Prediction of available soil nitrogen using Vis–NIR and MIR spectroscopy coupled with chemometric and machine learning approaches in inceptisols of India

Soil nitrogen is a fundamental indicator of soil fertility and crop productivity, and its rapid assessment is essential for efficient land management and sustainable agricultural practices. This study investigates the potential of Vis–NIR and MIR spectroscopy for predicting available soil nitrogen collected from the intensively cultivated Indo-Gangetic Plains of India. Soil samples were analyzed using the standard laboratory method and corresponding spectral signatures were processed using multiple preprocessing techniques, among which Standard Normal Variate (SNV) showed the greatest enhancement in model performance. Four regression algorithms Partial Least Squares Regression (PLSR), Random Forest (RF), Support Vector Regression (SVR) and Multivariate Adaptive Regression Splines (MARS) were calibrated for both the Vis–NIR and MIR spectral domains. Model evaluation demonstrated that MIR spectroscopy combined with PLSR yielded the highest prediction accuracy (R² = 0.84, RMSE = 9.95, RPD = 2.70), outperforming all Vis-NIR models and other machine-learning approaches. RF, MARS and SVR exhibited comparatively lower performance, with RPD values between 1.06 and 1.66. Based on the results, MIR spectroscopy combined with the PLSR algorithm is recommended as a rapid and reliable approach for estimating available nitrogen in Inceptisol soils of the Indo-Gangetic Plains. Adopting this technique can enhance nitrogen management while reducing time, cost and waste for sustainable soil fertility monitoring.