Date of Award


Document Type

Campus Access Dissertation



First Advisor

Jensen, John R


The research evaluated the capability of color-infrared (CIR) aerial photography integrated with high-posting-density LiDAR (Light Detection And Ranging) data for intertidal vegetation classification in the upper May River watershed near Bluffton, South Carolina. An object-based image analysis (OBIA) approach was adopted for mapping the following detailed tidal marsh classes: tall creekbank spartina alterniflora, intermediate spartina alterniflora, exposed mudflats, salicornia/short spartina and juncus roemerianus. A high-resolution Digital Terrain Model (DTM) was first developed from LiDAR bare earth data, based on which topographic features including slope and aspect were extracted. All the LiDAR-derived altimetry and intensity data were then combined with CIR spectral bands and input to the OBIA. A Segmentation experiment was designed with 200 different combinations of parameters using the multi-resolution image segmentation algorithm. Optimal parameter setting was analyzed based on a combination of unsupervised and supervised segmentation evaluation. Both unsupervised goodness measures and supervised metrics indicated that the scale factor of 40 was the approximate balance point between over- and under-segmentation, which was consistent with the visual inspection results. An empirical optimization experiment was also designed with four feature input options and two tree-based ensemble classification algorithms (i.e. Random Forest and Adaboost tree). Optimal scales were analyzed for each scheme and classification results from all eight schemes were compared. Interestingly, classification results from all eight schemes did not exhibit significant difference with segmentation scales <= 40. A feature importance analysis indicated that both Random Forest and Adaboost tree considered LiDAR altimetry and intensity information as important variables, although Adaboost gave them higher variable importance ranks. This research contributes to better understanding of the capability of integrated active and passive airborne remote sensing for improving the classification of tidal marsh vegetation in complex coastal ecosystems in South Carolina.