MODELING ERRORS FOR REMOTELY SENSED DATA INPUT TO CIS

Different views of spatial resolution and accuracy present a major obstacle to the integration of remote sensing and GIS. Accuracy in remote sensing is modeled using probabilities of class membership in each pixel; in vector-based GIS it is modeled using concepts such as the epsilon band. The problem of linking the two views of accuracy reduces to one of realizing a stochastic process which must satisfy conditions of prior and posterior probabilities, and spatial dependence. We propose two suitable methods, one storage intensive and the other computationally intensive. The methods can be adapted to incorporate various forms of prior knowledge