Statistical selection of tide gauges for Arctic sea-level reconstruction

• Published in: Advances in Space Research
• Main Authors: Svendsen, Peter Limkilde, Andersen, Ole Baltazar, Nielsen, Allan Aasbjerg
• Format: Article in Journal/Newspaper
• Language: English
• Published: 2015
• Subjects: Empirical orthogonal functions; Leverage; Principal component analysis; Function evaluation; Gages; Oceanography; Orthogonal functions; Quality control; Sea level; Tides; Empirical Orthogonal Function; In-situ observations; Ocean circulation models; Sea level variations; Sea-level reconstruction; Statistical properties; Statistical selection; Tide gages; Arctic; Arctic Ocean
• Online Access: https://orbit.dtu.dk/en/publications/edcb816e-7dfb-4103-ba56-42edcad89ca6; https://doi.org/10.1016/j.asr.2015.01.017

In this paper, we seek an appropriate selection of tide gauges for Arctic Ocean sea-level reconstruction based on a combination of empirical criteria and statistical properties (leverages). Tide gauges provide the only in situ observations of sea level prior to the altimetry era. However, tide gauges are sparse, of questionable quality, and occasionally contradictory in their sea-level estimates. Therefore, it is essential to select the gauges very carefully. In this study, we have established a reconstruction based on empirical orthogonal functions (EOFs) of sea-level variations for the period 1950-2010 for the Arctic Ocean, constrained by tide gauge records, using the basic approach of Church et al. (2004). A major challenge is the sparsity of both satellite and tide gauge data beyond what can be covered with interpolation, necessitating a time-variable selection of tide gauges and the use of an ocean circulation model to provide gridded time series of sea level. As a surrogate for satellite altimetry, we have used the Drakkar ocean model to yield the EOFs. We initially evaluate the tide gauges through empirical criteria to reject obvious outlier gauges. Subsequently, we evaluate the "influence" of each Arctic tide gauge on the EOF-based reconstruction through the use of statistical leverage and use this as an indication in selecting appropriate tide gauges, in order to procedurally identify poor-quality data while still including as much data as possible. To accommodate sparse or contradictory tide gauge data, careful preprocessing and regularization of the reconstruction model are found to make a substantial difference to the quality of the reconstruction and the ability to select appropriate tide gauges for a reliable reconstruction. This is an especially important consideration for the Arctic, given the limited amount of data available. Thus, such a tide gauge selection study can be considered a precondition for further studies of Arctic sea-level reconstruction.