Classification accuracy assessment

Classification accuracy refers to the correspondence between a class label assigned to a pixel and the true class. The true class is the label or category at a particular location that has been determined independently of the classification. This assessment was conducted to:

• enable the refinement of the classification to give the highest classification accuracy possible; and

• give an indication as to the likely accuracy of habitats derived by post-classification discrimination (Section 0).

In the latter case, the assumption is made that any data and decision implemented as part of post-classification discrimination will either increase or have no effect on classification accuracy. The results of directly testing mapping accuracy is presented in the form of an error matrix in Table 4.2c , showing classified versus true or reference assignments (Section 0). The Mangrove and Wetland classes were not assessed using this method because their results depended mainly from other sources of data (Table 4.2c).

Table 4.2c presents the errors in terms of user's accuracy, which is calculated as the number of correctly classified samples divided by the row total. The user's accuracy is termed as such because it represents the probability that a classified pixel represents that type on the ground (Cognalton 1991). Table 4.2c showed the highest inaccuracy occurred with Grassland being mis-classified as Shrubby Grassland (21.05%).