Interferometric synthetic aperture sonar (InSAS) produce bathymetric maps with large amounts of data. This introduces challenges in how the data should be stored and represented. The bathymetric maps are in raster format that can be converted to triangular irregular networks (TINs). A TIN representation of the map can reduce the amount of storage needed. However, it is not straight forward how the conversion should be performed while preserving the main characteristics of the mapped surfaces. We consider two decimation algorithms and one refinement algorithm that performs a raster to TIN conversion. They are point selection algorithms and based on Delaunay triangulations. We have implemented the algorithms and adapted them such that they also base their point selection on the coherence from an InSAS. The algorithms were applied on three classes of data: synthetic data without coherence, synthetic data with coherence, and real bathymetric data from an InSAS. The algorithms returned reduced data sets that contained less than ten percent of the input data and at the same time preserved essential features of their inputs. We found that the refinement algorithm performed the best to convert a raster to TIN in overall.