Tree-ring research and the theory of sampling: NFI-based samples strength and specificities
Tree-ring research studies depend heavily on samples, but the connection to the sampling theory remains loose. Tree-ring studies need to be based on samples of restrained size for several reasons: measuring tree-ring features remains labour-intensive, costly, and sampling cannot be seen as totally free of consequences for trees. Sampling, the art of selecting observational units to produce population-level estimates, comes with constraints that, when not respected, lead to biases and loss of confidence.
When constructing tree-ring series, several spatial scales are been crossed: the sites, which selection represents one sampling step; the trees, which selection was perhaps the most studied and proved so influential in many field.
The theory of sampling was developed initially for discrete and finite populations but recently also expended to continuous populations. Particularly suited to forests, these methods bring solutions to the sampling of one of the most difficult populations: the forest trees. National Forest Inventories (NFIs) are large-scale multipurpose surveys based on such probabilistic sampling and estimation methods, which ensure unbiased area-based estimations over large territories. The sampling framework of NFIs hence produce invaluable samples with desirable properties for inference, particularly ensuring balanced sampling and spatial representativity.
But, the advanced methods of NFIs also come at the price of a high complexity in the data structure, and the use of the samples they generate. A short list of points of concerns is recapitulated: loss of independence, unequal sampling probabilities, spatial and temporal correlations. Hence precautions in their use is necessary.