Determination of the optimal mathematical model, sample size, digital data and transect spacing to map CEC (Cation exchange capacity) in a sugarcane field

Abstract

The cation exchange capacity (CEC) is an important property because it influences soil structural stability, nutrient availability, pH and reaction to fertilisers. To assist Australian sugarcane farmers balance sugarcane-yield and minimise fertiliser run-off, the six-easy-steps nutrient management guidelines were developed. In this research we compare and contrast various aspects of digital soil mapping (DSM) of topsoil (0–0.3 m) and subsoil (0.6–0.9 m) CEC, including: choice of model (i.e. linear mixed model – LMM, regression kriging – RK, Cubist, random forest – RF and support vector machine – SVM), digital data (i.e. gamma-ray (γ-ray) spectrometry and apparent conductivity (ECa)) in combination or independent, transect spacing (i.e. 5, 10, 20, 30, 40, 60, 80 m) and number of samples (i.e. 120, 110,…, 10) for calibration. We test these using a validation (i.e. 40) data set. The comparisons were evaluated considering the agreement between measured and predicted CEC using Lin’s concordance correlation coefficient (LCCC) and accuracy using root mean square error (RMSE). The results indicate that for the DSM of topsoil CEC, the Cubist with an intermediate number of calibration samples (i.e. 80) using in combination both γ-ray and ECa was optimal in terms of agreement (LCCC = 0.79). For subsoil, a smaller number (i.e. 30) of soil samples for calibration was required to achieve good agreement (LCCC = 0.89). In terms of accuracy, the accuracy (RMSE = 5.42 cmol(+)/kg) of subsoil CEC was satisfactory, as it was less than half standard deviation (SD) (7.55 cmol(+)/kg) of measured CEC. While not the same for topsoil CEC, the accuracy (RMSE = 1.93 cmol(+)/kg) was not as satisfactory as it was over half measured topsoil CEC SD (1.68 cmol(+)/kg). The results also showed that while γ-ray alone was superior to ECa data for prediction, better results were achieved when both digital data were used in combination. In terms of a suitable transect spacing for collection of digital data to predict topsoil CEC, the small transects spacing (i.e. 5 m) was recommended. For subsoil prediction, larger transect spacing may still be appropriate (i.e. 5–60 m). The DSM approach overall enabled topsoil and subsoil prediction of CEC with good accuracy and small residuals, particularly at large calibration data sets (i.e. >80). The final DSM of topsoil and subsoil CEC therefore do allow for the implementation of the six-easy-steps nutrient management guidelines for lime. Specifically, the larger northern half (22.74 ha) requires a small (<2.5 t/ha) application rate with the southern half requiring intermediate (3–5 t/ha) to large (5 t/ha) rates.