In the paper, geographic phenomena varying continuously over geographical space (e.g., air pollution) are called geofields and queries concerning them are called geofield queries. Theoretical and experimental research showed that usually the most time consuming operation in geofield queries processing is generating the PNR representation of qualitative geofields. As the PNR representation is based on ordering a quadtree by Peano N space-filling curve, a central role in an implementation of the operator is played by square classifiers making decisions whether to assign classified quadrants to one of the intervals given in the query or split them and classify their children. Presented classifiers make the decision computing geofields model values in predefined points or in points chosen using a geofield variability model. The variability is predicted using conditional quantile functions,modeling dependence of change of geofield value module between points on the distance between the points.