Accuracy: Measures the agreement between a standard assumed to be correct and a classified image of unknown quality Precision: A measure of the sharpness or certainty of a measurement High accuracy means that bias is low; high precision means that the variability of estimates is low The usefulness of a map is related not only to its correctness but also to the precision with which the user can make statements about specific points depicted on the map Accuracy affects the legal standing of maps, the operational usefulness of data and the validity of research Sources of Classification Error In manual interpretations, errors are caused by the misidentification of parcels, excessive generalization, errors in registration, variations in detail of interpretation, and other factors The simplest causes of error are related to the misassignment of informational categories to spectral categories A simple landscape composed of large, uniform, distinct categories is easier to classify than one with small, heterogeneous, indistinct parcels arranged in a complex pattern Key landscape variables: Parcel size, Variation in parcel size, Parcel identities, Number of categories, Arrangement of categories, Number of parcels per category, Shapes of parcel and Radiometric & spectral contrast with surrounding parcels Error Characteristics Classification error: The assignment of a pixel belonging to one category to another 3 Common Error Characteristics 1) Errors are not distributed at random, but display a degree of systematic, ordered occurrence in space 2) Often erroneously assigned pixels are not spatially isolated but occur grouped in areas of varied size and shape 3) Errors may have specific spatial relationships to the parcels to which they pertain Error Matrix The standard form for site-specific error is the error matrix, sometimes known as the confusion matrix because it identifies not only overall errors for each category but also misclassifications by category To construct the error matrix, the analyst must compare 2 sources of data – the reference samples and the classified image – on a point-by-point basis to determine exactly how each of the validation samples is represented in the classification The validation data and the image should be acquired as close in time as possible Error of omission: Pixel 1 on the ground is assigned to Non-Pixel 1 on the map (The loss of a pixel) Error of commission: Non-pixel 1 on the ground is assigned to Pixel 1 on the map (The gain of a wrong pixel) User’s accuracy: From the perspective of the user, how accurate is the map? Corresponds to error of commission Producer’s accuracy: From the perspective of the mapmaker, how accurate is the map? Corresponds to error of omission