Predicting Soil Nitrogen Mineralization Rate in Agricultural Soils Using Near Infrared Reflectance Spectroscopy ( NIRS ): A Laboratory Incubation Study

For spring‐sown crops such as maize ( Zea mays L.), the extended growing season enables a substantial contribution of organically derived nitrogen (N), particularly from soil organic matter (SOM) mineralization. Consequently, the soil N mineralization rate (SMR) is a key determinant of maize N uptake and yield formation. Year‐to‐year and spatial variability in climate, soil properties, and management practices introduce considerable fluctuations in SMR, making it a major source of uncertainty in site‐specific N management. Near‐infrared reflectance spectroscopy (NIRS) offers a rapid, cost‐efficient and scalable tool with strong potential for improving SMR estimation and, consequently, site‐specific N management in maize production. In this study, 268 soil samples were collected across Germany with varying SOM content and incubated for 105 days to simulate the maize growing period, allowing laboratory quantification of SMR. Simultaneously, NIRS spectra (1100–2498 nm) were collected, preprocessed, and SMR was modelled using Partial Least Squares Regression (PLSR) and different machine learning (ML) algorithms. SMR was modelled more accurately by ML approaches, with non‐linear models such as the stacked ensemble model ( R 2 = 0.72, RMSE = 0.21 kg N ha −1 day −1 ) or random forest ( R 2 = 0.70, RMSE = 0.20 kg N ha −1 day −1 ) outperforming PLSR ( R 2 = 0.55, RMSE = 0.33 kg N ha −1 day −1 ). Incorporating ancillary soil and management data did not improve performance. NIRS‐ML modelling improved SMR prediction by up to 36%, with slight underestimation of laboratory‐measured SMR values due to indirect spectral relationship with NIRS. Nevertheless, the approach shows great potential as a rapid and scalable tool for SMR estimation, warranting further model refinement and validation across diverse environments.