Machine learning UV-vis spectroscopy for total nitrogen and phosphorus measurement in wastewater with wavelength-wise...

High-quality total nitrogen (TN) and total phosphorus (TP) data are essential for wastewater management and aquatic ecosystem protection, yet routine determinations rely on labor- and reagent-intensive wet chemistry. Ultraviolet-visible (UV-Vis) spectroscopy offers a rapid, low-consumable alternative, but TN/TP quantification in heterogeneous wastewater remains challenging because they are composite indices with weak, dispersed spectral features and strong matrix effects. Here we introduce a task-specific machine learning framework for TN/TP measurement via UV-Vis. TN is modeled with eXtreme Gradient Boosting (XGB) on pretreated spectra to capture nonlinearities with built-in regularization; TP, whose spectral cues are weaker and more dispersed, is addressed by a 1D-CNN-XGB hybrid pipeline, where the CNN distills shift-tolerant local motifs and reduces collinearity before nonlinear regression by XGB. The proposed method achieves R² = 0.9691 (RPD = 5.94) for TN and R² = 0.8738 (RPD = 3.43) for TP. Wavelength-resolved attributions with latent back-projection reveal distinct spectral drivers: TN is dominated by compact deep-UV motifs with a secondary UV band and minor near-UV features consistent with matrix/background variability, whereas TP follows a distributed, shape-based multi-band UV rule captured by the CNN. This framework outlines an effective and interpretable UV-Vis approach for TN/TP measurement in wastewater and may enable rapid, in-situ, low-consumable monitoring workflow towards field deployment.