5. Sensing Uncertainty

Image extraction methods, employing statistically based image classifiers, implement decision rules that partition spectral space into mutually exclusive categories. These methods perform well, particularly when measurements are precise and our conceptualizations of environmental process are unambiguous. In those specific situations, using Boolean logic to delineate crisp search spaces, support by traditional expressions of probability sufficiently capture items of interest reliably enough to permit their representation as thematic content. However, there are situations we encounter in the study of the environment where our ability to define the processes or parameters in exact terms is impossible and the environmental conditions we wish to categorize display a level of complexity that defies clear and certain classification. The conceptually simple problem of mapping degraded land or a deforested area are examples that illustrate the presence of “maybe” in our search for information and underscores the frustrations that follow from the inherent vagueness of our definitions and our adherence to ridged models that fail to conform to the lack of certainty that surrounds most environmental issues. In this chapter, we will explore approaches to image classification that departs from strict statistical methodologies. In our exploration, we will examine how these different models exploit the uncertainty and produce very different realizations of environmental conditions, presenting information in ways that communicate differently in the context of decision making.