Uncertainty estimation for databased bathymetry using a Bayesian network approach

Abstract

A Bayesian network approach has been developed to estimate uncertainties for gridded bathymetry models that store grids of large regional or worldwide bathymetry. A previous technique to estimate uncertainty for historical data uses Monte Carlo simulations to estimate uncertainty using soundings data, but it is infeasible to implement for gridded bathymetry models. The Bayesian network is designed to estimate uncertainty using gridded bathymetry and a distribution of navigation uncertainties for the survey data used to create the grid as inputs. This paper discusses how the Monte Carlo technique is used to train the network and how to implement it to estimate gridded uncertainty. A demonstration of this approach is presented that uses the data from the original paper for the Monte Carlo technique. Both Monte Carlo and Bayesian network estimators provide higher uncertainty with bottom slope. The Bayesian network estimate is generally greater than the Monte Carlo approach, which is by design for navigation safety, but is within an order of magnitude agreement.