Stochastic nature of landsat MSS data

Abstract

A multiple series generalization of the ARIMA models is used to model Landsat MSS scan lines as sequences of vectors, each vector having four elements (hands). The purpose of this work is to investigate if Landsat scan lines can be described by a general multiple series linear stochastic model and if the coefficients of such a model vary as a function of satellite system and target attributes. To accomplish this objective, an exploratory experimental design was set up incorporating six factors, four representing target attributes-location, cloud cover, row (within location), and column (within location)-and two factors representing system attributes-satellite number and detector bank. Each factor included in the design at two levels and, with two replicates per treatment, 2 7 or 128 scan lines were analyzed. The results of the analysis suggests that a multiple AR(4) model is an adequate representation across all scan lines. Further, the coefficients of the AR(4) model vary with location, particularly changes in physiography (slope regimes), and with percent cloud cover, but are insensitive to changes in system attributes.