Canonical Analysis : The Use of Transformed Landsat Data for Crop Type Discrimination

The primary goal of this thesis is to investigate the usefulness, accuracy, and efficiency of canonical analysis (a linear transformation technique) for the classification of crop types using multi-date multispectral scanner (MSS) Landsat digital data. The accuracy statistics and the computer processing efficiency of the crop type classification developed using a canonical transformation will be compared with accuracy and efficiency figures of a crop classification developed for the Columbia River and Tributaries Irrigation Withdrawals Analysis Project (Johnson, Loveland, Anderson, 1981) in which multi-date Landsat data covering the same area was classified. The Clarke, Oregon 7.5 minute USGS quadrangle was chosen as the study site because of the availability of timely Landsat data and the existing maximum likelihood classification from the Columbia River and Tributaries Project. In a general sense, the problem investigated is a method of gathering crop type information using remotely sensed data. As discussed in the Introduction, an important method for crop type identification is the classification of multi-temporal Landsat digital data. Using remotely sensed data in the form of Landsat computer compatible tapes (CCT\u27s) allows for the analysis of the full spatial resolution of Landsat data and the analysis and manipulation of the numeric sensitivity of digital data