A quantitative method combining data-level fusion of ultraviolet (UV) and fluorescence spectroscopy with opti-mized machine learning is proposed for online monitoring of total nitrogen (TN) in water. UV (200–400 nm) and fluorescence (425–800 nm) spectra were synchronously acquired with a custom dual light source system and fused after scattering correction, normalization, and denoising. Reference TN concentrations were determined by the na-tional standard method (HJ?636-2012). Lasso regression selected key wavelengths, and an Elastic Net model with Bayesian optimized hyperparameters was built. On 298 actual water samples, the model achieved a test set R2 of 0.9548 and an RMSE of 0.0784?mg?L?1; all relative prediction errors were below 20%, meeting the online monitoring specification (HJ?355-2019). Compared with single spectrum and other feature selection approaches, the proposed method shows significantly better accuracy, stability, and generalizability, offering a reliable solution for real time TN monitoring in complex water matrices.