Validating global landcover classification pattern using spatial resolution-independent landscape metrics

Abstract

The EROS Data Center is completing a global land-cover classification using 1 km data. Although 1 km data is very coarse for land-cover classification, it is the only means by which landcover can be classified at a global scale. A number of validation schemes have been devised and proposed for use in evaluating global landcover classification. These validation schemes are primarily concerned with assessing thematic accuracy. In addition, to the importance of evaluating thematic global classification accuracy, the authors have recognized the importance of evaluating changes in landscape pattern at coarser scales. Thus, they have developed landscape metrics that are capable of quantifying landscape pattern while being independent of spatial resolution. These metrics include measures of patch shape complexity and landscape fragmentation. The metrics are similar in concept to those proposed for landscape monitoring and assessment by the US Environmental Protection Environmental Monitoring and Assessment (EMAP). However, the metrics proposed by EMAP are dependent upon a number of factors with respect to remote sensing data, particularly, spatial resolution, level of classification, and sampling geometry, unlike the authors' metrics which are independent of these factors. Following each level of classification the pattern metrics could be calculated to determine if coarser resolution data are capturing the same type of changes in pattern in landcover categories as finer resolution data. Such information would allow researchers and decision makers to establish an error estimate on the use of 1 km data for capturing landscape patterns. It is important to determine if landscape pattern is detected because pattern has many implications on regional and global environmental processes such as biodiversity, animal migration, vegetation dispersal, predator-prey interactions, plant succession, nutrient cycling, and climate. By validating the 1 km data the authors be able to better assess the effects of pattern on global processes.