AbstractThe major focus of this thesis was to estimate Melaleuca biomass on tropical floodplains using remotely sensed data. The study areas were the Mary Floodplain System, northern Australia, and the floodplains of the Trans Fly Bioregion, southern New Guinea. In order to achieve the major focus, two specific key issues were addressed: the linking of field data to precise locations in remotely sensed imagery; and the selection of an appropriate remote sensing technology for mapping Melaleuca above ground woody biomass. In addressing these two issues, this thesis demonstrates that above-ground woody biomass of Melaleuca species on tropical floodplains can be determined from SAR data.
In determining the factors contributing to the accurate linking of ground data with image data for estimating a biophysical variable (biomass), a spatial statistic was applied to various remotely sensed image data sets. Neither the traditional root mean square error nor the spatial statistics currently employed in remote sensing applications address the problem of spatial registration errors in linking image data with a biophysical parameter on the ground. The statistic used in this study is a local statistic based on the average values within a window of increasing dimensions and assesses the variation occurring through a positional error element. This spatial statistic was used to: firstly, assess the effect of spectral and frequency orientation controls on determining the location and size of field plot sizes from remotely sensed data; and secondly evaluate spatial resolution controls and the impact of locational uncertainty on determining the location and size of field plots.
The results obtained from applying the spatial statistic for estimating an optimal field plot size and location to sample, taking into account positional errors, indicated that the optimal plot sizes and locations varied according to spectral bandwidth (location and width) and polarization (in the case of SAR data). It was shown that the statistic does not simply provide a measure of homogeneity (or conversely heterogeneity), but highlights spatially where the local variance is less than the global average. Therefore, the boundaries or ecotones of highly variable features and their adjacent habitats are well defined. These ecotone regions are displayed as the largest plot sizes by the statistic.
In relation to spatial resolution controls, the results of this study show that the optimum sample size is dependent on the image data as well as the ground characteristics. The results confirm that the highest resolution data provide the largest scope for selecting suitable plots in the varied environment of the test area only if the spatial registration is comparatively accurate. This fact was highlighted by the ADAR data by using a spatial error of 10 and 50 m. It was clearly demonstrated that the impact of registration error on suitable sample locations is considerable. Using an error estimate of 50 m, there are no areas in the data, where Melaleuca could be sampled for a quantitative comparison.
The results of the implementation of the spatial statistic to various image data sets shows that it is important that results from one sensor/image are not transferred to other sensors/images. It is suggested that the spatial statistic is implemented at the outset of any remote sensing investigation that links image data with field data relating to biophysical parameters.
It was determined through an extensive literature review that quad-polarised SAR data was the most suitable remotely sensed data by which to estimate above ground woody biomass of Melaleuca habitats in tropical floodplain environments. It is clear that the P-HV channel is most suitable for this task, as the L-Band channels are susceptible to saturation effects at lower biomass levels and the co-polarised Pband channels appear to be affected by understorey responses. The results of the JERS-1 analysis show some promise for examining Melaleuca encroachment in the Trans-Fly Bioregion. However, these results are not conclusive due to the time lag between the acquisition of image and field data, and the dynamic nature of the environment. There is the potential to use future space borne L-Band sensors with quad-polarisation for this task.
This study has also demonstrated that quantifying biomass relies on accurately locating sites and determining an appropriate plot size for field sampling. This was achieved through implementing the spatial statistic. The poor results obtained from the transect/point data highlight the importance of linking ground and image data in terms of plot size and location.
This research has formed a basis for further development of models and research related to quantifying the above ground biomass of wetland forest habitats in the tropics of the northern Australia and New Guinea. With the impending launch of JAXA’s ALOS satellite system late in 2005, there is the opportunity to obtain L-HH and L-HV data at 10 m resolution for the study regions. This will enable further SAR data capture for the Trans Fly Bioregion site, and field sites within the region to be collected using the spatial statistic. A suitable regression model can then be established using multi-date imagery. With a robust model, it is assumed rapid Melaleuca encroachment with the study region can be effectively quantified and monitored. Further work can be conducted on the effect of understorey composition and flooding on SAR backscatter. In future investigations, dry sites need to be incorporated more fully. The effect of understorey composition (aquatic grasses) needs to be examined in more detail also. Finally, the input of the above ground woody biomass model in the Mary Floodplain system could be linked with biogeochemical models relating to methane and carbon. These biogeochemical models do not exist at present but it is a relevant future research direction that would contribute to furthering the knowledge we have on the role of the remote floodplains of tropical northern Australia.
|Date of Award||Apr 2005|
|Supervisor||Greg Hill (Supervisor)|