[Rapid evaluation of primary nutrients during plant-field chicken manure composting using near-infrared reflectance...

The present study explored the efficiency of quantitative analysis for the contents of total nitrogen (TN), total phosphorus (TP) and total potassium (TK) during chicken manure composting with chrysanthemum residue in a plant field using a Thermo Nicolet Antaris near-infrared reflectance spectral apparatus equipped with InGaAs detectors (Thermo Nicolet Corporation, Chicago, USA). The samples used in this study were collected from different positions during composting and were scanned in polyethylene bags at 2 cm(-1) interval from 10 000 to 4 000 cm(-1) with 32 co-added scans. Regression models were developed using spectral data and reference data by partial least square (PLS). In order to enhance chemical information and reduce data systemic noise, different data preprocessing methods such as smoothing, 1st and 2nd derivative, standard normal variety (SNV) and multiplicative scatter correction (MSC) were tested. The optimum preprocessing method was selected with the lowest root mean square error of cross-validation (RMSECV). Outliers were removed on the basis of being labeled as compositional outliers by the criteria that the predicted-actual difference for the sample was three standard deviations from the mean difference. According to the concentration gradient of each parameter, all samples were divided into a calibration set (3/4 samples) and a validation set (1/4 samples). Leave-one-out cross validation was performed to avoid over-fitting on the calibration sets. Based on the values of determination coefficient (R2) and relative prediction deviation (RPD) in validation sets, the prediction results were evaluated as excellent for TP and TK, and approximate for TN. Hereinto, R2 and RPD values were greater than 0.82 and 2.0 for TN, and greater than 0.90 and 3.0 for TP and TK, respectively.