Vis– NIR Spectroscopy for Predicting Soil Active Carbon and Mineralizable Nitrogen Across Diverse Agroecosystems

This study evaluates the potential of visible and near‐infrared spectroscopy (Vis–NIRS) to predict two key indicators of soil biological activity, permanganate oxidizable carbon (POxC) and potentially mineralizable nitrogen (PMN), across a large and diverse regional soil dataset. A total of 1171 soil samples, collected between 2017 and 2024 from agricultural sites and gardens in Wallonia (Belgium), were analysed using both fresh and dry soil spectra in the 400–2500 nm range. Global and local partial least squares regression models were developed for each parameter using standardized spectral pre‐processing and validated on independent subsets. For POxC, both global and local models showed strong predictive performance (CCC = 0.90–0.94; SEP = 101.6–120.8 mg kg −1 ; RPIQ = 2.8–3.3), with the best dry local model reaching CCC = 0.960, SEP = 80.9 mg kg −1 and RPIQ = 4.3 after discarding spectral outliers based on Mahalanobis and Neighbor distances (GH ≤ 3, NH ≤ 1). PMN prediction was more challenging, with moderate performance (CCC ≤ 0.84; SEP ≥ 58.7 kg N ha −1 ; RPIQ ≤ 2.3), and local models consistently outperformed global ones. Adding the visible spectral domain did not systematically improve model accuracy for either parameter, indicating that NIR‐only acquisition may suffice. Comparing model error with the uncertainty of the reference methods showed that the two indicators approached the precision limits of their laboratory assays. In particular, a substantial part of the moderate PMN performance reflects the limited repeatability of the incubation reference method rather than the spectral model alone. The multi‐year, multi‐site design strengthens a regional spectral library but also introduces spatial and temporal variability that limits transferability, so predictions are most reliable for well‐represented soils. Overall, these findings support the integration of NIRS to quantify active carbon pools such as POxC and to screen PMN semi‐quantitatively, providing a rapid, cost‐effective complement to wet chemistry. Its progressive integration into routine, quality‐controlled soil analysis will benefit from expanding and diversifying regional spectral libraries.