Extraction of PARAFAC components
A strong peak and four weak shoulders were represented in the EEM spectroscopy of DOM from the wastewater sample at the preliminary grit chamber (Fig. 1a). Based on published literatures for EEM spectra of organic matter [30], peak T at λEx/Em = 270–290/330–370 nm might be associated with tryptophan-like fluorescence substance (TRLF), while shoulder B at λEx/Em = 260–280/300–330 nm might be relative to tyrosine-like fluorescence substance (TYLF). Shoulders A and C at λEx = 250–280 nm and 340–370 nm with the same λEm (430–470 nm) might be referred to UV fulvic-like (UV-FALF) and visible fulvic-like fluorescence substance (Vis-FALF), respectively. Shoulder M located between shoulders A and peak C might be microbial byproduct-like fluorescence substance (MBLF). The intensities of TRLF and TYLF were much higher than those of FALF and MBLF. This indicated that the former was dominant in DOM, which could be assigned with fresh and less degraded fresher and more recalcitrant materials derived from anthropogenic activities [31]. After the consecutive treatment, the intensities of shoulder B and peak T fell to a much larger extent than those of shoulders M, A and C (Fig. 1b).
The intensities of EEM spectra of POM from the wastewater at the PGC were much less than those of DOM (Fig. 1a and c), which attributed that about 60% of the total suspended solids should be removed after the preliminary grit chamber. Moreover, the still existing particles could be mostly disintegrated into smaller products by a hydrolysis process, which could be further degraded by bacteria and microbials [3]. Interestingly a distinct peak at λEx/Em = 240–260/360–390 nm in EEM spectra of POM could be concerned with TRLF (Fig. 1c and d), which had been partially degraded. In the wastewater treatment process, the trends of should B and peak T of POM were similar to those of DOM.
Four different fluorescent components were extracted from EEM spectroscopies of DOM and POM using PARAFAC modeling (Fig. 2). The component I (C1) with a single peak at λEx/Em = 275/330 nm similar to shoulder B, could be TYLF, and the component II (C2) with only a peak at λEx/Em = 285/350 nm similar to peak T, could be TRLF. The component III (C3) exhibited two peaks at λEx/Em = 240/400 nm (M1) and 300/400 nm (M2) resembled as shoulder M, and should be MBLF. The component IV (C4) also displayed primary and secondary peaks at λEx/Em = 360 and 275/450 nm comparable to shoulders C and A, and should be Vis-FALF and UV-FALF, respectively.
Evaluating removals of PARAFAC components
Figure 3a–c showed the abundances, relative proportions, and removal efficiencies of PARAFAC components of DOM in the WWTP. In the wastewater process, the decreasing order of total Fmax of C1 to C4 was PGC (13,321.03) > PRS (10,107.43) > ANA (3890.67) > FAC (3596.55) > ANO (3631.09) > AER (2935.13) > SES (2669.48), whose decreasing order of the removal efficiencies was ANA > (61.51%) > PRS (24.12%) > AER (19.17%) > SES (9.05%) > ANO (7.56%) > FAC (-0.96%). This indicated that DOM fractions could be mostly removed in the anaerobic tank. The Fmax of the C1 and C2 reduced to much greater extents than those of the C3 and C4, which elaborated that TYLF and TRLF were mostly removed, especially in the anaerobic tank. The C1%Fmax values were relatively constant (38.93%-49.88%) in the treatment process, while the C2%Fmax showed firstly increasing from 28.95% to 51.79% then decreasing to 41.42%. The C3%Fmax kept less than 10% (except for 14.66% in the preliminary grid chamber), so was the C4. Interestingly, the sum of the C1 and C2 were much more than 78.84% in a given unit, verified that TYLF and TRLF were dominant components of DOM in the wastewater. It was reported that the protein-like fluorophores are typically labile to biodegradation, and humic-like components require further treatments such as adsorption, coagulation, and advanced oxidation technologies [32, 33]. The removal efficiencies of C1 at the PGC to PRS were much higher than those at the rest sites, so were C3 and C4. This indicated that TRLF, MBLF and FALF were mostly degraded in anaerobic tank by anaerobic microorganisms, besides they were partially removed through absorption and sedimentation in the primary sedimentation tank. Furthermore, the removal efficiencies of C2 at the ANA and AER were much higher than those at the other sites, indicating that TRLF was mostly degraded in anaerobic and aerobic tanks by microorganisms.
