Probabilistic Detection of Morphologic Indicators for Beach Segmentation With Multitemporal LiDAR Measurements

Airborne light detection and ranging (LiDAR) surveys provide a rich data source of topographic information. However, beach monitoring with LiDAR data has been mostly limited to visualization and first-order measures derived from digital elevation models (DEMs). To exploit more information from multitemporal LiDAR data acquired over a beach, we extract surface features to detect morphologies that are indicative of shoreline change patterns. First, through cross-shore profile sampling of LiDAR-derived DEMs, the continuous 3-D beach surface is parameterized into several 1-D morphologic features progressing alongshore. Profiles are subsequently partitioned into binary erosion or accretion classes dependent on measured shoreline change between surveys. Then, a feature's class separability is quantified using information divergence measures nonparametrically constructed via Parzen windowing. The more interclass separation provided by a feature, the greater its discriminative potential, and the higher its ranking as a morphologic indicator. Rankings are computed across the survey epochs to evaluate performance stability of the features. Finally, the top-ranked features are implemented within a naive Bayes classifier to assess their ability to segment terrain more likely to erode. Results demonstrate the utility of the developed framework to systematically extract and incorporate useful morphologic information from LiDAR data for discerning patterns in beach change.