关键词: 成分数据, 空间插值, 零值替换, 对数比转换, 等角对数比, 土壤质地

Abstract: 【Objective】Spatial interpolation of compositional data needs to meet the four requirements including non-negativity, constant sum, error minimization and unbiased estimation. Soil texture is one of the compositional data. Taking soil texture data of Manas River oasis in Xinjiang as an example, the paper studied the log-ratio transformation approaches on spatial interpolation of compositional data.【Method】First, soil particles content of the soil samples were transformed by using the additive, centered and isometric log-ratio transformation approaches, respectively, then the ordinary kriging method was employed to perform the spatial interpolation. A zero replacement method was introduced to avoid the log-transformation of zero in soil particle composition.【Result】The constant sum of soil particles conent did not change after zero replacement. The kriging based on log-ratio transformation fulfilled the four requirements of compositional data interpolation while the kriging to soil particles separately did not. The kriging based on isometric log-ratio achieved the best results among the three log-ratio approaches, while no obvious differences were found among these approaches.【Conclusion】Zero replacement avoided the log-transformation of zero on the premise of unchanging the requirement of constant sum. The non-negativity, constant sum, error minimization and unbiased estimation of the four requirements of compositional data interpolation could be satisfied by ordinary kriging based on log-ratio transformation.

Key words: compositional data, spatial interpolation, zero replacement, log-ratio transformation, isometric log-ratio, soil texture