The waterbodies investigated in this study were located in the western part of Jilin, which belongs to the semiarid part of the Songnen Plain (Song et al., 2013). Two groups of lakes were investigated, i.e., the Chagan lake group and the Yuelianghu lake group. The Chagan lake group is made up of Lake Chagan (CGL), Xinmiaopao (XMP), Xindianpao (XDP) and Kulipao (KLP). The Yuelianghu lake group mainly includes Lake Yueliang (YLL), Talahong (TLH) and Xinhuangpao (XHP) (Fig. 1). The two groups are about 60 km away from each other, each of which includes both fresh and brackish waters. The primary economic value for these lakes is fisheries, agricultural irrigation and recreation. The average annual precipitation is about 391 mm, but the average evaporation is up to 1790 mm, resulting in water scarcity. Due to the area dominated by saline-alkali soil, the rainfall flush and agricultural catchment land use can result in an increase of lake salinities. These seven lakes are endowed with similar geological, hydrological and climatic settings; thus we presume that similar processes may control the CDOM components. In order to characterize the CDOM fluorescent components under seasonal variation using EEM–PARAFAC, 67 water samples were collected from the surface of the seven lakes in 1 L acid-cleaned plastic bottles during four field campaigns in June and August 2013 as well as in February and April 2014, respectively. These samples were collected during the ice-covering period using an ice drilling auger. The under-ice surface water came up when a hole was drilled in the ice layer by the auger. The ice shavings were collected in plastic bags and the under-ice surface water was collected in plastic bottles. The collected samples were held on ice and immediately transported to the laboratory in the city of Changchun in Jilin within 3–5 h. In the laboratory, these samples were filtered within 24 h and then kept at 4 ∘C until analysis within 2 days. Latitude and longitude of each sample location were recorded in situ using a Trimble Global Positioning System (GPS).

To characterize the basic parameters of water quality, salinity was measured by a DDS-307 electrical conductivity meter in the laboratory. Salinity was expressed on the basis of the UNESCO practical salinity unit. The pH was measured using a PHS-3C pH meter at room temperature (20 ± 2 ∘C) in the laboratory. Water turbidity was determined using the Shimadzu UV-2600PC UV–visible dual beam spectrophotometer with matching 3 cm quartz cells at room temperature (20 ± 2∘) with Milli-Q water as the reference (UV Talk Letter, 2013). To determine DOC concentrations, water samples were filtered through 0.45 µm filters and then measured using a Shimadzu TOC-5000 analyzer and a 1.2 % Pt on silica catalyst at 680 ∘C. Potassium hydrogen phthalate was used as a reference. The reproducibility of the analytical procedure was within 2–3 % for the current study (APHA/AWWA/WEF, 1998; Song et al., 2011).

In the laboratory, all the samples were filtered at low pressure, first through a precombusted Whatman GF/F filter (0.7 µm), and then through a prerinsed 25 mm Millipore membrane cellulose filter (0.22 µm) into glass bottles. Absorption spectra of the samples were measured between 200 and 800 nm at 1 nm increments using the Shimadzu UV-2600PC UV–visible dual beam spectrophotometer with a 1 cm quartz cuvette and Milli-Q water as a reference. The absorption coefficient aCDOM was calculated from the measured optical density (OD) of the sample using Eq. (1): aCDOM(λ)=2.303OD(λ)-OD(null)/γ, where γ is the cuvette path length (0.01 m) and the factor 2.303 converts from base 10 to base natural logarithm transformation. Some fine particles possibly remained in the filtered solution (Babin et al., 2003; Bricaud et al., 1995); therefore it was necessary to correct for scattering by fine particles and in this case, OD(null) is the average optical density over 740–750 nm where the absorbance of CDOM can be assumed to be zero.

