|Abstract||Two dates of C-HH SAR data acquired over the North Battleford, Saskatchewan, area in the early part (May 10 and June 12) of the 1987 growing season were used alone and in combination with a TM scene
(May 24) to investigate the crop classification ability of these data. Supervised maximum likelihood classification techniques were used to discriminate fields of early- and late-planted canola, grain, summer-fallow, pea, and lentil crop types.
Single-channel classifications produced accuracies ranging from 31 to 45 per cent, with the June 12 SAR data producing the best accuracy. Although the single channel overall classification accuracies were quite low, each of the crop types were
identified with accuracies of 60 per cent or better on one or more channels. Subsequently, the overall classification accuracy improved as the number of channels used in the classification increased. When TM bands 3, 4, and 5 were combined with both
dates of SAR data, an overall classification accuracy of 77 per cent was achieved. When a field classifier with a 51 per cent threshold is applied to these classification results, an overall accuracy of 75 per cent correct classification is achieved.
The results are discussed in relationship to the crop calendar and the potential for improvement by adding data acquired during later periods in the growing season. When RADARSAT is launched and multi-temporal SAR data are readily available, accurate
crop discrimination using SAR data augmented by TM data will be attainable. |