Accounting for analytical and proximal soil sensing errors in digital soil mapping

Digital soil mapping (DSM) approaches provide information by utilising the relationship between properties and environmental variables. Calibration of DSM models requires measurements that may often have substantial measurement errors which propagate to outputs need be accounted for. This study applied a geostatistical-based approach incorporates error variances in covariance structure spatial model, weights accordance with their accuracies assesses effects on outputs. The method was Western Cameroon, where samples from 480 locations were collected analysed for pH, clay organic carbon (SOC) using conventional mid-infrared spectroscopy methods. Variogram parameters regression coefficients estimated residual maximum likelihood under two scenarios: without taking into account. Performance scenarios compared validation metrics obtained three types cross-validation. Acknowledging impacted influenced variogram reducing nugget sill variance properties. Validation including mean error, root square model efficiency coefficient quite similar both scenarios, but prediction uncertainties more realistically quantified account errors, as indicated accuracy plots. There relatively small absolute differences predicted values up 0.1 1.6% 2 g/kg SOC scenarios. We emphasised incorporating improve uncertainty quantification, particularly when applying estimating Further development is extension non-linear machine learning Highlights Errors are usually not affect results. Measurement incorporated geostatistical Quantifying allows weigh accuracy. Accounting better