Application of Compositional Kriging and Several Types of Log-ratio Co-Kriging in a Case Study of Large-Scale Soil Texture Mapping

Objective To cope with compositional characteristics of soil texture data, compositional kriging and log-ratio cokriging were tested in a case study of large-scale soil texture mapping.

Method Based on soil texture data of 1863 samples collected in Guangdong Province, ordinary kriging, cokriging, compositional kriging method and four types of log-ratio cokriging, were used to map soil texture, and accuracies of the methods were evaluated.

Result The maps produced using ordinary kriging had only 23% of the region met the basic requirement of constant sum, and their accuracies were the worst. Cokriging indirectly solved the problem of constant sum for soil texture data, and generated good accuracies. Compositional kriging did not only directly solve the problem of constant sum for soil texture data, but also generated the best accuracies. The log-ratio cokriging could also directly solve the problem of constant sum, but generated accuracies slightly better than ordinary kriging and lower than the other methods. Among the log-ratio cokriging methods, the isometric logarithmic ratio transformation performed the best.

Conclusion Compositional kriging performed the best in large-scale soil texture mapping, followed by cokriging and the log-ratio cokriging combined with isometric logarithmic ratio transformation. In the future, large-scale soil texture mapping based on geostatistical methods still needs to further explore the use of environmental covariates to improve the accuracy, and consider the influence of sampling and other factors.