A priori data set screening to improve efficiency of LiDAR processing for shallow water bathymetry International Journal of Remote Sensing

High data volumes and the time required to process LiDAR point clouds to identify bathymetric points create a potentially large lag between data acquisition and use for shallow water mapping. In this study, a method was developed for a priori identification of areas (500 m by 500 m ‘tiles’) in a data set that are unlikely to contain bathymetric pulse returns and therefore do not need to be processed. Using an airborne LiDAR data set centred on Key West, Florida (United States) containing 1374 tiles, a logistic regression model was developed to predict if a tile contained extractable bathymetry (according to standard operating procedures of the United States National Oceanic and Atmospheric Agency (NOAA)) using quantifiable characteristics of depth frequency histograms as predictors. Results indicated that tiles that do not contain extractable bathymetric pulse returns could be identified with 90% accuracy. A post-modelling ‘spatial reassignment’ of individual tiles based on the characteristics of neighbouring tiles provided only a minor accuracy improvement. The methodology was validated on a Miami Beach LiDAR data set containing 120 tiles. Results were comparable to the Key West results, although the logistic regression model had to be re-calibrated for Miami Beach. To operationalize the results and eliminate the need to process all tiles a priori, a progressive tile-sampling approach is suggested. Furthermore, operational use of this a priori tile screening approach also requires consideration of expected uses of bathymetric maps and risk tolerance relative to the different consequences of false negative (FN) and false positive (FP) errors. For the Key West data set comprised of 1374 tiles of which 36% did not contain extractable bathymetry, screening tiles and then processing non-excluded tiles for bathymetric extraction was estimated to reduce total time by 489 hours (161 human/manual hours and 328 computer hours) compared to not screening and processing all 1374 tiles. Grant no. NA20NOS4000196