Abstract: In most areas there exist some soil field samples “( existing samples data”) which were accumulated through two historical soil surveys and/or specific field studies. The accuracy of soil property map predicted based on this kind of samples is lower, because the size of these samples is limited and the global representativeness is poor. However, these samples are valuable resource because they do capture the relationship between soil and environment conditions. This paper presents a stepwise and effective method to design additional samples by integrating existing samples. The aim of the method is to design as few additional samples as possible to complement the global representativeness of the existing samples and improve the accuracy of predictive soil mapping. The detail process is as follows. First, we determine the spatial extent which can be represented by the existing samples. Second, each location (grid) can be considered as a candidate for additional sampling in the area which the existing samples do not cover and the area which can be covered by each candidate was then calculated. Third, the location which can represent the largest additional area was chosen as the first additional sample. Finally, the above procedure was iterated until the new sample set covered the whole study area. This integrative sampling scheme can not only determine the number and the locations of additional samples, but also give the order of the samples. It provides samplers with the important information on how many points can be sampled reasonably when the available resources are limited.

Key words: existing samples, global representativeness, individual representativeness, prediction uncertainty, stepwise additional sampling designing scheme