Uncertainty of landscape indices for spatial structure analysis

Abstract

Landscape indices or metrics have been widely used to relate spatial patterns to ecological processes or ecological functions. Landscape indices have several uses and applications. However, there is evidence that landscape indices are very sensitive to remote sensing imagery analysis. The aim of this study was to assess the effect of accuracy of satellite imagery classification on landscape indices for the analysis of spatial structure. We used the supervised classification method with different classification strategies based on spectral and ancillary information. Eight images classified using different strategies were produced from a single satellite image. Landscape indices at landscape and class level were computed. Overall, results showed that images with higher classification accuracy have lower coefficient of variation of landscape indices. Some landscape indices as Shannon and Simpson diversity, percentage of like adjacencies, and aggregation index showed lower sensitivity to the increase of classification accuracy. However, other indices as patch density and core area were more sensitive. Sources of error in the satellite imagery analysis are very diverse. Hence, caution is advised in the analysis of spatial structure using landscape indices.