Habitat map of seagrass cover derived from a supervised moderate-spatial-resolution multi-spectral satellite image, integrated with manual delineation and coincident field data, Moreton Bay, 2011

A supervised classification was applied to a Landsat TM5 image. This image was acquired 9:40 am, on the 27th July 2011 (5.14 am low tide at Brisbane Bar). The image classification was applied on areas of clear waters up to three metres depth and for exposed regions of Moreton Bay. Field validation data was collected at 4797 survey sites by UQ. GPS referenced field data were used as training areas for the image classification process. For this training the substrate DN signatures were extracted from the Landsat 5 TM image for field survey locations of known substrate cover, enabling a characteristic "spectral reflectance signature" to be defined for each target. The Landsat TM image, containing only those pixels in water < 3.0m deep, was then subject to minimum distance to means algorithm to group pixels with similar DN signatures (assumed to correspond to the different substrata). This process enabled each pixel to be assigned a label of either seagrass cover (0, 1-25 %, 25-50 %, 50-75 % and 75-100 %). The resulting raster data was then converted into a vector polygon file. Species information was added based on the field data and expert knowledge. Both polygon files were joined by overlaying features of remote sensing files with the EHMP field data to produce an output theme that contains the attributes and full extent of both themes. If polygons of remote sensing were within polygons of field data the assumption was made that the remote sensing polygon was showing more detail and the underlying field polygon was deleted.

本研究对一幅陆地卫星5号专题制图仪（Landsat TM5）影像开展监督分类任务。该影像拍摄于2011年7月27日上午9时40分，布里斯班航道当日低潮时刻为凌晨5时14分。本次分类的覆盖范围限定为水深不超过3米的清澈水域，以及莫顿湾（Moreton Bay）的裸露区域。昆士兰大学（University of Queensland，UQ）在4797个调查点位采集了野外验证数据，以经GPS定位的野外实测数据作为影像分类流程的训练样本区域。本次训练中，研究人员从陆地卫星5号TM影像中提取已知底质覆盖类型的野外调查点位处的数字数值（Digital Number，DN）特征，据此为每一类目标地物定义特征性的“光谱反射特征（spectral reflectance signature）”。后续仅保留水深小于3.0米的水体像素，对该陆地卫星TM影像应用最小距离均值分类算法，将具有相似DN特征的像素归为一类，假设其对应不同的底质类型。通过该流程，每个像素被赋予相应的分类标签，涵盖海草覆盖度的五个等级：0%、1%~25%、25%~50%、50%~75%以及75%~100%。所得栅格分类结果随后被转换为矢量多边形文件。研究人员基于野外实测数据与专家经验，为矢量文件添加了物种相关属性信息。通过将遥感矢量要素与EHMP野外调查数据进行空间叠加，将两类矢量文件进行融合，生成同时包含两类数据属性与完整覆盖范围的输出专题图层。若遥感矢量多边形完全位于野外调查矢量多边形内部，则默认遥感多边形包含更精细的地物细节，遂删除对应的底层野外调查多边形。