Sampling for High-Resolution Soil Mapping

When doing sensing for high-resolution soil mapping, one has to decide on the disposition of the sensor, which is a special case of spatial sampling. To optimise the pattern of measurements, a cost model and a quality model are proposed. The quality model reflects the coverage of the geographic space, and this is illustrated with some practical experiments. Optimisation of sensing patterns is worked out for two different types of sensing equipment. If the sensor variable differs from the target (management or decision) variable, then a model is needed to predict the target variable from the ancillary data. So in that case, one also has to decide how and where to sample for calibration data. This ‘calibration sampling’ differs from ‘sensor sampling’, as now coverage of the predictor space rather than the geographic space is important. In addition, the handling of extremes is an issue here. Existing methods for calibration sampling are reviewed and a suggestion is made for a new approach, based on fuzzy cluster analysis, which might avoid some of the shortcomings of existing methods.