A CDOM absorption spectrum (aCDOM(λ)) can be expressed as an exponential function (Babin et al., 2003; Bricaud et al., 1995): aCDOM(λi)=aCDOM(λr)exp⁡[-S(λi-λr)], where aCDOM(λi) is the CDOM absorption at a given wavelength λi, aCDOM(λr) is the absorption estimate at the reference wavelength λr (440 nm) and S is the spectral slope of the CDOM absorption. According to Helms et al. (2008), S is calculated by fitting a linear model to the data over a wavelength range of 275 to 295 nm (S1) or 350 to 400 nm (S2). To eliminate the inter-laboratory variability, the slope ratio SR= S1/S2 is defined to indicate the molecular weight and photobleaching of CDOM (Helms et al., 2008; Zhang et al., 2010).

The EEM analysis of CDOM was conducted using a Hitachi F-7000 fluorescence spectrometer (Hitachi High-Technologies, Tokyo, Japan) with a 700-voltage xenon lamp. The scanning ranges were 200–450 nm for excitation, and 250–500 nm for emission. Readings were collected in the ratio mode at 5 nm intervals for excitation, and at 1 nm intervals for emission, using a scanning speed of 2400 nm min-1. The band passes were 5 nm for both excitation and emission. A Milli-Q water blank of the EEM was subtracted to eliminate the water Raman scatter peaks (McKnight et al., 2001; Stemdon et al., 2003; Zhang et al., 2010, 2011).

The inner-filter effect, which results from reabsorption and excitation of the fluorescence itself, can reduce the fluorescence intensity by 5 % (Larsson et al., 2007; McKnight et al., 2001). In order to eliminate the inner-filter effect, the EEM was corrected for absorbance by multiplying each value in the EEM by a correction factor based on the assumption that the average path length of absorption of the excitation and emission light is one half length of the cuvette (McKnight et al., 2001; Zhang et al., 2010). The correction function is expressed as follows: Fcorr=Fobs×10(Aex+Aem)/2, where FCorr and Fobs are the corrected and uncorrected fluorescence intensities and Aex and Aem are the absorbance values at the respective excitation and emission wavelengths.

The measured fluorescence intensity is dependent on the concentration of the dissolved fluorophores in waterbodies. Finally, the fluorescence intensities of all samples' EEMs were normalized to the area under the Milli-Q water Raman peak (λex = 350 nm, λem = 371–428 nm) measured daily (Lawaetz and Stedmon, 2009). The contour figures of the EEMs were plotted using the MATLAB 10.0 software package (MathWorks, Natick, Massachusetts, United States).

PARAFAC, a three-way method, is applied to divide the CDOM fluorescence into separate fluorescent signals (Andersen and Bro, 2003; Stedmon and Bro, 2008). According to Stedmon and Bro (2008), a similar PARAFAC analysis is carried out in the present study using the DOMFluor toolbox in MATLAB with the “N-way toolbox for MATLAB” (Andersson and Bro, 2000). Before PARAFAC modeling, the excitation wavelengths from 200 to 220 nm and the emission wavelengths from 250 to 300 nm were deleted because of their poor quality. In order to remove the effect of Rayleigh scatter on PARAFAC modeling, the missing values (NaN – “not a number”) were inserted in the regions (Ex-20⩽ Em ≦ Ex+20 and 2Ex-20≦ Em ≦2Ex+20; unit: nm) which are significantly influenced by the first- and second-order scattering from the measured spectroscopic data (Hua et al., 2007; Stedmon and Bro, 2008).

To determine the appropriate number of PARAFAC components, the split-half validation procedure was executed to verify whether the model was valid by comparing the emission and excitation loadings from each half (Stedmon and Bro, 2008). Split-half analysis is the most effective method for implementing the PARAFAC models, in which the EEM is randomly divided into four groups of equal size, and then analyzed for two split halves (1–2 and 3–4 half) respectively. If the correct number of components is chosen, the excitation and emission loadings from the two groups should show the same shape and size (Bro, 1997, 1999). The fluorescence intensity of every component was represented by Fmax (Raman unit: nm-1) (Stedmon and Markager, 2005).

Statistical analysis was conducted using the SPSS 16.0 software package (Statistical Program for Social Sciences). Regression and correlation analysis was used to describe the relationship between the CDOM absorption coefficient, DOC concentration, salinity and Fmax. A model II-ANOVA (analysis of variance) was performed to determine whether the seasonal variability is higher than between-lake variability. The difference is considered to be statistically significant when p values are less than or equal to 0.05.

