Low-Cost NIR Spectroscopy Versus NMR Spectroscopy for Liquid Manure Characterization

Accurate characterization of liquid manure properties, such as dry matter (DM), total nitrogen (TN), ammonium nitrogen (NH₄-N), and total phosphorus (TP), is essential for effective nutrient management in agriculture. This study investigates the use of near-infrared spectroscopy (NIRS) within the 941-1671 nm range, combined with advanced pre-processing and machine learning techniques to accurately predict the liquid manure properties. The predictive accuracy of NIRS was assessed by comparison with nuclear magnetic resonance (NMR) spectroscopy as a benchmark method. A number of 51 liquid manure samples were analyzed in the laboratory for the reference manure properties and scanned with NIRS and NMR. The NIR data underwent spectral pre-processing, which included two- and three-band index transformations and feature selection. Partial least squares regression (PLSR) and LASSO regression were employed to develop calibration models. According to the results, using cohort-tuned models, NIRS showed fair predictive accuracy for DM (R² = 0.78, RPD = 2.15) compared to factory-calibrated NMR (R² = 0.68, RPD = 0.81). Factory-calibrated NMR outperformed for chemical properties, with R² (RPD) of 0.89 (1.74) for TN, 0.97 (5.70) for NH₄-N, and 0.95 (2.64) for TP, versus NIRS's 0.66 (1.68), 0.84 (2.45), and 0.84 (2.51), respectively. In this study with 51 samples, two- and three-band indices significantly enhanced NIRS performance compared to raw data, with R² increases of 34%, 57%, 25%, and 33% for DM, TN, NH₄-N, and TP, respectively. Feature selection efficiently reduced NIR spectral dimensionality without compromising the prediction accuracy. This study highlights NIRS's potential as a portable tool for on-site manure characterization, with NMR providing superior laboratory validation, offering complementary approaches for nutrient management.