Controlling data uncertainty via aggregation in remotely sensed data

Aggregation may be used as a means of enhancing remotely sensed data accuracy, but there is a tradeoff between loss of information and gain in accuracy. Thus, the choice of the proper cell size for aggregation is important. This letter explores the change in data accuracy that accompanies aggregation and finds an increase in image thematic accuracy with increasing cell size, resulting from 1) reduction in the impact of misregistration on thematic error and 2) mutual cancelation of inverse classification errors occurring within the same cell. A model is developed to quantify these phenomena. The model is exemplified using a vegetation map derived from an aerial photo. The model revealed a major reduction in effective location error for cell sizes in the range of 3-10 times the size of mean location error; reduction in effective classification error was minor.