On the accuracy of digital bathymetric data

The global 5‐arcmin gridded topography data ETOPO‐5 are based on contour maps rather than original soundings and have large errors. The artificial statistical distribution generated by digitizing contours makes these data unsuitable for use in regression models for depth‐age variations. Their amplitude spectrum is bounded by a (frequency) −4 power law, so that they should not be used when the gravity‐topography transfer function is important, such as in studies of flexural isostasy and mantle convection. I assess the accuracy of 14,491,069 digital ship soundings in 2253 cruise surveys collected between 1955 and 1992 in the Lamont‐Doherty Earth Observatory on‐line data base by analyzing 329,058 crossover errors (COEs) at intersecting ship tracks. Five percent of cruises with internal COEs yield root‐mean‐square COE amplitudes exceeding 500 m; all of these have errors in digitizing two‐way travel time from analog precision depth recorder traces. Twenty‐eight cruises were found which had errors caused by misinterpretation of the nominal sound velocity used when travel times were reported as nominal depths. Two nominal sound velocities in common use differ by 2.5%, an amount which is often undetectable, producing uncertainties in depth of this magnitude. Ship data have been acquired at different rates over time, with the peak of activity in the early 1970s. Although present technologies can yield very accurate data, these are acquired at a rate which is small with respect to the total available data; the cumulative median global COE has remained constant at 26 m since the late 1970s. Most recent data acquisition has been in the northern hemisphere oceans, and the oldest and least accurate data are in the southern oceans where median COEs are 100–250 m. The majority of the data in the South Pacific were acquired before the advent of satellite navigation.