The water quality parameters, i.e., pH, salinity, and turbidity, for the 67 water samples collected from June 2013 to April 2014 in the western part of Jilin are displayed in Table 1. When the set of samples from various field trips was pooled together, the waters had high pH values and high salt contents. The highest salinity was present when the lakes were frozen in February 2014, whereas relatively constant values (around 0.40) were exhibited in the other three seasons. In addition, the waterbodies were highly turbid. The highest turbidity was present in June 2013, and then reduced in August 2013, and the lowest value was recorded in February 2014. Compared with February 2014, the turbidity had almost no change in April 2014 (Table 1).

Based on the EEM “peak picking” technique, the key fluorescence peaks can be observed in 67 water samples: two humic-like and two protein-like substances (Coble, 1996; Stedmon et al., 2003). The humic-like components are the mixture of the humic-like acids of aromatic and aliphatic compounds from terrestrial substances, and aquatic humic-like substances of phytoplankton origin. With respect to the protein-like components (i.e., tyrosine-like and tryptophan-like substances), they mainly consist of dissolved amino acids. As an example, Fig. S1 in the Supplement displays the EEMs of samples from Lake Xindianpao in different seasons. The peaks comprise two humic-like fluorescence peaks: one in the ultraviolet range (Ex/Em = 220–240/410–430 nm) and the other in the visible range (Ex/Em = 300–340/410–450 nm); and protein-like fluorescence peaks: tyrosine-like (Ex/Em = 210–230, 270–280/310–330 nm) and tryptophan-like (Ex/Em = 220–230, 280–300/350–370 nm).

In our study, four separate fluorescent components (Fig. 2a–d) and the excitation and emission loadings (e–h) of the four components identified by EEM–PARAFAC are summarized in Fig. 2 and Table 2. The first fluorescent component (C1) was a biological degradation humic-like component comparable to humic-like peaks (M and N) in marine and in phytoplankton degradation experiments for inland waters (Coble, 1996; Zhang et al., 2009). Component 2 was consistent with the humic-like peaks (A and C) defined by Coble (1996). Component 3 resembles the tryptophan-like (T) component as found by Baker et al. (2004) and Hudson et al. (2007). Component 4 is likely related to the tyrosine-like component (B) (Hudson et al., 2007). Components 3 and 4 represent autochthonous semi-labile CDOM associated with bacteria activity and phytoplankton degradation (Borisover et al., 2009; Stedmon et al., 2003). In particular, there was a shoulder (i.e., deflection but not peak) at the excitation wavelength 310–330 nm in component 4 and 330–340 nm in component 3, which may be due to the residual Raman peaks in some water samples (Fig. 2c–d). In this study, not all of the four components were present in all of the samples.

