The advent of topobathy lidar has dramatically increased the data collected in a lift, with obvious consequences for data timeliness. For similar reasons, acoustic bathymetry processing has transitioned from selection/classification of individual observations to computer-assisted depth estimation, using the observation uncertainties to weight the contribution to depth, and perform unsupervised classification into self-consistent observation clusters.

We adapt here an acoustic tool to topobathy data. From a data-adaptive resolution determination, this method estimates a variable-resolution DTM from raw data, where each node can have multiple, simultaneously valid, estimates of depth. Conventional depth disambiguation is ineffective with the estimated 75% outlier rate; a new vector-quantized hidden Markov model technique is therefore proposed. This machine learning method develops a clean depth reconstruction, while allowing for a “no reconstruction” solution in areas with no real depth observations.

This offers efficient, automatic processing, and the opportunity to align bathymetric processing methods, with obvious benefit for training, output compatibility, and workflow management.