This paper describes on-going co-operative research by the New South Wales (N.S.W.) Department of Agriculture and IBM Australia Limited. The aims of the project are to investigate if Landsat digital data can be used to map and monitor agricultural resources, particularly crop acreages and production, over large areas.
From the outset, multi-temporal whole scene computation has been a feature of the technical approach. Software modifications for large data volumes were carried out allowing supervised maximum likelihood classification to be used throughout the work. Considerable amounts of agronomic ground truth have been collected over wide latitudes for use in training the computer system as well as for accuracy assessment of classification results.
A number of Landsat scenes have been studied, with three to five acquisitions being registered for each. Preliminary analyses were conducted on historical data of Tamworth (N.S.W.) and Narrabri (N.S.W.) scenes. More intensive studies were undertaken for the 1980-81 wheat season in the Narrabri scene and are currently being undertaken in the Horsham (Victoria) scene. The techniques used and results from these analyses are discussed in this paper. Classification accuracy has been very encouraging showing excellent potential for use in crop area and production estimates.
Various problems emerged which required special attention and these are discussed. They included loss of data because of cloud cover, registration accuracy, training techniques, best combinations of bands and dates for classification, confusion classes, computation of very large volumes of data and classification accuracy assessment.
This work is demonstrating valuable applications of Landsat data in Australia. Technology transfer to other users is under way and it is anticipated that further technological development will lead to large scale adoption of remote sensing techniques for monitoring agricultural resources in Australia.
Date of this Version