These fresh and brackish water in Jilin in Northeast China are endowed with similar geological, hydrological and climatic settings; thus it is presumed that similar processes may control the CDOM components. When a model II-ANOVA using season and lake as random effect factors was performed, it shows that the seasonal variability (F > Fcrit, p < 0.05) is higher than between-lake variability. Therefore, the water samples from different lakes for every season were pooled together in order to study the seasonal variation of the fluorescent components. As shown in Fig. 3a, the average fluorescence intensity of the four components had seasonal variation. When all the water samples in different seasons were pooled together, the average value of total fluorescence intensity was 2.05 ± 0.93 nm-1, corresponding to the intensities of 0.71 ± 0.32 (C1), 0.33 ± 0.11 (C2), 0.50 ± 0.24 (C3) and 0.51 ± 0.26 (C4) nm-1 for different components. These results can demonstrate that the fluorescence intensity was dominated by C1, implying most of the CDOM for the seven inland lakes originated from the degradation of phytoplankton and microorganisms. The protein-like components (C3 and C4), related to bioavailability and microbial activity of CDOM, had almost the same magnitude. In all four seasons, the fluorescent component C2, which was terrestrially imported to waterbodies, contributed less to total fluorescence than the other three. The total fluorescence intensity differed under seasonal variation, varying from 2.54 ± 0.68 nm-1 in June to 1.93 ± 0.70 nm-1 in August 2013, and then increased to 2.34 ± 0.92 nm-1 in February and reduced to the lowest intensity of 1.57 ± 0.55 nm-1 in April 2014 (Fig. 3c). The intensities of four fluorescent components (i.e., 0.75 ± 0.17 (C1), 0.32 ± 0.06 (C2), 0.69 ± 0.24 (C3) and 0.77 ± 0.20 (C4) nm-1) (Fig. 3d) from the samples collected in June 2013 exhibited similar trends to that for the pooled data set. These values were higher than the seasonal average except C2 (0.32 ± 0.06 nm-1). This can be explained by enhanced activities from plant degradation and microbial activities, but fewer terrestrial substances were imported to the waterbodies in June; therefore, the fluorescence intensity of C2 was lower than the seasonal average. Compared to the fluorescence intensity in June, the three fluorescence intensities (0.65 ± 0.14 (C1), 0.33 ± 0.16 (C3) and 0.52 ± 0.36 (C4) nm-1) from the samples collected in August 2013 were reduced, but an increased value was recorded for C2 (0.42 ± 0.05 nm-1) (Fig. 3d). Particularly, the fluorescence intensities of two protein-like components showed an obvious difference. This can be attributed to substantially increased precipitation up to 180 mm in July from June to August 2013 (Fig. 3b); therefore, floods occurred when rainfall continued to increase in August. Gradually, DOM contained in terrestrial CDOM was flushed by rainfall to the lakes; therefore, the C2 (0.42 ± 0.05 nm-1) fluorescence intensity became higher. In accordance with Cheng et al. (2010), the rainwater CDOM for this study was largely characterized by protein-like components (Cheng et al., 2010). The fluorescence intensity of the rainwater CDOM was very weak, and the rainwater CDOM also had a much lower humic-like concentration (Fig. S2b). The intensities of the other three components decreased because of dilution resulting from heavy rain and relatively weak microbial decomposition of plants.

The highest C1 (1.02 ± 0.38 nm-1), presented in February 2014, and the C2 (0.39 ± 0.12 nm-1) intensity remained almost the same as that in August 2013. However, the protein-like components indicated that the C3 (0.57 ± 0.25 nm-1) intensity was higher than the C4 (0.35 ± 0.17 nm-1) intensity, which was opposite to the results from other months (Fig. 3d). In cold winters, the surface waters formed a thick layer of ice covering the lake waters. Because the ice cover reduced light penetration and restricted gas exchange between the underlying water and atmosphere, vigorous biological activity in the lakes would be reduced at low temperature and low light level (Thomas, 1983; Uusikiv et al., 2010; Wharton Jr., et al., 1993). Although the biological activity was very weak, there could still be a bit of production of C1 and C3 in lake water. Also, dissolved materials were left in the underlying surface waters and little terrestrial matter was imported to the lakes once covered by ice (Stedmon et al., 2007). Therefore, the C1 and C3 in the water of the lakes beneath the ice layers would have been produced and accumulated simultaneously, whereas the C2 remained the same. Obviously, the fluorescence intensity of component 1 reached the highest value for the winter samples. As shown in Fig. S2a, another striking feature for the winter samples was that the fluorescence of CDOM in the ice was dominated by the tyrosine-like C4 component, which is consistent with the findings of Barker et al. (2009, 2013) and Stedmon et al. (2007). It showed that the C4 component was left in the ice cover when the lakes were frozen. Therefore, it is not surprising that the intensity of component C4 for water beneath ice layers was reduced and the concentrated C3 showed a much higher fluorescence intensity. In April 2014, the intensities of four fluorescent components (0.47 ± 0.17 (C1), 0.25 ± 0.08 (C2), 0.40 ± 0.16 (C3) and 0.45 ± 0.13 (C4) nm-1) (Fig. 3d) exhibited similar seasonal trends, though these values were much lower than the average. Our interpretation is that the ice CDOM was characterized by the tyrosine-like component (C4) (Fig. S2a), and the fluorescence intensity of C4 contributed by the ice meltwater was very weak. However, the underlying lake CDOM included both humic-like (C1 and C2) and protein-like (C3 or C4) components. When the ice in the lakes melted into water during warming weather and biological degradation, and human activity was weak, the lake CDOM was diluted by the ice meltwater and the fluorescence intensity reached its lowest value in early spring.

