Introduction

Dissolved organic matter (DOM) is a heterogeneous mixture consisting of carbohydrates, proteins, lignins, organic acids, and various uncharacterized structures such as humic substances (HS) (Thurman 1985). DOM is considered to be an important element for aquatic ecosystems owing to its multiple functionalities, including nutrient sources, light attenuation, changing metal toxicity, and microbial respiration (Mulholland 2003). DOM is introduced into aquatic systems through external (allochthonous) and internal (autochthonous) sources.

From the standpoint of DOM cycling, a coastal environment can be categorized into a transition zone where both external and internal sources are mixed together and various biogeochemical transformation processes occur. Several factors associated with upstream catchments such as hydrology, geomorphology, land use/land cover, soil types, and seasonality are likely to govern the quantity and the quality of the DOM that is exported into the coastal environments (Stedmon et al. 2003; Kowalczuk et al. 2009; Yamashita et al. 2011). In nearby oceanic water, meanwhile, other processes such as microbial degradation, photo-oxidation, and algal productivity were proposed as critical factors to determine the temporal and spatial variations in DOM (Romera-Castillo et al. 2011; Guo et al. 2011; Maie et al. 2012).

The Shiwha Lake, an artificial coastal lake, is located on the west coast of South Korea adjacent to the industrial cities of Ansan and Siheung. This lake is surrounded by a 12.6-km-long dike (constructed in 1994) as part of the development of 173 km2 of a tidal flat. This estuary originally provided good habitats for aquatic organisms, receiving water discharge from the upstream watershed through streams and runoff (KICOX 2011). However, the dike construction has led to a serious deterioration of water quality in the region due to the insufficient circulation of the enclosed lake water and the continuous input of anthropogenic pollutants from nearby industrial complexes (KORDI 1999). The average chemical oxygen demand (COD) in the midregion of the lake before the construction of the dike was 4 mg L−1, but it increased up to 15 mg L−1 after the completion of the construction. To cope with the problem, the South Korean government designated Lake Shihwa as a special management zone, and it opened a water gate in a middle of the dike in 2001 to allow water exchange with the adjacent sea (Ra et al. 2011). In 2011, a tidal power plant equipped with its power capacity of 254 MW was constructed to promote seawater circulation and to supply electric power to the nearby industrial complex (Cho et al. 2012). In spite of the recent effort, however, continuous pollution input from the upstream industrial complexes and the urban areas has been a concern for maintenance of a sustainable ecosystem for this area (Yoo et al. 2009; Moon et al. 2012).

Excitation-emission matrix (EEM) fluorescence spectroscopy has been widely utilized as the simplest and most effective method for probing the composition and the sources of DOM in aquatic environments. Stedmon et al. (2003) used parallel factor analysis (PARAFAC) to successfully decompose a large data set of EEMs into several separate components representing its unique fluorescence features. Many subsequent studies proved that combining EEM with PARAFAC provided more reliable and plentiful information on the fluorescence signals of DOM in various aquatic ecosystems (Murphy et al. 2008; Yamashita et al. 2010; Yamashita et al. 2008; Guo et al. 2011). There were many reports that successfully explored the spatial and temporal variability of DOM in coastal regions using EEM-PARAFAC analyses (Yamashita et al. 2011; Chari et al. 2012).

A few studies have been reported on ecological and environmental evaluations of the Shiwha Lake since the construction of the tidal power plant, in which the accumulation of heavy metals in sediments and bioaccumulation of benthic organisms were highlighted (Ra et al. 2011; Won et al. 2012). To date, however, there has been no investigation on DOM in the coastal region affected by the operation of a tidal power plant. It is thus required to examine the temporal and spatial variability of DOM in the lake, tracking its sources. Therefore, this study aimed at (1) exploring the distributions of DOM and its spectroscopic characteristics within Lake Shiwha and the upper streams at two different seasons (i.e., the monsoon and the premonsoon seasons) and (2) tracking the sources of the lake DOM by employing various spectroscopic techniques including EEM-PARAFAC and multivariate statistical analysis.

Materials and methods

Study area and sample collection

The Shiwha Lake is located on the west coast of South Korea (between 126° 31′–127° 00′ E and 37° 11′–37° 23′ N, Fig. 1). The depth reaches 17.9 m in the middle zone with the average value of 4.5 m, and the area is approximately 79.4 km2. The total watershed area is 476.5 km2 and the lengths of the inflowing streams are approximately 10 km. The water-holding capacity of the lake reaches up to 323,769,000 m3 (Cho et al. 2012). Approximately 13,000 mills involving mechanical (48 %), electrical/electronic (19 %), petrochemical (8.3 %), steel (5.7 %), and textile companies (3.5 %) are currently operated in the upstream industrial complex areas. The total population of the two cities of Shiheung and Ansan exceeds 1 million (KICOX 2011). The average rainfall in this region for the year 2012 was 1,719 mm. Intense rainfalls occur mainly in the monsoon season from June to September. The amount of rainfall during this period of 2012 accounts for 82.4 % of the total annual rainfall (www.wamis.go.kr).

