Application des modèles statistiques spatio-temporels aux échantillonnages forestiers successifs

Abstract

When surveying the same forest on several successive occasions, sampling intensity may be reduced without any loss of precision by taking into account the spatial and temporal structures of the estimated variable. The theory of regionalized variables (RV) generalizes and improves the estimators derived from the classical sampling with partial replacement (SPR) theory. A general model accounting for both temporal and spatial structures is presented in the context of successive inventories. The best linear unbiased estimators (Blue) are derived by using the kriging technique. A comparison of RV and SPR estimators on a simple numerical example reveals that the variance can be overestimated with the classical estimators. An application to a forest decline survey illustrates how this theory may be used to choose and optimize a sampling strategy. Finally, the general interest as well as some practical problems of RV theory are discussed.