Landscape Complexity and Remote Classification in Eastern Coastal Mexico: Applications of Landsat‐7 ETM⁺ Data

Abstract

This paper addresses multiple semi‐automated methods for discriminating landuse/landcover (LULC) classification in the absence of field data. The study site comprises a heterogeneous mixture of physiographic landscape elements in a remote area of eastern Mexico. Classification of modified Tasseled Cap transformations as well as a hybrid unsupervised‐supervised classification technique were employed using a Landsat 7ETM+ scene; the panchromatic band was reserved for accuracy assessment. Using a random stratified sampling scheme, overall accuracies ranged from 61.5% to 83.3% and Khat accuracies varied from 56% to 81.1%. The classifications derived from modified Tasseled Cap transformations did not appropriately discriminate landscape components, possibly owing to landscape heterogeneity. Classification results were improved through the use of landscape stratification in the attribution phase in the hybrid classification procedure, recommended for complex environments. Lastly, this work underscores the emerging importance of reserving Landsat 7ETM+ panchromatic information for accuracy assessment when field data are lacking.