Fig. 1
figure 1

Study area and the sampling locations

Full size image

A total of 15 sampling sites were selected to fully reflect the spatial variability of the water quality and DOM of the lake and the surrounding areas (Fig. 1). Sampling was conducted twice in 2012 during two separate nonrain events of the premonsoon (May 14–15) and the monsoon seasons (September 5–7). For each sampling season, eight samplings were made from the surface and the bottom layers at four different lake sites (L1–L4). Five samples (I1–I5) were collected from the channels receiving treated wastewater and surface runoff from the industrial complexes. Four samples (U1–U4) were collected from the upper streams surrounded by forested and agricultural areas. Two sampling sites (E1 and E2) represented an estuary region, which connects the upper streams with the lake. The collected samples (4 L) were refrigerated and stored in an ice box, moved to a laboratory, and kept in a refrigerator. All sample analyses were carried out within a week after their collection except for biochemical oxygen demand (BOD), which was completed within 12 h.

General water quality and DOM spectroscopic measurements

Measured general water quality parameters included BOD, COD, total nitrogen (TN), total dissolved phosphorus (TDP), and chlorophyll a (Chl a). All the measurements were made based on standard protocols (APHA 2005).

An aliquot (100 ml) of each collected sample was filtered using a prewashed 0.45-μm membrane filter (cellulose acetate, Advantec) for DOM analyses. Dissolved organic carbon (DOC) concentrations (mg C L−1) were determined using a Shimadzu V-CPH analyzer. Ultraviolet (UV)–visible spectra from 250 to 600 nm were recorded using a UV–visible spectrophotometer (Evolution 60, Thermo Scientific) with a 1-cm quartz cuvette. Specific UV absorbance (SUVA) values of the samples were calculated by multiplying DOC concentration-normalized UV absorbance at 254 nm by a factor of 100 (i.e., 100 × UV254 / DOC) (Weishaar et al. 2003). The spectral slopes of S275–290 and S350–400 were obtained by the ratios of the two best fit regression lines from the natural log-transformed absorption spectra of the filtered samples at the wavelength ranges between 275 and 295 nm and between 350 and 400 nm, respectively (Helms et al. 2008). In addition, an absorption coefficient at 375 nm (a375) was measured for chromophoric DOM (CDOM) concentrations.

Fluorescence EEMs were measured with a luminescence spectrometer (LS-55, PerkinElmer) by scanning the emission spectra from 280 to 550 nm at 0.5-nm increments and stepping through excitation wavelengths from 250 to 500 nm at 5-nm increments. Excitation and emission slits were adjusted to 10 and 5 nm, respectively, and the scanning speed was set at 1,200 nm min−1. To limit second-order Rayleigh scattering, a 290-nm cutoff filter was used for all the measurements. When the UV absorbance of the samples at 254 nm exceeded 0.05 cm−1, the corresponding samples were appropriately diluted to avoid the inner filter correction (Hur et al. 2007). The background fluorescence EEM from a blank solution (Milli-Q water) was also taken into account. Fluorescence intensity was normalized in quinine sulfate equivalent (QSE) units, where 1 QSU corresponded to the maximum fluorescence intensity for 1 μg L−1 of quinine in 0.1 N H2SO4 at the excitation/emission of 350/450 nm (Coble et al. 1998). The fluorescence index (FI) was calculated as the ratio of the emission intensities at 450 to 500 nm for an excitation wavelength of 370 nm (McKnight et al. 2001). FI values have been used to differentiate between allochthonous and autochthonous DOM sources. Humification index (HIX) was calculated according to a definition suggested by Zsolnay et al. (1999) with an exception of using the excitation wavelength of 255 nm instead of 254 nm. Higher HIX values are typically associated with the characteristics of soil-derived DOM such as a higher content of aromaticity, a lower rate of mineralization, and a lower percentage of oxygen-containing functional groups (Fuentes et al. 2006).

PARAFAC modeling

For this study, PARAFAC modeling was performed using MATLAB 7.0 (MathWorks, Natick, MA, USA) with the DOMFluor Toolbox (http://www.models.life.ku.dk). The details of the modeling are well described elsewhere (Stedmon et al. 2003; Kowalczuk et al. 2009; Nguyen et al. 2013). The split-half analysis was used to validate the identified components and the number.

Statistical analyses

Statistical analyses including correlation coefficients and PCA were performed using XLSTAT (Addinsoft, New York, NY, USA). Differences in parameters were evaluated using the p values based on Student’s t test. The relationships among variables were evaluated using regression and correlation analyses. Shapiro-Wilk test was performed to confirm the normality of the measured parameters.

