Developing a pedotransfer function for the prediction of nitrogen mineralization in the agricultural soils of Quebec

Effective nitrogen (N) management is crucial for maximizing crop yields while minimizing environmental impacts. In the agricultural systems of Quebec, soil organic matter mineralization supplies a significant portion of crop N demand, but direct quantification is challenging and costly. This study utilized zero-N trial data to evaluate a two-pool zero-plus first-order kinetic model for predicting growing season N mineralization (GSNM) based on total N (TN) and potentially mineralizable N (PMN). Additionally, machine learning-based pedotransfer functions (PTFs) were developed to predict TN, PMN, and GSNM from easily measurable soil properties using a large soil health dataset ( n = 3117). The kinetic model showed strong agreement between predicted and observed soil N supply, especially with the inclusion of deeper soil layers and early-season mineralization estimates. Recursive feature elimination identified total carbon (TC) and clay as the best predictors for TN, yielding a Lin's concordance correlation coefficient (CCC) of 0.93 and a coefficient of determination ( R 2 ) of 0.86, while soil respiration (SR) and pH best predicted PMN (CCC = 0.89, R 2 = 0.80). For GSNM, SR, TC, and pH were the top predictors (CCC = 0.91, R 2 = 0.83). The developed PTFs provide a practical framework for estimating soil N-pools where direct data is unavailable, ultimately improving site-specific N management decisions. These tools support more efficient fertilizer use and minimize environmental losses. Future research should focus on integrating soil management practices into the development of PTFs and considering spatial and landscape variability through digital soil mapping.