A MULTI-SCALE APPROACH FOR BENTHIC HABITAT MAPPING OF SHALLOW WATER REGION OF GUIUAN, EASTERN SAMAR USING A 4-BAND HIGH RESOLUTION WORLDVIEW-2 SATELLITE IMAGE USING OBJECT-BASED IMAGE ANALYSIS

摘要

High resolution multi-spectral satellite images had been optimized in mapping the shallow water region of the coastal environments. This helps in the monitoring and management of marine systems and coastal resources over large areas. The objective of this research is to utilized a 4 band (Red, Green, Blue, NIR) WorldView-2 satellite image of Guiuan, Eastern Samar to produce a benthic habitat map using object-based image analysis. The image had undergone two major steps: pre-processing and classification. The pre-processing comprised of several steps essential to correct the raw satellite image file for benthic habitat mapping. First, the image's reflected radiance was calibrated and corrected from water surface and atmospheric scattering. Then, land and cloud masking was performed. Next. Hedley's deglinting formula was applied to remove the effects of sun glint in the image by utilizing the NIR band of the dark pixels. The water column correction was adapted from Lyzenga to remove the effects of the absorption and scattering of photons in the water. A depth invariant matrix was calculated and applied to the band pairs of good water reflectors. The inputs for the multi-scale classification were the pre-processed WorldView-2 satellite image, and color transformation outputs from RGB to HSV and RGB to HLS outputted into 3 layers using Prinicipal Component Analysis . For the classification, a multiscale segmentation was applied using a hierarchical top to bottom scheme. Starting with a bigger scale parameter for shallow water region and smaller scale for the benthic community zones. Different segmentation parameters such as weights, shape, and compactness were performed to enhance the object found in each zones. A classification algorithm was utilized for the WorldView-2 Image called Support Vector Mechanics(SVM) using eCognition, an object image analysis software. By using the training data set, the mean layer values, their standard deviations and their textural characteristics were inputted for the benthic classes' object features. Ground measurements from video tows and sampling confirmed the accurate representation of the classified benthic habitat maps.