Imputation strategy in spatial data

In area frame surveys a mixed approach consisting in the observation of the surroundings of a not reachable point combined with orthophoto interpretation needs to be used in the locations that are particularly remote and difficult to reach. In this situation a simplified nomenclature has to be applied for some land cover categories due to the difficulties in properly distinguishing among specific classes. In the estimation phase the resulting observations can be considered affected by a sort of partial non response phenomenon. Classification of land cover indeed is available only at aggregated level. A donor based methodology is proposed to impute this missing detailed information. Assuming that neighbouring points are affected by spatial autocorrelation, potential sets of donors are identified among points within different distance thresholds. Capability of the method of correctly imputing the missing information is discussed and its robustness assessed in terms of two different costfunctions both based on the maximum distance observed among potential donors and recipient points.