Monitoring seagrass : an investigation with multi-temporal satellite imagery in Boullanger Bay, Tasmania

Seagrass plant communities play an ecologically important role in Australian and Tasmanian coastal regions. They provide key ecological functions such as: organic matter provision; assimilation of energy into ecosystems; nutrient trapping and cycling; shore line protection and formation; substrate sediment stabilization; enhanced biodiversity; and trophic transfers to adjacent habitats in tropical and temperate regions. The temperate Tasmanian seagrass communities, however, have experienced a period of decline in their abundance due to various disturbances as have other coastal areas in the world. Seagrasses are subject to rapid environmental changes arising from not only natural causes but also human-induced pressures, such as sea level rise, coastal development, sewage discharges and sediment runoff. In spite of the value of these aquatic plant communities, virtually no monitoring of the abundance and distribution of seagrasses habitats, in particular at a large geographic scale has been conducted in Tasmania, such as Boullanger Bay. Information on the extent and status of submerged aquatic vegetation (SAV), here largely seagrass meadows, at multi-spatial and multi-temporal scales are needed to support effective conservation and management. Due to resourcing challenges for boat or diver based monitoring, production of such information on the natural dynamics of SAV can only feasibly be tackled by the application of remote sensing techniques. However, a lack of knowledge about the efficacy of remote sensing techniques for seagrass mapping and monitoring persists. In the response to this need, this thesis summarises the development of an appropriate image processing scheme that produced viable data for interdisciplinary purposes in this particular location. Methods for mapping and monitoring SA V habitats distribution in Boullanger Bay at multiple spatial and temporal scales are trialled. Three case studies were conducted, including: (1) comparison between two hybrid image classification approaches for the investigation of method effectiveness; (2) change detection analysis of land cover classes in Boullanger Bay at two different spatial scales to determine the contribution of the moderate spatial resolution of Landsat and Advanced Land Observing Satellite (ALOS); and (3) change detection analysis to determine the efficacy of the moderate spatial resolution and annual temporal resolution of Landsat in both intertidal or subtidal seagrass dominated environments. Multi-temporal thematic map series of change detection results were produced over 18 years, from 1990 to 2008. The spatial and temporal changes in the occurrence of SAV meadows in Boullanger Bay were identified and presented that show the extent and distribution of SAV habitats and their rate of change. Case Study 1 investigated the efficacy of the remote sensing technique performed in the case studies was investigated. Two hybrid approaches: Independent Component Analysis (ICA) based Maximum Likelihood Classifier (MLC) approach and Principal Component Analysis (PCA) based ISODATA approach were compared to assess their ability to classify land cover objects. 'Error Matrix', image classification accuracy assessment technique demonstrated the better image classification accuracy of ICA based MLC approach (Overall accuracy: 88.4%, Kappa coefficient: 0.86) than PCA based ISODATA approach (Overall accuracy: 82.7%, Kappa coefficient: 0.79). Case Study 2 provided 'from - to' change between the classified land covers produced from Case Study 1. Write Function Memory Insertion (WFMI) change detection approach is used to effectively visualise the 'from - to' change between the land cover types. The study indicated that the saltmarsh/seagrass boundary was relatively stable over the study period. Conclusions about the relative change of habitats across the whole Boullanger Bay study site are limited due to image processing issues related to cloud and deeper water confounding the results. For SAV, it appears that there has been a decadal scale decline between 1990 and 2000 and then the areas remain stable through to 2008. In Case Study 3, firstly, an ICA based Multiple-date Composite Image (MCI) change detection analysis was performed to identify the spatial and temporal changes in the intertidal habitats, especially the Zostera muelleri seagrass, in the Welcome Inlet area. A relatively stable overall coverage was identified with fluctuating losses and gains in SA V meadows in many areas throughout the monitoring period at rates ranging from annual to decadal. Secondly, a WFMI approach was used for change detection analysis in the open subtidal area of the Boullanger Bay to identify the stability of subtidal SAV meadows from 1990 to 2008. The method revealed the very high stability of sand patches (i.e. uncolonised areas) within the dense Posidonia australis seagrass meadows over the 18 year period. Issues of the accuracy of thematic maps derived from Landsat and ALOS imagery were identified, including misclassification of land cover types in deep water areas i.e. > circa 7 m. However, the overall efficacy of the satellite sensors for mapping and monitoring SA V meadows in Boullanger Bay was supported.