Comparing mapping approaches at subcatchment scale in northern Thailand with emphasis on the Maximum Likelihood approach

Among the mapping approaches investigated only the Maximum Likelihood approach and the Classification Tree method are suitable for upscaling procedures. Both approaches require a calibration area which can only be mapped by transects or grid-based random sampling. These approaches are more efficient in the case of known ex-ante soil variability, so that the distance for transects, grid cell size, and amount of sampling points can be chosen in a proper way. It is useful to apply local knowledge in order to elevate ex-ante knowledge of soil information. In one pilot study local knowledge-based-soil sampling could reduce the amount of sampling points within a calibration area to less than the half used for transects and randomized mapping. The Maximum Likelihood approach applied afterwards provided an accuracy of at least 70% and hence is as accurate as the Classification Tree method. It seems to be most useful to combine the different approaches.