Improvements in the mapping of shallow marine habitats through predictive spatial modelling

Abstract

The UK has an obligation to monitor certain sites around its coastline due to various International environmental designations. Remote sensing is ideal as these areas are often difficult and hazardous to access. Traditionally, classification has been based on the spectral responses of targets, but indistinct spectral signatures between targets often leads to poor classification accuracy. To improve this, other methods of classification such as predictive spatial modelling (PSM) incorporate ‘knowledge’ of a study area in addition to the target’s spectral information. Well documented natural zonation of shallow marine habitats make them especially suitable for classification using PSM techniques. Using the ERDAS Knowledge Engineer, a series of knowledge-based rules relating to exposure, depth and substrate were tested, combined with QuickBird high spatial resolution remotely sensed data, to improve the classification of a study area in the Sound of Harris, Outer Hebrides, Scotland. Results show that after preliminary supervised classification based on spectral parameters, the additional rules enabled further subdivision of spectrally similar classes, and the simultaneous classification of both subtidal and intertidal biotopes in one process. This is an improvement on previous attempts which subset and separately classified the subtidal and intertidal areas. While the approach shows potential, considerable biotope confusion still exists, which resulted in a low overall accuracy of just under 40%. The need for further discriminating factors to improve the classification, as well as a more robust knowledge based approach, are highlighted as potential future improvements.