Statement |
A semi-automated object-based image analysis (OBIA) was performed to produce the map. OBIA is a two-step approach consisting of segmentation and classification (Blaschke, 2010), implemented in the software package eCognition v8.7.2. The backscatter image is segmented into objects (sections of the image with homogenous backscatter characteristics). For each of these objects, mean values of the primary acoustic data layers and their derivatives were calculated.
Each stage in the process is numbered and described in detail below.
Stage 1. Data Preparation
Prior to analysis, bathymetry data were re-sampled onto a grid at 2 m resolution, whilst backscatter data were re-sampled at 1 m resolution. Default ÔÇÿno dataÔÇÖ values in the Floating Point Geotiff files were transformed to null values within ArcGIS v10.1.
Stage 2. Derivatives calculated
A range of derivatives were calculated from bathymetry and backscatter datasets. Only the backscatter dataset, however, was used for classification.
Stage 3. Segmentation
Segmentation divides an image into meaningful objects based on their spectral and spatial characteristics. The resulting objects can be characterised by their various features, such as layer values (mean, standard deviation, skewness, etc.), geometry (extent, shape, etc.), texture and many others.
The input layers used were the primary acoustic data layers (bathymetry and backscatter strength) and the derivatives roughness and BPI25.
Segmentation was carried out using the multi-resolution segmentation algorithm in eCognition. This is an optimisation procedure that starts with an individual pixel and merges it consecutively with neighbouring pixels to form an object. The process continues until a threshold value for a scale is reached. The threshold is set by the operator, who determines the variability allowed in the objects.
The goal of segmentation is to create meaningful objects that represent areas of homogenous values in the map image. The size of the objects is influenced by the scale parameter mentioned above and the heterogeneity of the image. For a fixed value of the scale parameter, a homogeneous area of seabed will have larger objects than a heterogeneous area. Likewise, for a fixed seabed heterogeneity, larger values of the scale parameter produce larger objects than smaller values. The scale parameter was selected using the Estimation of Scale Parameters (ESP) tool. The tool calculates local variance (LV) of object heterogeneity within a scene for increasing scale parameters at user-defined intervals. The threshold for rate of change of LV relative to the data properties in the entire image, can be used to indicate the scale level at which the image can be segmented in the most appropriate manner (Dragut et al., 2010). The final segmentation was carried out at pixel level on backscatter strength, bathymetry, BPI25 and roughness with the scale parameter set at 10.
Stage 4. Classification
For each of the objects created, mean values of the primary acoustic data layers and their derivatives were calculated (e.g., the mean backscatter value for the grid cells lying within the object) for further statistical analysis and modelling. Objects and their feature mean values were exported as a GIS shapefile for further use in assigning their corresponding sediment class and producing a broadscale habitat map. Though ÔÇÿHigh Energy Infralittoral RockÔÇÖ and ÔÇÿModerate Energy Infralittoral RockÔÇÖ broadscale habitats were identified from video and still analysis it was not possible to predict the distributions of these habitats due to their similar reflectivityÔÇÖs, and a lack of discriminatory evidence. With this in mind they were grouped to create one habitat of ÔÇÿInfralittoral RockÔÇÖ for prediction and classification purposes. Furthermore, ÔÇÿSubtidal Macrophyte Dominated SedimentÔÇÖ was identified as part of the video analysis. This particular broadscale habitat is highly associated with subtidal coarse sediments with one being discriminated from the other based on biological components rather than physical ones. To this end remote techniques were not suitable for the discrimination of these two habitats, from one another, as investigation of the data did not allow for an environmental discriminator to be identified. Classification of ÔÇÿSubtidal Macrophyte Dominated SedimentsÔÇÖ was therefore reduced to classification as ÔÇÿSubtidal Coarse SedimentÔÇÖ. |