Juan Carlos A. Graciosa, Renante R. Violanda, Annie G. Diola, Wenifel P. Porpetcho, Danilo T. Dy, Roland Emerito S. Otadoy
Mangroves are salt-tolerant plants located at the fringes of tropical and subtropical zones. They play key roles in mitigation of natural hazards such as shoreline erosion reduction and protection from storm surges and tsunamis. In addition, mangrove ecosystems are one of the most diverse and most productive ecosystems on earth. In light of the increasing frequency and intensity of tropical cyclones hitting the Philippines due to a warming earth and the reported shrinking of mangrove forests, an inventory of mangrove vegetation shielding the islands is urgently needed. In the past, there were efforts to map mangrove vegetation in the Philippines using Landsat imageries. However, the resolution was insufficient to estimate the mangrove patches at the municipality level. Recently, airborne LiDAR (Light Detection and Ranging) has provided opportunities for detailed mapping of natural resources such as mangroves. In this study, we propose a context-based method to extract mangrove patches from LiDAR data. The algorithm was implemented in MATLAB. The algorithm is context-based due to the fact that the parameters used are based on the actual nature of mangrove patches. It only utilizes two derivatives: the digital terrain model (DTM) and the canopy height model (CHM), which were generated using LAStools. The method primarily uses the following parameters: tidal height, canopy height, Shannon entropy, and patch area. The tidal height was used for estimating the intertidal zone. The CHM was used for excluding vegetation that are too short or too tall to be mangroves. The Shannon entropy is a measure of pixel disorder or variation. Areas with much pixel variation (e.g. edges in an image) exhibit large entropy while uniform areas (e.g. in the middle of mangrove patches) have small entropy. Patch areas that were too small to be mangrove communities were ignored. As a test case, the algorithm was applied to extract the mangrove patches in Pinamungajan, Cebu, Central Philippines. We also used k-means classifier to delineate the mangrove patches. The method shows comparable results as the k-means classifier (accuracy = 97.52%). The overall accuracies based on field validation and the orthorectified photos concurrently taken during the LiDAR survey using the context-based approach was 93.32%. In addition, the context-based algorithm executes significantly faster compared to the aforementioned classifier-based method.
Phil-LiDAR Research Center, University of San Carlos, Cebu City, Cebu, 6000, Philippines; Theoretical and Computational Sciences and Engineering Group, Department of Physics, University of San Carlos, Cebu City, Cebu, 6000, Philippines; Department of Biology, University of San Carlos, Cebu City, Cebu, 6000, Philippines
