Comparation of Spatial Interpolation Methods on Slowly Available Potassium in Soils

Abstract

Soil nutrients are important parameters for maintaining the plant growth. Conventional in-situ methods are usually labor-intensive, low efficiency and time-consuming. Conversely, spatial interpolation is an extremely efficient method to derive the spatial distribution trend from discrete ground truth data. In this study, a total of 402 sample data of soil nutrients was collected in the wheat fields of Tongzhou District, Beijing, China. Slowly available potassium (SAK) was used as the interpolation factor. Four methods were comparatively employed to evaluate the availability and accuracy, specifically including inverse distance weighted (IDW), ordinary kriging (OK), spline function (Spline) and Trend interpolation. There are three models of OK, spherical model, exponential model and Gaussian model. Cross-validation was performed to compare the interpolation accuracy using the mean error (ME) and root mean square error (RMSE). The results show that OK is superior to the other methods, in which the exponential model is the best. In comparison with spherical function and Gaussian function, exponential model has the best performance with the ME and RMSE of 176.7084 and -1.1283, respectively.