It is widely accepted that spatial or spatio-temporal dependencies must be acknowledged appropriately in data that are spatially and/or temporally aligned. However most of the respective regression models still rely on the assumption that the dependence structure does not vary over space, i.e. the resulting models are spatially stationary. While this simplification considerably facilitates estimation since information can be borrowed across different observation points with the same configuration in space, it is typically too simplistic in forest stands. Nonstationarity can, for example, arise due to changing environmental conditions or due to constraints in the spatial domain where, for example, specific site characteristics may act as boundaries modifying the common spatial dependence patterns. The general research objective is then to develop flexible and practical ways of allowing for nonstationarity in spatial components of semiparametric regression models. Hence, points that locally have similar configurations but are placed at different locations in the observation domain are allowed to have distinct dependence patterns.
