Northern Territory University, Darwin

Magen Geyer

1997

More specifically, the invasive weed, Mimosa pigra and native vegetation in the Oenpelli floodplain were assessed.

The review of Landsat TM's ability to map land cover types led to the image classification comprising bands 2, 3, 4 and 5. The classification process involved the hybrid method. Two algorithms, the minimum distance to means (MDM) and the maximum likelihood classifiers (MLC) were used to classify herbaceous vegetation and bare clay soil areas.

The effectiveness of chemical control was analysed. The distribution of Mimosa infestations inhabiting the floodplain were destroyed after the application of herbicides. This allowed native vegetation to slowly recuperate in the flood plains. Thus, the monitoring of these vegetation changes were observed. The area of change of native floodplain vegetation and bare soil was estimated in relation to the rate of Mimosa expansion.

Actual field data could not be collected. This was compensated by visual interpretations of the final land cover classification maps made by an expert who had field knowledge of the study site. In addition, the accuracy of both the MDM and MLC algorithms were compared by error matrices.

The accuracy assessments for six of the eight images were highly acceptable, with overall mapping assessments greater than 85'7o. Accuracy assessment for both the 1994 images were substantially lower. Overall mapping accuracies of 79@'o and 77@'o were calculated for the MDM classification and MLC, respectively. Sources of errors were identified, primarily due to TM's inability to distinguish between floodplain vegetation types. However, analysis of the results concluded that Landsat TM has high potential for meeting the current and future needs in relation to mapping and monitoring land cover types.

Supervisors

Professor Greg Hill, Northern Territory University
Dr Waqar Ahmad, Northern Territory University
Dr Garry Cook, CSIRO W&E