Sampling with Multiple Objectives and the Role of Spatial Autocorrelation

Abstract

The handling of mixed type multivariate data sets is discussed on an example from the ‘Man and Biosphere’ project, Davos, Switzerland. The data comprise variables of different spatial resolution, aggregation and reliability. Correlo-grams are computed for data subsets describing different landscape features on differing scales. Some variables show spatial independence, others exhibit correlation among adjacent sampling localities. Periodicity is detected in the distribution pattern of some animal species, and soil types form coenoclines. While some results derived by previous modelling reflect local patterns, others suggest wide ranging relevance. Optimum sampling intensity therefore should not only depend on the aims of the study, but be also influenced by the nature of the variables. For investigations with multiple objectives, the simultaneous use of a combination of sampling designs is suggested. Quadrat size, grid width, and even investigation areas may vary. The commensurability of the designs can be achieved by simultaneously running the operations at the different aggregation levels for the relevant variables.