UBC Theses and Dissertations
Application of the Getis statistic to the monitoring of riparian zones using mult-temporal radarsat images Eichel, Frank Herbert
Monitoring the condition of a large number of dispersed riparian management areas on northern Vancouver Island is hampered by poor ground access, the high cost of flying, and a lack of resources. Furthermore, monitoring them with high spatial resolution satellites, such as IKONOS, QuickBird, and Terra, is frustrated by persistent cloud cover. A possible alternative is to use radar imagery to facilitate monitoring. Satellitebased radar sensors, including Canada's RADARSAT, generate their own illumination and consequently they are able to acquire imagery at any time of the day or night. The imagery can also be acquired during stormy weather since the radar satellite's illumination is not obstructed by clouds. Radar backscattering from forested areas typically exhibits low spatial autocorrelation because these areas are structurally quite heterogeneous. On the other hand, freshly logged cut-blocks appear to be comparatively homogeneous on radar images. Consequently backscatter from these areas should exhibit higher spatial autocorrelation. Due to similar characteristics exhibited by windthrown areas within riparian management areas and along cut-block boundaries, it was postulated that the Getis statistic could be applied to multi-temporal RADARSAT images to detect both the increased backscatter and higher spatial autocorrelation associated with windthrow damage in these areas. The areas that were examined in this thesis are located on northern Vancouver Island in a triangular area between Port Hardy, Port McNeill, and Port Alice. The area is entirely forested and is under active management for timber production. Depending on location, existing stands of mature timber are comprised of one or more species of western hemlock, western red cedar, and Sitka spruce. Regenerated areas are dominated by western hemlock and western red cedar with wetter areas being reforested with Sitka spruce. Four high spatial resolution RADARSAT images provided the subscenes used in this study. These images, with a pixel spacing of 3.125 metres and a resolution of approximately 8 metres, were acquired in Fine 2 Mode in December 1996, August 1997, November 1998, and March 1999 on identical ascending orbits under similar rainy conditions. To better understand the behaviour of the Getis statistic with radar data, it was first used to detect increased levels of backscatter resulting from clearcutting. It was then applied to riparian zones to determine if it could detect increased backscatter resulting from windthrow damage. Each subscene was converted to a series of Getis value images by passing five kernels ranging from 3x3 to 11x11 in size over the image. A single Getis value representing the highest local spatial autocorrelation was selected for each pixel from the five Getis value images and then written to a maximum Getis (MaxGetis) value image. The associated MaxGetis Distance images indicated that, for the data used in this research, clearcut areas were just as heterogeneous as forested areas. Therefore, without an improvement in data homogeneity, it was not possible to determine the extent of disturbances such as clearcutting and windthrow on the premise that they exhibited high spatial autocorrelation over more extensive areas than adjacent forest. However, the characteristics of the MaxGetis image were such that the difference between areas of high and low backscatter was significantly enhanced and consequently multi-temporal composites of MaxGetis images were found to be especially useful for visualizing areas exhibiting high backscatter. Although thresholding MaxGetis difference images provided a means for determining where significant increases in backscatter had occurred, its reliability for detecting fresh windthrow damage or any other structural change in the landscape for that matter was thrown into doubt by the high number of apparent false alarms appearing in areas where no disturbance was known to have occurred. These false alarms arose as a result of a 95% thresholding level being applied to the difference image to isolate significant increases in MaxGetis values. Raising the threshold to 99% substantially reduced the number of false alarms although a substantial number remained. This led to the conclusion that the nature of the data used in this research does not allow thresholded MaxGetis difference images to be used as a reliable means of detecting significant increases in backscatter due to some form of disturbance. This research also determined that backscatter emanating from windthrow areas is considerably weaker than it is from freshly logged areas, thus making it difficult to distinguish windthrow damage from forest. Consequently, it is suggested that further research is required to determine if other types of radar data, such as multi-look and fully polarimetric data, are more suitable for detecting windthrow damage.
Item Citations and Data