Variation in landscape metrics derived from multiple independent classifications
This study assessed how 60 independent land cover classifications of the same Landsat thematic mapper (TM) scene by students in an introductory image analysis course influenced classification accuracies, patch number, and mean patch size. Overall accuracies ranged from 39-62\% and were not correlated to the selected landscape metrics. This emphasized the need to evaluate class-specific accuracies which correlated more strongly with the metrics. Analysts with a greater number of pixels classified as deciduous trees,’conifers, other vegetation, and impervious/disturbed areas had more and larger patches .of these classes. However, a greater number of corn pixels decreased the number of corn patches and increased their mean size. Further, increasing the number of water pixels created new but smaller patches of water. These results indicate that the variation among analysts can have a subtle, but significant, effect on the values of landscape metrics. This should be useful for researchers and decision-makers who typically need to interpret values derived from only a single classification.
Lake Buena Vista, Florida