[Geophysical Research Letters] Supporting Information for [Considerations for Estimating the 20th Century Trend in Global Mean Sea Level] B.D. Hamlington1 and P.R. Thompson2 [1] {Center for Coastal Physical Oceanography, Old Dominion University, Norfolk, USA.} [2] {Department of Oceanography, University of Hawai‘i at Mānoa, Honolulu, Hawaii, USA.} Contents of this file Text S1 Introduction The file text01.pdf contains a description of the methods used for selecting, editing, and analyzing the tide gauge data. Text S1. 1. Methods We analyze the tide gauge datasets used in three recent sea level reconstructions: Ray and Douglas [2011; RD2011 hereafter], Church and White et al. [2011; CW2011 hereafter], and Hay et al. [2015; H2015 hereafter]. All three studies begin with tide gauge data from the Permanent Service for Mean Sea Level (PSMSL) Revised Local Reference 1 (RLR) dataset, but each implements very different gauge selection choices and quality control criteria when forming the historical tide gauge dataset used in their reconstructions. We begin with monthly RLR data spanning 1900-2013 and form three datasets by subsetting the RLR dataset based on the gauges used in each reconstruction. This information was obtained online for CW2011 and via personal communication for RD2011 and H2015. To isolate the effect of tide gauge selection, we apply a consistent automated quality control procedure to the data from each set and then use the same simple method to calculate GMSL for each case. The quality control algorithm is consistent with Church and White et al [2004] and Calafat et al. [2014]: (1) All monthly values flagged by PSMSL have been removed; (2) Gaps of one to two months were linearly interpolated; (3) Continuous sections of data shorter than two years were removed; (4) Records were split when a 250 mm month-tomonth jump occurs if that record has fewer than ten such jumps. The ICE-5G v1.3 glacial isostatic adjustment (GIA) [Peltier et al., 2004] was applied to all of the tide gauge records. No other correction for vertical land motion was made. GMSL for each of the three sets is calculated via a simple arithmetic mean over the tide gauge data in each set. We are primarily interested in the effect of gauge selection on 20th century trends and this method has been shown to be adequate when computing global long-term trends. In addition, this method is linear and transparent, which allows us to diagnose the effect of particular selection choices on the estimated trends. Due to the lack of a consistent datum for all of the tide gauge records, the arithmetic mean is applied to the first differences of monthly tide gauge data, which is then reintegrated to form the GMSL time series. 2 We estimate trends from these time series over the 1900-2013 period using a linear least squares fit, and confidence intervals (95%) are obtained using a standard error estimate [Acton, 1966]. The effective degrees of freedom have been estimated based on the decorrelation timescale of the GMSL time series. We avoid a more thorough treatment of the error, as it does not affect the conclusions of the manuscript, but note some differences in our error estimate in comparison to the three reconstruction studies. The errors provided for the reconstructions are intended to represent errors in the accuracy of the reconstructions relative to the true trend in GMSL. In contrast, we are concerned with the precision (not accuracy) of the trend given differences in gauge selection. Thus, the standard error estimate based solely on the least squares fit serves the purpose of this paper. A more realistic estimate of the error would involve accounting for spatial correlations among the tide gauges, leading to a further reduction in the degrees of freedom. This, however, is left for future work. 3