The concentration of DOC, CDOM absorption coefficients and the slope ratio SR are shown in Table 3. The DOC concentrations ranged from 835.83 to 7345.83 µmol L-1, with an average value of 3133.05 ± 1504.14 µmol L-1 during the period from June 2013 to April 2014, demonstrating seasonal dynamics that can be attributed to hydrological, climatic and landscape variations (Song et al., 2013). The highest average DOC concentration (4587.03 ± 1666.83 µmol L-1) was present in February 2014 (ice-covered period); whereas relatively constant values of approximate 2500 µmol L-1 were observed in the ice-free season. The relative high DOC concentration in the ice-free season was caused by the evapo-condensed effect due to the prolonged sunshine duration for the lakes in the Songnen Plain. The higher DOC concentration in winter can be attributed to the accumulated DOC left in the liquid phase when ice formation took place, resulting in the higher DOC concentration in the underlying water (unpublished material). Generally, the absorption coefficient a(350) is used as a proxy for characterizing CDOM concentration (Guo et al., 2010; Zhang et al., 2011); a(280) is related to DOC biodegradation (McDowell et al., 2006), and a(254) can be used to characterize the optical properties of DOC aromaticity (Jaffé et al., 2004; Weishaar et al., 2003). The highest averaged CDOM absorption coefficients, a(350), a(280) and a(254), were also present in February 2014, corresponding to the highest DOC concentration. The SR values of the two wavelength ranges (275–295 over 350–400 nm) were used to represent DOM molecular weight (Helms et al., 2008). The lowest mean of SR was present in August 2013, suggesting that the relatively weak microbial decomposition of plants and lots of terrestrially imported substances through rainwash resulted in the higher average molecular weight of DOC.

When the whole data set (N= 67) was pooled together, there were significantly positive linear relationships between a(254), a(280), a(350) and Fmax for two humic-like components (C1 and C2), respectively, but mostly, such correlations were not observed for the protein-like components (Fig. 4a and b, Table 3). These results were in accordance with previous investigations (Zhang et al., 2010, 2011). Components 1 and 2 were strongly linearly correlated with each other (R2= 0.628) (Fig. 4c), indicating that the concentrations of the two humic-like components were controlled by common sources (Baker and Spencer, 2004). There was a weak relationship (R2= 0.051) between the protein-like components (C3 and C4), possibly because of a complex origin of CDOM such as rainfall in summer, ice in winter and organic pollutants derived from domestic, agricultural and industrial sewerage, which represent the complex origins of CDOM. However, there was almost no correlation between the humic-like and protein-like components. The linkage of a fluorescence signal to DOC was very complicated because of the seasonal impacts, i.e., increased rainfall, algal blooms and ice cover, which affect the DOC concentration. Due to both steady and labile CDOM fluorescent components in DOC, the fluorescent signal changed with the ratio of fluorescent and non-fluorescent CDOM components (Henderson et al., 2009). A weak relationship (R2= 0.411) (Fig. 4d) was found between DOC and component 3, likely from the decay of plants through microbial activity or the pollution from human and animal waste.

Different from the findings by Yamashita et al. (2008) for ocean water, this study did not find obvious correlation between salinity and EEM–PARAFAC-extracted components, with the exception of C3 (R2= 0.469) (Table 4 and Fig. 4f). The most important finding for the water samples collected in different seasons from the Songnen Plain is a significant relationship (R2= 0.930) between salinity and DOC (Fig. 4e). This is because DOC is evapo-condensed from spring to autumn and freeze-accumulated in winter in the semiarid region. A prolonged sunshine duration can result in an evapo-condensed DOC concentration in the ice-free season. On the other hand, the DOC accumulates when the lakes freeze in winter, leaving DOC in the liquid phase.