Automatic feature extraction from airborne lidar measurements to identify cross-shore morphologies indicative of beach erosion

• Main Authors: Starek, M.J., Vemula, R.K., Slatton, K.C., Shrestha, R.L., Carter, W.E.
• Format: Conference Proceeding
• Language: English
• Subjects: change detection; Data engineering; Data mining; data segmentation; Entropy; Feature extraction; Laser radar; lidar; Monitoring; Morphology; Optical pulses; Sampling methods; Sea measurements; shoreline erosion
• ISSN: 2153-6996; 2153-7003
• DOI: 10.1109/IGARSS.2007.4423354

Airborne lidar data were acquired along St. Augustine Beach, Florida six times between August 2003 and June 2006. To identify sub-aerial morphologies indicative to beach erosion, the data sets were mined extensively by extracting several morphological features using cross-shore profile sampling. For each profile, the features were grouped into erosion or accretion classes and their class-conditional probability density functions (PDFs) estimated via Parzen windowing. PDF separability was ranked using symmetric and normalized measures of relative entropy (i.e. divergence). Results were compared to a simple median metric. The more interclass separation provided by a feature, the greater its potential as an indicator for erosion or accretion. Over short time periods (>1 month), beach slope and beach width ranked highest by providing the most separation and therefore high potential as indicators for erosion. Over longer time periods (>1 year), deviation-from-trend, which is the shoreline's deviation from the natural strike of the beach, ranked highest. This is significant in that the pier region's deviation from the natural trend is believed by coastal researchers to be a strong contributing factor to it being an erosion "hot spot". The method we have developed provides a systematic framework to mine high-resolution airborne lidar data over beaches, detect erosion-prone areas, and numerically rank a feature's potential as an indicator for erosion.