Figure 3d–f exhibited the abundances, relative proportions, and removal efficiencies of PARAFAC components extracted from POM in the WWTP. In the wastewater units, the Fmax sum of C1 to C4 in the PGC (765.73) was the highest, followed by the PRS (555.42), ANO (494.62), ANA (489.88), FAC (376.89), AER (259.32) and SES (182.87). However, the removal efficiencies of POM fractions in the site AER was the highest (31.19%), followed by the SES (29.48%), PRS (27.47%), FAC (23.80%), ANA (11.80) and ANO (-0.97). This indicated that POM fractions could be deeply removed through adsorption and sedimentation in the primary and secondary sedimentation tanks, and degraded by aerobic microorganisms in the aerobic tank [3, 32]. Much higher proportions of the Fmax of the C1 and C3 were removed than those of the C2 and C4 in the successive treatment units (Fig. 3d), which indicated that TYLF and MBLF were highly much degraded. The mean of the C1%Fmax (37.98 ± 4.49%) was the highest in the wastewater treatment process, followed by the C3 (33.95 ± 7.94%), C2 (23.97 ± 5.22%) and C4 (4.11 ± 1.55%). Noticeable, the sum of the C1 and C3 were much more than 67.22% in each unit (Fig. 3e), expounding that TYLF and MBLF were representative components of POM in the wastewater. The removal efficiency of the C1 was highest at the site AER, followed by the PRS, FAC, ANA, SES and ANO, and the removal efficiency of the C2 was the highest at the site AER too, followed by the PRS, FAC, SES, ANO and ANA (Fig. 3f). These protein-like substances were mainly removed in the aerobic tank. This attributed that the protein-like should be broken into smaller products through the hydrolysis process [3], which could be further metabolized by bacteria and microbial. The descending order of the C3 removal efficiencies was SES > ANA > ANO > FAC > PRS > AER, indicating that MBLF were mostly removed in the second sedimentation tank. The descending order of the C4 removal efficiencies was ANA > AER > FAC > PRS > SES > ANO, explaining that FALF substances were mostly degraded by anaerobic microbial in the anaerobic tank.
Inter/inner dynamic-variations of PARAFAC components
There were six peaks in PARAFAC components (Fig. 2), whose changing order could be identified by hetero 2D-COS and 2D-COS in the seven successive treatment units. This could reveal inter/inner dynamic-variations of PARAFAC components of DOM or POM from the wastewater in the WWTP.
Figure 4 exhibited synchronous and asynchronous maps of the hetero 2D-COS based on excitation loadings of PARAFAC components of DOM. There was a positive relationship between peaks B and T in both the synchronous map and asynchronous map (Fig. 4a, b), indicating that the changing order was B → T according to Noda’s rule [30]. Peak B had positive correlations with peaks A and C in the synchronous map (Fig. 4c), while negative correlations with the peaks A and C in the asynchronous map (Fig. 4d), elaborating that the changing order was A and C → B. Peaks A and C presented positive correlations with peak M1 in either synchronous or asynchronous maps (Fig. 4e, f), proving that the changing order M1 → A and C. There was a positive relationship between peaks M2 and A in the synchronous map and a negative relationship in the asynchronous map (Fig. 4e, f), explaining that the change order was A → M2. Peak M2 had a positive correlation with peak C in the synchronous and asynchronous map (Fig. 4e, f), indicating that the changing order was M2 → C. In summary, the changing order of the six peaks was M1 → A → M2 → C → B → T, indicating that the continuous dynamic variation of MBLF occurred in the successive treatment units, while disconnected dynamic variations of TRLF and TYLF. This indirectly proved that TRLF and TYLF could be removed in the anaerobic/anoxic units.
