Applying spatial models to grid-based data relies on strong hypotheses, notably, stationarity. We propose a method to explore the hypothesis of stationarity and pinpoint parts of the grid-based data, where the hypothesis is violated.

We apply these methods to field trials in plant breeding, where grid-based data corresponds to measurements and coordinates of plant varieties in the field. The hypothesis of stationarity may be violated if spatial patterns exist in the residuals in certain areas of the field (i.e., residual blocks), due to environmental effects.

Our methods represent grid-based data in a tree graph using quad and binary tree. Residual blocks are then recovered using tree segmentation of the tree indexed data.