At the interface between domain knowledge and statistical sampling theory: Conditional distribution based sampling for environmental survey (CODIBAS)

Abstract

In environmental research, there is an extensive knowledge of the underlying processes that cause a distinct spatial landscape pattern of system properties. Statistical sampling theory deals with how a dataset must be constructed that allows for the transferability of the insights from the collected data to the system. Stratified and balanced sampling schemes applied for environmental survey, seek to reduce the necessary amount of data to capture the spatial heterogeneity of a landscape. The particular design’s specification depends on the field of application. Against this background, the author strives to draw the attention to the conceptual shortcomings of conditioned Latin hypercube sampling (cLHS) in the context of soil survey and digital soil mapping. Furthermore, a new sampling design is presented which (1) combines the advantages of both, stratified and balanced designs, (2) shows consistency in the application of pedogenetic theory, and (3) which can be obtained by some simple modifications to the computer code of cLHS. Overall, this manuscript shall promote a vivid discussion in the Pedometrics community concerning the consideration of scientific domain knowledge in statistical sampling theory.