Testing the accuracy of remote sensing land use maps

Some of the main aspects that need to be considered in a remote sensing sampling design are: (1) the frequency that any one land use type (on the ground) is erroneously attributed to another class by the interpreter; (2) the frequency that the wrong land use (as observed on the ground) is erroneously included in any one class by the remote sensing interpreter; (3) the proportion of all land (as determined in the field) that is mistakenly attributed by the interpreter; and (4) the determination of whether the mistakes are random (so that the overall proportions are approximately correct) or subject to a persistent bias. A sampling and statistical testing procedure is presented which allows an approximate answer to each of these aspects. The concept developed and described incorporates the probability of making incorrect interpretations at particular prescribed accuracy levels, for a certain number of errors, for a particular sample size. It is considered that this approach offers a meaningful explanation of the interpretation accuracy level of an entire remote sensing land use survey.