A GENERALIZED LINEAR MODEL APPROACH FOR BEACH CHARACTERIZATION WITH MULTITEMPORAL ALSM MEASUREMENTS

Airborne laser swath mapping (ALSM), also known as airborne scanning lidar, provides a rich data source of topographic information. When acquired with sufficient temporal coverage, the high spatial resolution information in the lidar data can reveal patterns in beach change otherwise unforeseen. In this regard, data mining and pattern classification techniques offer great potential to progress the analysis of lidar data for beach monitoring. In this research, we develop a probabilistic-based classifier to examine the influence of varying beach morphology on shoreline change patterns observed in multitemporal lidar data acquired over a Florida beach. Through cross-shore profile sampling of lidar-derived digital elevation models (DEMs), the continuous 3D beach surface is parameterized into several 1D morphologic features progressing alongshore. Profiles are subsequently partitioned into binary erosion or accretion tending classes dependent on measured shoreline change between surveys. A generalized linear model, logistic regression, is then used to model the influence of morphology (features) on probability of shoreline response (erosion or accretion) alongshore for each survey epoch. To handle spatial correlation in the binary responses, a generalized estimating equation approach is utilized. Lasso penalization is then applied to reduce collinearity and automatically eliminate non-important features. Those morphologies more influential to the shoreline response patterns are systematically detected, and the resultant models can be used to characterize beach terrain configurations more conducive to erosion.