Accuracy Assessment

Accuracy assessment is an essential component of the investigation to quantify whether the data quality of one classification method is superior over others (Sader et al. 1995). Accuracy assessment is the process in which the image is partitioned into a set number of groups (classes) based on the values of the pixels in one or more image channels. The accuracy was set at more than 80 % as the threshold for accurate values. A total of 255 reference pixels per class were selected in a stratified random sampling approach for each class to assess the accuracy. Accuracy assess­ment involves identifying a set of sample locations (ground verification points) that would be visited in the field. Then the land cover identified in the field was compared to that mapped in the image during supervised classification for the same location by means of error or confusion matrices (Jensen 1986; Stehman 1996). Based on the confusion matrices, different accuracy measures were calcu­lated: producer’s accuracy, user’s accuracy, and overall accuracy. To summarize the classification results, overall accuracies with 95 % confidence intervals were also generated. KAPPA analysis yields (Khat statistic) were also calculated to measure agreement or accuracy as suggested by Congalton (1991).

13.2.2 Landscape Pattern Analysis

There are more than a hundred indices for quantifying landscape pattern. A group of selected metrics (or indices) can be useful to interpret the landscape changes and considered relative to the type of patches (Apan and Peterson 1998). Landscape pattern metrics were run on images for the year studied. This analysis is designed to generate spatiotemporal indicators of landscape pattern as part of the scale-pattern – process paradigm (Walsh et al. 1998). Six landscape metrics were chosen in this study: class area (CA), percent of landscape (PLAND), number of patches (NP), patch density (PD), mean patch area (AREA_MN), and largest patch index (LPI) (Table 13.2). The spatial metrics were generated by FRAGSTATS version 3.3, a software package that calculates a number of spatial structures at three levels: entire landscape, class, and patch levels (McGarigal and Marks 1995; McGarigal et al. 2002).