coefficient: NA

rank: NA

scope: climate

fraction: NA

rank: NA

scope: climate

direction: decrease

interpDirection: decrease

scope: climate

seasonality: subannual

variable: temperature

variableDetail: sea surface

variableGroup: temperature and salinity

scope: climate

scope: climate

basis: We develop and test a transfer function constructed using linear regression analysis of SBV coral d18Oanomaly variations and instrumental SSS over the period 1970?2007 CE [Thirumalai et al., 2011]. We performed a calibration-verification exercise (Figure 6) between d18Oanomaly and SSS at SBV to assess the robustness of the transfer function [Quinn and Sampson, 2002]. The slopes in equations (1)?(3) are within error of each other; however, the intercepts are slightly different. Despite this small difference, the similarity of the lines (Figure S2) provides confidence that the transfer function developed in this study can be used to reconstruct past changes in salinity at this locality based on coral d18O variations. We also note that there is a signal in the residuals, which likely reflects a small temperature component in the coral d18O anomaly signal (Figure 6, bottom), most likely due to the small changes in seasonality over this time period (Figure 4c).

coefficient: 1.973

direction: positive

fraction: NA

inferredMaterial: seawater

integrationTime: 1

integrationTimeBasis: Geochemical variations versus depth were converted to variations versus time using AnalySeries software [Paillard et al., 1996]. Geochemical variations in d18O were used to determine a first-order age model, with the maximum intra-annual peaks in d18O being assigned as the coldest month of the year (August), beginning with the year 2007, when the living coral was cored. This first-order age model places d18O variations in the time domain, with uneven time increments (Dt). A second-order age model, with a monthly Dt, was created using the AnalySeries software program, which was verified by comparing years with anomalous d18O values to known ENSO events where possible. This second-order age model is the final age model used in all plots and data analysis.

integrationTimeUncertainty: ~1?2 months in any given year, no errors given for annual chronology

integrationTimeUncertaintyType: chronological

integrationTimeUnits: month

mathematicalRelation: linear

rank: 1

scope: isotope

seasonality: subannual

variable: seawaterIsotope

variableGroup: EffectiveMoisture

variableGroupDirection: negative

variableGroupOriginal: d18O_seawater

basis: We found that the correlations with d18O are 0.72 (with total pseudocoral), 0.47 (with SST component), and 0.68 (with SSS component), p < 0.01 for all three (Figure S1), indicating that SSS changes represent a larger fraction of the variance in the coral d18Oanomaly signal than SST changes, as expected from the larger magnitude of interannual SSS variations at this site. Testing several different percent contributions of SSS and SST to create the pseudocoral, we determined that a combination of 35% SST and 65% SSS results in the closest representation to the observed coral d18Oanomaly values. This was calculated by creating a pseudocoral that consisted of percent SST/SSS contributions that ranged from 100/0% to 0/100%, in increments of 5% (i.e., 100/0, 95/5, 90/10a€¦ 5/95, 0/100). The pseudocoral consisting of 35/65% gave the highest correlation with the measured d18O time series, which provides the percent contributions of SST and SSS to the time series. This relationship was determined over the period 1970a€“2007, and is limited by the length of the instrumental SSS data set. We assume stationarity in the proportional contributions of SST and SSS to the coral d18O signal because SSS data needed to evaluate this assumption are lacking in the pre- 1970 period. However, the assumption of stationarity of a proxy-instrumental relationship developed over the instrumental time period affects all proxy-based climate reconstructions that extend beyond the instrumental period. Thus, lacking additional instrument data and/or another independent SST- or SSS-only proxy there is no easy way to reduce the uncertainty of the empirically derived proxy relationship over the calibration-verification interval.

coefficient: NA

direction: negative

fraction: 0.22

inferredMaterial: seawater

integrationTime: 1

integrationTimeBasis: Geochemical variations versus depth were converted to variations versus time using AnalySeries software [Paillard et al., 1996]. Geochemical variations in d18O were used to determine a first-order age model, with the maximum intra-annual peaks in d18O being assigned as the coldest month of the year (August), beginning with the year 2007, when the living coral was cored. This first-order age model places d18O variations in the time domain, with uneven time increments (Dt). A second-order age model, with a monthly Dt, was created using the AnalySeries software program, which was verified by comparing years with anomalous d18O values to known ENSO events where possible. This second-order age model is the final age model used in all plots and data analysis.

integrationTimeUncertainty: ~1?2 months in any given year, no errors given for annual chronology

integrationTimeUncertaintyType: chronological

integrationTimeUnits: month

mathematicalRelation: linear

rank: 2

scope: isotope

variable: temperature

variableGroup: Temperature

variableGroupDirection: negative

variableGroupOriginal: T_water

coefficient: NA

fraction: NA

rank: NA

scope: isotope

coefficient: NA

rank: NA

scope: climate

fraction: NA

rank: NA

scope: climate

scope: climate

coefficient: NA

fraction: NA

inferredMaterial: seawater

rank: NA

scope: isotope

coefficient: NA

fraction: NA

rank: NA

scope: isotope

coefficient: NA

fraction: NA

rank: NA

scope: isotope