Near infrared spectral analysis of cattle slurries from Lombardy (Northern Italy) breeding farms

Composition of animal slurries is a basic knowledge for their management within a fertilization plan in order to define their right dose, to optimize supplemental addition of nutrients to soil by mineral fertilizers and to minimize adverse effects on the environment, overall nitrate release to groundwater. In most farms nutrients releases by slurries are calculated on the basis of tabulated mean values so that significant estimation mistakes are made. Farmers aren’t attracted by the conventional analyses, because of their high cost and the long time to get a response. Furthermore, slurries have a low economical value and their composition changes during time, so that analytical results have a short life. Rapid and cheap analytical methods, whether less precise than conventional methods, appear to be an useful way to improve slurries management. The ability of NIR spectroscopy to estimate cow slurries properties was tested on a set of 101 samples collected during spring 2003 from cow breeding all over Lombardy, Northern Italy. Before analyses samples were homogenized at 10-50 um particle size. The analysed slurry variables were: dry matter (DM), ashes (A), Kjeldahl nitrogen (KN), ammonium nitrogen (NH4+-N), pH, electrical conductivity (EC), total phosphorous (TP), total potassium (TK), total carbon (TC) and not volatile nitrogen (NVN). KN, NH4+-N, pH and EC were determined on the fresh, liquid, samples while the other variables were determined on samples dried at 105°C and milled at 0.5 mm. The ranges of variables, expressed as g/kg of fresh weight, were: 16-162 for DM, 6-45 for A, 0.9-7.3 for KN, 0.3-2.5 for NH4+-N, 0.1-1.4 for TP, 1.1-4.4 for TK, 4.5-70.3 for TC, 0.4-3.6 for NVN, 5.5-31.8 mS/cm for EC and 6.6-8.1 for pH. Homogenized samples, heated at 25°C, were scanned (from 1100 to 2498 nm, 2nm resolution) with cuvette refilling twice using a horizontally positioned Foss Tecator NIRS5000 spectrometer. Spectra were averaged for each sample, corrected for scattering, smoothed and 2nd order derived. Outliers detection (10 samples) was conducted by Principal Components Analysis (PCA), selecting samples by Q and Hotelling’s T2 tests. Spectra population was divided by Mahalanobis distance of neighbours samples in calibration and validation sets. Regressions against measured variables were determined by Partial Least Squares (PLS) analysis, looking for the smallest calibration population needed to obtain the best results. Some variables were predicted fairly well by NIRS in terms of RMSEP and prediction R2 (DM: 0.95, 0.95; A: 0.37, 0.74; KN: 0.32, 0.90; NH4+-N: 0.18, 0.83; TC: 3.74, 0.92; NVN: 0.21, 0.94; TP: 0.13, 0.78) whereas pH, EC and K didn’t show great correlation with spectral data. By applying genetic algorithms (GA) to variables and spectra, was selected a subset of wavelengths able to detect each considered variable, without losing prediction accuracy, in order to individuate a smaller group of informative bands (15-30), useful to be used on a cheapest and portable multi-filter apparatus for on-field content determinations.