Digital soil mapping based site-specific nutrient management in a sugarcane field in Burdekin

Abstract In the highly productive Burdekin valley, the soil is heterogeneous. To optimise productivity, for irrigated crops such as sugarcane, essential nutrients such as calcium (Ca) and magnesium (Mg) need to be applied. To assist sugarcane farmers, nutrient management guidelines for these mineral elements have been recommended (i.e. Six-Easy-Steps) based on the exchangeable (Exch.) Ca and Mg. However, these are applied using ‘one size fits all’ approach, which lead to fertiliser use inefficiencies, given the alluvial nature of the landscape; where sandy soil characterises the ephemeral creeks (Chromosols) and clay soil types (Sodosols and Vertosols) of the plains. Herein we used and compared regression kriging (RK) and linear mixed models (LMM) to create digital soil maps (DSM). We also compared the efficacy of various proximal sensed γ-ray spectrometry and EM data. Using measures of bias (mean error - ME) and precision (root mean squared error - RMSE) of predictions, as well as the Lin's concordance, we determine which model and data was most useful using a leave-one-out cross-validation. The results for Exch. Ca showed that while RK approach had a strong concordance (0.81), unbiased (0.01) and precise (0.06), the LMM outperformed RK, given the better concordance (0.87) and bias (0.00). The MSPE of the final LMM DSM (0.01) was also smaller compared to the RK DSM (0.0123). Moreover, both DSM was superior to the traditional Soil Order map (0.0171). The results for Exch. Mg were equivalent. We also conclude, that while good concordance was achieved using either γ-ray (Lin's = 0.79) or EM (0.83) for Exch. Ca, using both proximal sensors was optimal. The results showed that pedometric approach can be used to generate a DSM of Exch. Ca and Mg in a sugarcane field. In terms of soil use and management, the infertile sandy textured soil associated with the prior stream channels and characterised by small Exch. Ca (