Mapping mangrove forest from LiDAR data using object-based image analysis and Support Vector Machine: The case of Calatagan, Batangas

Efforts have been made in conserving the mangrove forests of Calatagan, Batangas. As an aid in the conservation and environmental assessment of the said forest, this study presents a method in producing fine scale map of these protected areas. The study site covers the northern part of Calatagan consisting of 12 barangays, seven of which are in the coastal zone. Serving as input in image classification, airborne LiDAR (Light Detection and Ranging) data measuring approximately 72 km2 was provided by DREAM LiDAR. Combinations of multi-threshold and multiresolution segmentation enabled the pre-processing stage before moving on with the classification. It was observed that non-ground features are best segmented by giving weights to pit-free CHM, average intensity and number of returns arithmetic. Rule-based approach was implemented in identifying artificial surfaces such as buildings and fences; then, the Support Vector Machine (SVM) algorithm was applied in classifying the remaining non-ground objects. Fourteen features were selected for the training phase of SVM: eight from LiDAR-derived layers (number of returns, vertical features and intensity) and six from geometry features. Applying accuracy assessment based on test and training area (TTA) mask derived from validation points, the producer´s and user´s accuracy for mangrove, artificial surfaces and other vegetation were above 0.87. As a result, the overall accuracy of the produced fine scale mangrove map was 0.91 with a Kappa Index of Agreement of 0.85. The mapping product can then be used by local government unit and management agencies as a reference in baseline assessment of mangrove resources, and in planning and implementing further conservation efforts.