Automated cleaning and uncertainty attribution of archival bathymetry based on a priori knowledge

Hydrographic offices hold large valuable historic bathymetric data sets, many of which were collected using older generation survey systems that contain little or no metadata and/or uncertainty estimates. These bathymetric data sets generally contain large outlier (errant) data points to clean, yet standard practice does not include rigorous automated procedures for systematic cleaning of these historical data sets and their subsequent conversion into reusable data formats. In this paper, we propose an automated method for this task. We utilize statistically diverse threshold tests, including a robust least trimmed squared method, to clean the data. We use LOESS weighted regression residuals together with a Student-t distribution to attribute uncertainty for each retained sounding; the resulting uncertainty values compare favorably with native estimates of uncertainty from co-located data sets which we use to estimate a point-wise goodness-of-fit measure. Storing a cleansed validated data set augmented with uncertainty in a re-usable format provides the details of this analysis for subsequent users. Our test results indicate that the method significantly improves the quality of the data set while concurrently providing confidence interval estimates and point-wise goodness-of-fit estimates as referenced to current hydrographic practices.