This study evaluated the potential of near-infrared spectroscopy (NIRS) combined with multivariate analysis to predict soil nitrogen (N) content in both wetland and dryland systems of Java Island, Indonesia. A total of 145 soil samples were collected across four provinces and analysed using a Frontier FT-NIR spectrometer in the 750-2500 nm range. Spectral data were pre-processed using several methods, and partial least squares regression (PLSR) models were calibrated and validated against reference Kjeldahl measurements. Among all approaches, asymmetric least-squares (ALS) baseline correction coupled with PLSR achieved the best predictive performance, with R & sup2; = 0.882, RMSE = 0.0401% N, and RPD = 2.84 on the independent validation set. Characteristic wavelengths were identified around 1400, 1900, and 2200-2330 nm, corresponding to overtone and combination bands of N-H and O-H vibrations. The results confirm that NIRS offers a rapid, non-destructive, and cost-effective alternative for soil nitrogen assessment under diverse agroecosystems. This approach can support precision nutrient management in tropical smallholder farming systems and provides a foundation for developing portable, field-ready detection tools.