Applying provincially derived pedotransfer function and digital soil map predictions of nitrogen indices at the field scale.

There is a pressing need to provide crop producers with tools for estimating soil nitrogen (N) mineralization at the local level and to support variable N rate technologies. With provincial scale datasets being used to train predictive models, there is an opportunity to observe how these predictions perform at the field-scale; especially in fields below provincial soil quality benchmarks. This study examined how provincially derived decision support tools (DSTs) for estimating growing season N mineralization performed at the field-scale for informing N fertilizer recommendations. DSTs included a 30m resolution digital soil map (DSM), and novel pedotransfer Functions (PTFs) developed using machine learning and predictor variables from a provincial database. Six agricultural fields were sampled for comparison with DST predictions, as well as to create novel infield maps (5m resolution) of N indices. DSTs were successful at placing field N mineralization potential in context to provincial averages, achieving PTF predictions with a concordance (CCC) as high as 0.91 in comparison to direct soil measures, and 71% to 100% of these direct measures within the 90% confidence interval of DSM predictions. Uncertainty decreased substantially using infield data, and infield maps of N indices achieved a CCC as high as 0.83. Field-scale predictors for N mineralization were different from DSM results, with relief variables having high importance. This study demonstrated how provincially derived PTF and DSMs of N indices could be used to inform N fertilizer recommendations. Development of field-scale N maps also introduces a powerful tool for guiding variable rate N applications.