Some observations on crop profile modelling

Publisher Website
Google Scholar
Abstract
LANDSAT data are typically available for a number of overpasses during a growing season. There is currently considerable interest in modelling the so-called crop profile for such multitemporal data by a non-linear profile function of some (spectral) index for the spectral bands. The derived coefficients are used in a subsequent allocation procedure. This paper outlines some results obtained from an evaluation of the approach for crop data from the wheatbelt of Western Australia. Specifically, the degree of separation of crop classes from pasture classes, as measured by the discriminant root, is compared for analyses based on the original bands, on various spectral indices and on fitted coefficients from a crop profile function describing the temporal change in these indices. For the data considered, a marked loss of discrimination is found for analyses based on various spectral indices, when compared with those based directly on the corresponding discriminant functions (where the linear combination is not constrained to be the same for each time). Analysis of the coefficients for the non-linear profile functions fitted to the indices results in further loss of separation.