A comparison of spatial autocorrelation indices and landscape metrics in measuring urban landscape fragmentation

The combined use of remote sensing based land cover classification and landscape metrics has provided a positive step toward gaining a comprehensive understanding of the features of landscape structure. However, numerous limitations of land cover classification indicate that the utilization of classified thematic maps in landscape pattern analysis is questionable and may even lead to large errors in subsequent analyses. Instead of generating and employing detailed land cover classification maps, the utility of local spatial autocorrelation indices derived directly from satellite imagery to measure landscape fragmentation was examined. Since local spatial autocorrelation can capture spatial pattern at a local scale, it can be expected to detail the spatial heterogeneity for various parts of a landscape instead of providing a single value as in the case with the global measures. This study compares the traditional landscape metrics to the use of satellite imagery based local spatial autocorrelation measures in quantifying landscape structure over Phoenix urban area. Two local spatial autocorrelation indices, the Getis statistic and the local Moran's I were employed in evaluating landscape pattern, using normalized indices as the inputs. Results show that there is a clear relationship between local spatial autocorrelation indices and FRAGSTATS metrics. Both the Getis statistic and the local Moran's I can serve as useful indicators of landscape heterogeneity, for the entire landscape, and for different land use and land cover types. The paper provides a feasible methodology for urban planners and resource managers for effectively characterizing landscape fragmentation using continuous dataset.