Objective class definitions using correlation of similarities between remotely sensed and environmental data

Abstract

This letter presents a natural approach for selecting a subset of environmental variables to define classes prior to mapping them efficiently using remotely sensed data. The procedure is based on correlation calculations between a set of 'environmental' similarity matrices and an 'image' similarity matrix. The correlation is a modified Spearman rank correlation. Using a SPOT-HRV image and seven environmental variables, an application in a coral reef environment showed that only three environmental variables are relevant to defining classes prior to mapping.