A1325
Title: Auto-correlation-driven environmental sampling: an adaptive approach
Authors: Linda Altieri - University of Bologna (Italy) [presenting]
Daniela Cocchi - Dipartimento di Scienze Statistiche (Italy)
Abstract: Spatially balanced techniques are commonly used in environmental sampling. These methods return well-spread samples across the study area, which is considered a highly efficient approach for estimating the total or mean of a variable of interest. We have shown that such techniques perform well only if the variable is positively correlated over space. Conversely, in environmental data, it is common to encounter negative correlations, such as when plants or animals compete for soil or natural resources. We propose a new approach to spatial sampling, which sequentially adapts the selection of units to the type of correlation of the study variable. Our method is based on Spatially Correlated Poisson Sampling, incorporating a novel weighting system that either encourages or discourages the selection of units at specific distances, depending on the strength and nature of the correlation at those distances. The estimation of this association is adaptively refined as the sampling progresses. When the spatial arrangement follows a random or negatively correlated structure, our approach results in estimates with a significantly smaller estimation error. We demonstrate the effectiveness of this method through a case study using ecological data.
