Quadrat size dependence, spatial autocorrelation and the classification of community data

Abstract

Listen

Translate article

The paper evaluates spatial autocorrelation structure in grassland vegetation at the community level. We address the following main issues: 1) How quadrat size affects the measurement of spatial autocorrelation for presence/ absence and cover data? 2) What is the relationship between spatial autocorrelation and classification? 3) Is there a temporal change of spatial autocorrelation in the vegetation studied? We found that multivariate variogram shape, variance explained and the sill are different for presence/absence data and cover data, whereas quadrat size increases apparently introduce a stabilizing effect for both. Spatial stationarity is detected for species presence, and non-stationarity for cover. A new graphical tool, the clusterogram is introduced to examine spatial dependence of classification at various numbers of clusters. We found that spatial autocorrelation plays a crucial role in the classification of vegetation and therefore we suggest that its effect should not be removed from clustering. Mutual interpretation of variogram and clusterogram shape may be informative on the number of meaningful clusters present in the data. Spatial autocorrelation structure did not change markedly after 23 years for presence/absence data, indicating that the vegetation of the study area is stationary in time as well. The present study demonstrates that traditional quadrat data are suitable for evaluating spatial autocorrelation, even though field coordinates are recorded several years after sampling is completed.