In-situ recommendation of alternative soil samples during field sampling based on environmental similarity

Field sampling is an essential step for digital soil mapping and various sampling strategies have been designed for achieving desirable mapping results. Some unpredictable complex circumstances in the field, however, often prevent some samples from being collected based on the pre-designed sampling strategies. Such circumstances include inaccessibility of some locations, change of land surface types, and heavily-disturbed soil at some locations, among others. This may result in the missing of some essential samples, which could impact the quality of digital soil mapping. Previous studies have attempted to design alternative samples for the selected ones beforehand to address this issue. It cannot solve the problem completely as those pre-designed alternative samples could also be inaccessible. In this paper, we propose a dynamic method to recommend alternative samples for those unavailable samples in the field. The identification of alternative samples is based on the environmental similarity between an unavailable soil sample and its alternative candidates, as well as the spatial accessibility of these candidates. For the convenience of fieldwork, the proposed method was implemented to be a mobile application on the Android platform. A simulated soil sampling study in Xuancheng county, Anhui province of China was used to evaluate its performance. From a sample set for the study are, 30 samples were assumed to be inaccessible. For each of them, an alternative soil sample from the set was recommended using the proposed method. A deviation analysis of silt and sand content at the depth of 20 ~ 40 cm between the soil samples and their alternatives shows that the deviation on silt content is less than 20% for half of the soil samples. A larger deviation on sand content might be attribute to the limited alternative candidates in this virtual experiment. In a second experiment, we randomly selected a number of existing soil samples and replaced them with their corresponding alternative soil samples. This created 1000 hybrid sample sets. Each hybrid sample set was then used for digital soil mapping with iPSM. An evaluation using 59 independent soil samples indicated that the RMSE and MAE with the hybrid sample sets were close to that with the original sample set. The proposed method proved to be able to recommend effective alternative samples for those unavailable samples in the field.