A review of spatial prediction methods for soil particle-size fractions [土壤粒径空间预测方法研究综述]

作者：Shi W.; Zhang M.

来源：Acta Geographica Sinica, 2022, 77(11): 2890-2901.

DOI：10.11821/dlxb202211013

• Geostatistics
• High accuracy surface modeling
• Machine learning
• Soil interpolation
• Spatial accuracy

Soil particle-size fractions (PSFs), including sand, silt, and clay, are key parameters for land-surface process simulation and ecosystem service evaluation. More accurate interpolation of soil PSFs can help better understand the simulation of the above models. As compositional data, soil PSFs have special demands of the constant sum (1 or 100%) in the interpolation process, and the spatial distribution accuracy is mostly affected by the performance of spatial prediction methods. Here, we provided a framework for the spatial prediction of soil PSFs, and reviewed a series of methods in the steps of this framework, including methods of log-ratio transformation of soil PSFs (additive log-ratio, centered log-ratio, symmetry log-ratio, and isometric log-ratio methods), spatial interpolators of soil PSFs (geostatistical methods, regression models, and machine learning models), validation methods (probability sampling, data splitting, and cross-validation) and indices for accuracy assessments in soil PSF interpolation and soil texture classification (rank correlation coefficient, mean error, root mean square error, mean absolute error, coefficient of determination, Aitchison distance, standardized residual sum of squares, overall accuracy, Kappa coefficient, and precision-recall curve) and uncertainty analysis (prediction interval, confidence interval, standard deviation, and confusion index). In addition, we summarized several ways to improve the prediction accuracy of soil PSF, such as normalizing the data distributions through effective data transformation, choosing suitable prediction methods based on the data distribution characteristics, improving mapping accuracy and distribution reasonability through the combination of auxiliary data, improving interpolation accuracy through hybrid models or joint modeling for multi-components. Finally, we proposed the future research fields of the spatial prediction methods of soil PSFs, including considering the principles and mechanisms of data transformation, developing joint simulation models and high accuracy surface modeling methods for multi-components, and combining soil particle size curves with stochastic simulations. Our review highlights the importance of spatial prediction methods for soil PSFs, and also provides a clear framework for improving the performance of these methods for other researchers in this field.

• 单位
  中国科学院; 资源与环境信息系统国家重点实验室; 中国科学院大学