Multibeam Sponsors

Welcomes you to the Icebreaker Party

Computer-aided quality assurance of high-resolution digitized historic tide-gauge records

Hein, Barjenbruch, Blasi, Mai


There is no doubt that sea-level data of high quality are needed for different purposes from daily practical routines like navigation, operational modelling for storm-surge warnings, fundamental statistics and research on climate change. At the recent meeting of the GLOSS Group of Experts (7-11 November, 2011,Paris) the rescue of tide gauge data which currently stored in non-computer forms (charts, tabulations, etc.) was addressed (Circular Letter, IHB File No S3/2705). Our study reports the difficulties connected with the digitalization of tide gauge data in paper form. The crucial challenge is situated in the quality control of the data. Generally, these data are so extensive, that automatic methods, so called Computer-aided quality assurance (CAQ) must be used to identify failed digitization, data gaps or distortions of water levels.

The analogues sea-level records include many ambiguities and errors in the time series, which may disturb the automatic data processing. We analyse the suitability of methods to detect this uncertainties. Different statistical methods like spectral analysis or fuzzy-logic show good results in detecting outliers and filling gaps. Gaps in time series are a serious problem, and every method that is used to fill them can give only estimations. For the time being, the fuzzy-logic methods investigated to find a suitable solution.

Additionally, the Lomp-Scargle-Power-Spectrum (LSPS) calculate the power-spectral-density function across all gaps, which give an in-depth understanding of the underlying physical processes. By converting the important frequencies back to the time domain a straight forward equidistant coastal-hydrological time series will be the result. The next step of the CAQ is to find high-water and low-water data in magnitude and time – automatically, which is not an easy task. Presently, several methods are being investigated to find the best solution. The last step is to find breakpoints in the time series, which allows homogenisation and trend detection.