Sensitivity of correlation structure of class- and landscape-level metrics in three diverse regions

Abstract

Sensitivity of landscape metrics to selection of spatial scale (i.e., resolution or areal extent), land-use categories, and different landscapes has led to unreliable conclusions for practitioners of landscape analysis and modeling. Unlike previous studies that mostly considered such metrics and assessed the effect of each factor separately, our study focuses on the sensitivity of the correlation structure of different sets of landscape metrics as a whole under different situations via principal component analysis (PCA). We used the congruence coefficient (rc) to calculate the changes in factor structures under different situations. We used 16 class-level and 15 landscape-level metrics of 900 village-based and 150 town-based samples that were collected from three regions. Five cell sizes, two land-use classes, and two sets of land-use metrics were also considered. We did not control the cell sizes, sample extent, and different landscapes in the sensitivity analysis to study the interactive relationships between different factors. All factors strongly influence the correlation structure of the landscape metrics, with each factor demonstrating a unique influence. Changing cell size significantly affects the correlation structures in the plain region, especially in croplands and built-up lands. Town-based results show a relatively more stable correlation structure than village-based results (except in land-use categories). Different land-use classes show different responses to changing cell size, sample extent, and sets of landscape metrics in different regions. These results show the great interactive influences of these factors, which have often been overlooked in previous studies. The conclusions drawn from fixed factors may be conditional and inapplicable to other situations. The sensitivity of the correlation structure in diverse regions may improve our understanding of landscape metrics as a whole and can provide further insights into the correlation structure of landscape metrics for land-use management and monitoring.