Figure 5 showed synchronous and asynchronous maps of hetero 2D-COS and 2D-COS based on excitation loadings of PARAFAC components of POM. Peak B was positively related to peaks M1 and M2 in the synchronous map (Fig. 5a), while negative related to peaks M1 and M2 (Fig. 5b). This suggested that the varying order should be M1 and M2 → B. Peak T was positively relative with peak M1 in both synchronous and asynchronous maps (Fig. 5c, d), indicating that the varying order was T → M1. Peak T was positively relative with peak M2 in the synchronous map, while negatively relative with peak M2 (Fig. 5c, d). This suggested that the varying order was M2 → T. Peaks M1 and M2 had positive correlations with peaks A and C, while negative correlations with peaks A and C (Fig. 5e, f). This proposed that the varying order was A and B → M1 and M2. Peak A represented a positive correlation with peak C (Fig. 5g), while a negative correlation with peak C in the asynchronous map (Fig. 3h). This indicated that the varying order was C → A. Based on the above results, the varying order of the peaks was C → A → M2 → T → M1 → B, indicating that FALF showed a successive variation in the wastewater treatment process, while TRLF showed an undulated variation. This indirectly validated that TRLF of POM could be degraded in the anaerobic and anoxic units.
Latent transformation of PARAFAC components
An SEM based on the hypothetical model could be developed as a modeling with an endogenous latent variable and four observed variables. Meanwhile, the former was associated with the removal efficiencies of Fmax sum of PARAFAC components of DOM or POM, and the latter was concerned with the Fmax of C1 to C4.
The modeling with Chi-square = 52.887, Degree of freedom = 3 and Probability level = 0.000 showed a marginal acceptance for the latent transformation of DOM fractions (Fig. 6a), for the Chi-square with less than 5.0 was available [34, 35]. This could attribute that C3 or C4 had a weak effect on the total PARAFAC component efficiencies, as proved by relatively small path coefficients (0.22 or 0.22). This indirectly that the percentages of the C3 and C4 were smaller than those of the C1 and C2. The C1 with a path coefficient of -1.00 had a strongly negative direct effect on the removal efficiencies, indicating that C1 should be removed in the anaerobic tank [32]. C2 with the path coefficient of 0.35 showed a positive effect on the removal efficiencies, elaborating that C2 were continuously removed in the treatment process. C3 with the path coefficients of 0.93 or 0.88 showed a strongly positive direct effect on the C1 or C2, indicating C3 showed an indirect effect on the removal efficiencies. This indirectly verified that TYLF and TRLF could be degraded by microorganism, especially in the anaerobic tank [32]. Furthermore, C4 showed an indirect effect on the removal efficiencies through C1, as insight into the path coefficient of -0.35. This indirectly evidenced that C1 could be degraded partially into C4 in the wastewater treatment.
The model deduced from PARAFAC components of POM was referred as the Chi-square = 36.556, Degree of freedom = 3 and Probability level = 0.000 represented a rough acceptance too (Fig. 6b). This could contribute to a poor direct influence of the C3 with the path coefficient of 0.18 on the removal efficiencies, and a weak direct influence of C4 with the path coefficient of 0.13 on the C2. This indirectly validated that C1 and C3 was the representative component of POM, instead of the C2 and C4. C1 with the path coefficient of 0.72 displayed a direct positive influence on the removal efficiencies, demonstrating that TYLF could be unceasingly removed in the successive treatment units. However, C2 and C4 with the path coefficients of -0.37 and -0.61 respectively, displayed direct negative influences on the removal efficiencies, demonstrating that the variations of C2 and C4 were apparently unstable (Fig. 3c). C3 displayed a directly positive influence on the C1 or C2, as confirmed by the relatively large path coefficients (0.84 and 0.57). This indicated that C3 displayed an indirectly positive influence on the removal efficiencies through the C1 and C2. Undeniably, this indirectly proved that TYLF and TRLF should be discomposed by microorganism too. Interestingly, C4 displayed an indirectly positive influence on the removal efficiencies by the C2 with the path coefficient of 0.13. This could attribute that C2 and C4 should be partially degraded into dissolved organic matter.