Near-infrared spectroscopy as a sensing technology for compositional analysis of cattle manure
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3172212Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
- Master’s theses (BioVit) [414]
Sammendrag
The feasibility of laboratory near infrared spectroscopy (NIRS) for determination of dry matter (DM) and fresh weights of total nitrogen (N), ammonium-N, organic N, phosphorus (P) and potassium (K) in cattle manure was investigated. Calibrations for DM concentrations of organic N, P and K were also evaluated. Forty-eight manure samples were collected from Norwegian cattle farms and measured in reflectance in the range 1990 to 15500 cm-1. Calibrations were developed by partial least square regression using spectra from 4000-9000 cm-1 and validated by two independent test sets. By cross-validation, the contents of DM, total N, ammonium-N and fresh weights of organic N, P and K were predicted with R2 values of 0.95, 0.91, 0.87, 0.91, 0.82 and 0.78, respectively. The ratio between the analyte standard deviation and the root mean square error of cross-validation (RPD) was 4.82, 3.56, 2.80, 3.61, 2.40 and 2.30 for DM, total N, ammonium-N and fresh weights of organic N, P and K, respectively. Dry matter concentrations of organic N, P and K were poorly predicted by cross-validation with R2 values of 0.23-0.73 and RPD ratios of 1.33-2.00.
Validation by an independent test set gave predictions of DM, total N, ammonium-N and fresh weights of organic N, P and K with R2 values of 0.86, 0.79, 0.93, 0.42, 0.76 and 0.76, respectively. The corresponding RPD values were 2.35, 1.91, 3.44, 1.05, 2.00 and 1.81, respectively. Removal of a shared sample outlier in the prediction of DM, total N and K (fresh weights) increased the R2 and RPD values to 0.95 and 3.95, 0.96 and 3.83, 0.90 and 3.08, respectively. We concluded that laboratory NIRS has potential to predict DM and fresh weights of total N, ammonium-N and K. Fresh weight predictions of P can be useful for approximate determinations, while predictions of organic N, P and K on a DM basis are not recommended. The feasibility of predicting fresh weight organic N needs further investigation.
The effect of sample temperature (5 and 20⁰C) on prediction performances was investigated by using calibration models developed at 20⁰C to predict samples measured at both temperatures. Prediction errors obtained from samples measured at ~20⁰C were overall lower than the corresponding measured at ~5⁰C, and significantly (p<0.05) lower in the prediction of DM and K. According to these results, an effect of sample temperature should be considered during calibration development and future predictions.