Spectra of NIR of a total 104 commercial organic fertilizers. doi:10.1371/journal.pone.0088279.g002 

Context in source publication

Context 1

... minimum, maximum, mean, and standard deviation (SD) of the quality indices of commercial organic fertilizers in the calibration sets and validation sets are shown in Table 1. This data set represented a wide range of compositions. For all the 104 sampled fertilizers, the mean value and SD of the quality indices were 3.95 6 1.70% of moisture, 357.63 6 76.90 g TOM/kg, 19.52 6 16.99 g TN/kg, 4.87 6 9.54 g WSOC/kg, 2.48 6 5.09 g WSN/kg, 6.40 6 0.92 pH, 5.56 6 2.55 mS/cm of EC, and 61.30 6 20.03% of GI. This wide variability in the quality indices of the commercial organic fertilizers allowed us to successfully build a correlation between the NIR spectra and the compost quality indices [3]. Samples with a difference between the reference and predicted values were considered outliers and thus excluded during the calibration process. The removal of outliers was on the basis of being labeled as compositional outliers based on the criterion that if the predicted versus actual difference for a sample was 3 SD or more from the mean difference [13]. For the quality indices, two outliers were removed for the moisture, TOM, and pH; 0 for TN, EC, and GI; 7 for WSOC; and 10 for WSON (Table 1). All the NIR spectra of the collected commercial organic fertilizers could be divided into two groups of signal with different slopes under 1,400 nm, i.e., one group presented an increased curvature and the another one was more flat. Meanwhile, the former had a significant absorbance peak at wavelengths of approximately , 1,420 nm, but while the latter had only a small absorption at this position. The second significant spectral peak was at approximately 1,950 nm (Fig. 2). The absorbance band at 1,420 nm is usually assigned to the O–H and aliphatic C–H, while the band at 1,950 nm is associated with the amide N–H and O–H [20,22,23]. Note that the absorption peaks were heavily overlapped, mainly because the near-infrared spectrum contains all strength information of the chemical bond, chemical composition, electronegativity, etc. Meanwhile, other interference information, such as scattering, diffusion, special reflection, surface gloss, refractive index, and reflected light polarization, affects the near-infrared spectrum [24,25]. Thus, the quantitative predictions are difficult directly through NIR spectra alone. Multivariate analyses are required to discern the response of properties of commercial organic fertilizers from spectral characteristics with the support of chemometric methods, e.g., PLS analysis. Based on the guideline proposed by Saeys et al. [26], the accuracy of the predictions for the calibration model is classified as excellent when r 2 . 0.90, good when 0.81 , r 2 , 0.90, moderately successful when 0.66 , r 2 , 0.80, and unsuccessful when 0.50 , r 2 , 0.65. Meanwhile, according to Albrecht [9] and Chang et al. [27], the accuracy of the PLS model and prediction was considered good for RPD . 2, acceptable for 1.4 , RPD , 2, and unreliable for RPD , 1.4. In this study, the results of the NIRS calibration and validation for the quality indices of commercial organic fertilizers are listed in Table 2 and Figures 3–4. The NIR calibrations allowed accurate predictions of the TOM, WSON, pH, and GI ( R 2 = 0.73–0.93 and RPD = 1.47–2.96). The results were less accurate for the moisture ( R 2 = 0.91, r 2 = 0.79, RPD = 2.22), TN ( R 2 = 0.98, r 2 = 0.80, RPD = 2.25) and EC ( R = 0.99, r = 0.74, RPD = 2.27). However, the WSOC had the worst prediction, with R 2 = 0.88, r 2 = 0.76 and RPD = 2.10. Therefore, predictions were moderately successful for the moisture, TOM, TN, WSON, pH, EC, and GI, but unsuccessful for WSOC. Previous studies have demonstrated that the NIR-PLS was successful in predicting some parameters such as nitrogen (N), carbon (C), C/N, humic acid, pH, respiration, and composting time during the composting process. For example, Saeys et al. [26] developed calibrations using the PCA and PLS regressions for the moisture ( r 2 = 0.91, RPD = 3.22), TOM ( r 2 = 0.90, RPD = 3.00) and TN ( r 2 = 0.86, RPD = 2.63) in pig manure using a mobile spectroscopy instrument. Huang et al. [28] obtained calibrations for the moisture ( r 2 = 0.98, RPD = 7.48), pH ( r 2 = 0.62, RPD = 1.63), EC ( r 2 = 0.90, RPD = 3.10), and TN ( r 2 = 0.97, RPD = 6.11) in animal manure (cattle, chicken, and pig manures) composts using the NIR-PLS method. Vergnoux et al. [18] obtained excellent calibrations using PCA and PLS regressions in sewage sludge compost for the moisture content ( r 2 = 0.91), TN ( r 2 = 0.98), and pH ( r 2 = 0.92). Albrecht et al. [9] evaluated the biological and chemical changes during the composting process of green waste and sewage sludge using NIR and found that the NIR calibrations successfully allowed accurate predictions of N, C, the C/N ratio, humic acid (HA), pH, and composting time, but were less accurate for the OM, protease, acid, and alkaline phosphatase and unsatisfactory for fulvic acid. Soriano-Disla et al. [29] obtained moderately successful predictions for WSOC in compost ( r 2 = 0.75, RPD = 1.70) and in sewage sludge ( r 2 = 0.60, RPD = 1.60). However, these investigations were conducted in samples during composting, and no report has applied the NIR-PLS to predict the indices of commercial organic fertilizers. An obvious difference between samples from the whole composting process and those from commercial organic fertilizers is the wide variability of the ingredients used in the elaboration of the composting heaps. Therefore, obtaining good correlations was more difficult for commercial organic fertilizers used in this study. The schematic of rapidly evaluating the quality of commercial organic fertilizers using near infrared spectrometer was given in Figure S2. The results in this study indicated for the first time that the indices of commercial organic fertilizers could be well- evaluated by the NIR with PLS regression method. In this study the NIR spectroscopy combined with PLS analysis has been developed as an alternative method to traditional chemical analysis for rapidly and accurately predicting the essential quality indices of commercial organic fertilizers. In general, the NIR-PLS technique provided accurate predictions of the TOM, WSON, pH, and GI; less accurate results for the moisture, TN, and EC; and the worst results for WSOC. As a result, we suggest the NIR spectroscopy with PLS analysis may be used as a valuable industrial and research tool to rapidly and accurately assess the quality of commercial organic fertilizers. These photos suggest that the commercial organic fertilizers are more evenly than samples from the composting process. ...