Wingecarribee Swamp is the largest montane peatland on mainland Australia at an altitude of 680m above sea level and at a length of almost 5 kilometres. The swamp has significant conservation value as it provides habitats for a number of endangered and vulnerable flora and fauna. On 8 and 9 August 1998 the Wingecarribee Reservoir catchment experienced a major storm event that lead to substantial washout and collapse of the swamp. It was estimated that the collapse moved 5 to 8 million cubic metres of sediment into the adjoining reservoir. The collapse caused large sections of the swamp to fragment and become fissured. The Sydney Catchment Authority (SCA) and Department of Environment and Conservation (DEC) manage the swamp under the Wingecarribee Swamp and Special Area Plan of Management (WSSAPM). The WSSAPM identifies the need to protect and restore the ecological integrity of the swamp.
The main objective of this study was to develop a cost/benefit remote sensing framework to monitor the ecological integrity of Wingecarribee Swamp. A selection of commercially available remotely sensed datasets were chosen for this study based on their spatial and spectral resolutions. Two satellite sensors, Landsat 7 ETM+ and SPOT5 10m XI, were selected as they are readily available in Australia and they differ in the number of bands and spatial resolution. An airborne multi-spectral sensor, digital multispectral videography (DSMV), represented a dataset with a high spatial resolution, but low spectral bandwidth. The final dataset was an airborne CASI hyperspectral image with high spatial and spectral resolutions.
Aerial photographic interpretation (API) of 1:24,000 infrared photography was used in conjunction with field survey data to produce a community map for Wingecarribee Swamp to an accuracy of 82.4%. A total of sixteen plant community types and two physical classes were described using a hierarchical vegetation classification system developed in this study. This community map provided the base for selecting training sites and validating the classification of the multi- and hyperspectral imagery.
All image types evaluated in this study were unable to map all eighteen community types identified in the API. The results of the study showed that the spatial and spectral resolutions of the sensors played a major role in the classification accuracies achieved by each image type. The highest classification accuracy of 78.7% was produced by using spectral angle mapper (SAM) on a minimum noise fraction (MNF) transformation of the CASI imagery. However the CASI classification was only able to map ten of the eighteen community types. Landsat 7 ETM+ was considered inappropriate for mapping community types on the swamp due to its moderate spatial resolution. SPOT5 XI produced the highest classification accuracy from the two satellite sensors investigated in this study (72.5%). The spectral resolution of DSMV was the limiting factor in the separation of community types on the swamp, especially in the case of Salix cinerea (Willow). The best spectral separation of community types was found in the visible red to shortwave infrared wavelengths. Shortwave infrared bands were found only in the satellite sensors, Landsat 7 ETM+ and SPOT5.
The maximum likelihood classifier (MLC) was considered the best algorithm for mapping community types on the swamp when using multispectral data. The MLC performed very well when used on the CASI hyperspectral data, achieving only 0.5% less than the more intensive SAM technique when using the MNF image. The results of this study showed that the pre-processing techniques were more important in improving classification accuracies than the choice of classification algorithm. This was especially evident in the increase in classification accuracies when texture was used on the DSMV and MNF on the CASI. The further processing of data from its raw state increases the time required to produce classification maps. The increase in accuracy must be weighted against the time to get from image capture to resulting classification map.
The cost/benefit analysis of the results of this study demonstrated the importance of evaluating a range of technologies and techniques in developing a framework for monitoring medium-term projects. Although the classification of CASI SAM-MNF produced the best overall accuracy result, it was only 2.7% better than the DSMV, but at an extra cost of $33,000. Similarly, an additional 3.5% accuracy was achieved using DSMV instead of SPOT5 at double the cost.
The evaluation of remotely sensed imagery for mapping community types on Wingecarribee Swamp presented a framework for monitoring several key actions under the Wingecarribee Swamp and Special Area Plan of Management. This study has shown that high spatial and spectral resolution imagery was required to accurately map community types on the swamp. However, the sensors evaluated in this study did not produce classification accuracies that were considered acceptable for monitoring community types.
Future studies should evaluate remotely sensed imagery that has a spatial resolution of less than 10m and spectral bands that include a shortwave infrared wavelength for better discrimination of community types. The capture of SPOT5 10m XI and panchromatic 5m data simultaneously could provide the most cost effective dataset for monitoring the ecological integrity of the swamp due to its high spatial resolution and inclusion of a shortwave infrared band.