‘neutral-positive’ category (index > −0.5) to reflect the predominant state in the reconstructions. Phase lengths, frequencies (percent time), and distributions are computed for observations and reconstructions over multiple intervals (e.g., CE 1–2011; CE 1351–2004).

- A 2000-year Law Dome ice-core-based IPO reconstruction shows that IPO negative phases are short (7 ± 5 years) and infrequent (≈10% of the time), representing departures from a predominantly neutral-positive state that persists for decades (61 ± 56 years). Observational-era assumptions of roughly symmetric 20–30-year positive and negative phases are not supported over the Common Era. - Comparison with an independent 654-year tree-ring IPO reconstruction (Buckley et al., 2019) reveals limited coherence at the 95% level and modest correlation once autocorrelation is accounted for, yet both show predominantly neutral-positive conditions and lack strong interdecadal/multidecadal periodicity. - Negative phases occur far less frequently than positive phases in the reconstructions (≈10–14% vs. ≈29–51% of time, depending on series/period) and are shorter than suggested by observations (reconstructed 5–7 years vs. observed mean 19.5 years for the two observed negative phases). Negative-phase prevalence is higher in the last ~661 years than in the prior ~1350 years. - Observations (1877–2014) include four positive phases (mean ≈14.3 years) and two negative phases (mean ≈19.5 years), but the short record poorly represents Common Era PDV and over-represents negative phases. - Hydroclimatic impacts in eastern Australia: Rainfall distributions show positive vs. neutral years are statistically indistinguishable, while negative years tend to be significantly wetter than neutral and positive years. Consequently, grouping neutral and positive into ‘neutral-positive’ is justified for risk assessment. - Rainfall reductions from negative to positive IPO years across eastern Australia domains are 6–14%; from negative to neutral-positive years 7–12%. Examples (1900–2014): Eastern Australia 65 mm/y (10%) negative→positive and 55 mm/y (9%) negative→neutral-positive (means: all years 609 mm/y, negative 647, neutral-positive 592); Murray–Darling Basin 47 mm/y (9%) for both contrasts (means: 472, 503, 456); New South Wales 69 mm/y (12%) and 67 mm/y (12%) (means: 522, 566, 499); Queensland wet season 80 mm/y (14%) and 67 mm/y (12%) (means: 529, 576, 509); Victoria 40 mm/y (6%) and 45 mm/y (7%) (means: 645, 674, 629). - Due to nonlinear rainfall–runoff relationships, a 2–4% reduction in annual rainfall (all years vs. neutral-positive years) implies roughly 6–16% lower annual runoff/reservoir inflows, posing significant water security risks if neutral-positive conditions dominate. - The mid-20th-century observed negative IPO phase was unusually long relative to Common Era reconstructions, further indicating that the observational period is atypical.

The study addresses the core question of long-term PDV/IPO behaviour by providing a millennial-scale, annually resolved reconstruction demonstrating that the Pacific climate system predominantly resides in a neutral-positive state with only occasional, short-lived negative excursions. This reframes PDV-driven climate risk: planning based on observational-era phase symmetry likely overestimates the frequency and duration of negative phases. For eastern Australia, where PDV strongly modulates hydroclimate, neutral-positive conditions align with increased drought risk, lower mean rainfall, and significantly reduced runoff relative to negative phases, informing water security and infrastructure planning. The findings also intersect with debates on PDV drivers by showing muted periodicity in reconstructions and altered negative-phase prevalence over the last ~661 years, consistent with hypotheses that external forcing (volcanic, anthropogenic aerosols) may pace decadal variability. The proposed ‘neutral-positive’ threshold offers a practical framework for risk analyses in regions influenced by the IPO. The study highlights the need to understand the initiation and persistence of negative phases, especially given their apparent greater prevalence in more recent centuries and potential future anthropogenic influences.

Common Era reconstructions indicate that PDV, as expressed by the IPO, is characterized by a predominantly neutral-positive state with infrequent, short negative phases, contradicting the commonly assumed symmetry of 20–30-year phases. The observational period—especially the mid-20th-century negative phase—appears atypical relative to the past two millennia. This has quantifiable implications for climate risk management across the Pacific Basin: risk assessments should emphasize the neutral-positive state and its associated hydroclimatic outcomes (e.g., elevated drought risk in eastern Australia). Two future scenarios are posited: (1) PDV reverts to its long-term neutral-positive dominance, rendering risk analyses based on mid-20th-century observations questionable; or (2) the mid-20th-century negative anomaly signals a transition to more frequent/prolonged negative phases, implying a different, but equally important, recalibration of risk. Future work should prioritize identifying mechanisms that initiate and sustain negative IPO phases and assess contributions from anthropogenic aerosols and greenhouse forcing, using long (>500-year) reconstructions to quantify phase statistics and their changes through time.

- Observational IPO indices span too few phase changes to robustly determine phase statistics or underlying mechanisms, limiting direct comparison and calibration confidence. - Disagreement between independent PDV reconstructions (e.g., phase timing and sign mismatches, limited coherence) is common, reducing certainty in specific phase histories even as broader prevalence patterns align. - Autocorrelation reduces effective degrees of freedom in correlation analyses, complicating statistical inference. - Ice-core dating uncertainty increases with depth (albeit constrained by volcanic markers); potential errors of several years in older sections could affect precise phase boundaries and lengths. - Despite efforts to maintain sampling resolution and use methods robust to missing or unevenly sampled data, residual inhomogeneities or proxy-specific noise may remain. - The reconstruction exhibits a small, statistically significant negative trend; authors argue it reflects more frequent/longer negative phases in the last 500 years rather than proxy drift, but this interpretation carries uncertainty. - Hydroclimatic case-study in eastern Australia relies on observational rainfall datasets and phase classifications using smoothed indices; regional variability and nonstationarity may affect transferability.