Results and discussion

Spatial and seasonal variations of general water quality parameters

The collected samples exhibited large variations in the organic matter concentrations ranging from 0.7 to 5.9 mg L−1 for BOD, from 5.2 to 72.6 mg L−1 for COD, and from 1.8 to 23.9 mg L−1 for DOC, respectively (Table 1). The small portion of BOD to COD indicates that the organic matters are mostly constituted by nonbiodegradable fractions. The nutrient concentrations varied from 0.5 to 4.7 mg L−1 and from 0.023 to 0.177 mg L−1 for TN and TDP, respectively (Table 1). Irrespective of the sampling periods, the lake samples (L1 to L4) presented lower levels of the organic matter and the nutrient concentrations than those of the upstream catchments (U1 to U4 and I1 to I5) (t test, p < 0.05). The exceptions were BOD and TDP concentrations in September. The relatively low concentrations in the lake may be attributed to the dilution effect driven by the exchange of the lake water with the nearby seawater. It has been reported that the exchanged water reaches 16 million tons per day, accounting for 8.9 % of the storage capacity of the lake (Cho 2005). Except for BOD, the estuary area showed the concentration ranges of the water quality parameters falling between those of the upstream catchments and the lake (Table 1).

Table 1 Concentrations (mean ± standard deviation) of selected water quality parameters for the four different sampling regions at two different sampling periodsa
Full size table

Seasonal variations in the general water quality parameters were compared for each categorized sampling region. The organic pollution of the upstream catchments, which were represented by BOD, COD, and DOC concentrations in average, was more serious in May than in September (p < 0.05), and the same trend was observed for the nutrient concentrations (i.e., TDP and TN). In the premonsoon season, the pollutants from the industrial and municipal effluents were the least diluted by precipitation. In contrast, the dilution effect could be more pronounced in September, in which much higher precipitation (232 mm) was reported (www.wamis.go.kr) than in May (9 mm). The lower salinity values of the upstream catchments in September versus May (t test, p < 0.05) also supported the enhanced dilution effect in the monsoon season. However, the lake samples exhibited the opposite trend for the seasonal variations. For example, except for DOC, all the measured parameters of the lake displayed higher concentration levels in September versus May (t test, p < 0.05) (Table 1). This observation may be explained by the increased pollutant loadings into the lake for the rainy season.

Our results based on the selected water quality parameters demonstrated that the major pollutant sources of the lake stemmed from the upper streams and the industrial channels (i.e., the upstream catchments). It was also notable that the water quality of the lake might be strongly affected by the dilution from the seawater exchange, and it varied seasonally probably due to the catchment runoff. Previous studies on metal and organic contamination of this lake have also shown that the deterioration of the lake water quality originated from anthropogenic activities occurring in the upper catchments (Ra et al. 2011; Moon et al. 2012).

Spectroscopic properties of DOM

SUVA has been used as a measure of DOM aromaticity in water samples (Weishaar et al. 2003). For this study, the average SUVA values exhibited large spatial and seasonal variations, ranging from 0.7 to 2.9 L mg C−1 m−1 (Table 2). However, no seasonal difference was found for the lake. Instead, the samples from the upper streams showed lower SUVA values in September than in May (t test, p < 0.05), and the estuary had the same trend. The results suggest that more nonaromatic DOM components could be brought into the estuary presumably from the upstream forested and agricultural areas in the rainy season. No distinct seasonal difference was found for the industrial channels.

Table 2 Spectroscopic properties (mean ± standard deviation) of DOM for the four different sampling regions at two different sampling periods
Full size table

Significant positive relationships between CDOM absorption and DOC concentrations were previously reported in aquatic environments, and the CDOM absorption coefficient at 375 nm (i.e., a375) has often been used as a surrogate for colored DOM (Fichot and Benner 2012; Asmala et al. 2012). In this study, the average a375 values of the lake were 1.1 and 0.7 m−1 in May and September, respectively (Table 2), which fell within the ranges reported from Danish estuaries and coastal water (Stedmon et al. 2000) and from the open sea (Kowalczuk et al. 2005). The average a375 values of the catchment areas and the estuary had much higher levels, ranging from 6.7 to 9.5 m−1, indicating that the external DOM sources may contain a larger fraction of chromophoric organic matter within the total pool of DOM compared to the lake DOM. Photobleaching may be involved in lowering a375 values in the lake (Stedmon et al. 2000).

The spectral slope of S275–295 ranged from 13.8 to 22.9 μm−1 (Table 2), which fell into the range reported from three Finnish estuaries (Asmala et al. 2012). The spectral slopes of the lake were higher than those of the upstream catchments for both sampling periods (t test, p < 0.05), which was consistent with the previous studies reporting the higher spectral slope values of water samples with increasing salinity (Helms et al. 2008; Asmala et al. 2012; Fichot and Benner 2012). Based on an inverse relationship previously established between the spectral slope and the molecular weight of DOM (Helms et al. 2008), our result suggests that the molecular size of the lake DOM may be smaller than those of the upstream areas probably due to photobleaching.

The spectral slope ratio (i.e., SR or S275–295/S350–400) can be utilized as an indicator of photobleaching (Helms et al. 2008). For this study, the slope ratios presented the range between 1.0 and 2.4 (Table 2). Seasonal differences were observed for all the four classified regions of the lake, and the catchment areas with higher values shown in May versus September (t test, p < 0.05) and the estuary had the same trend. This result could be attributed to the seasonal difference in the local solar radiation. Longer monthly sunshine hours were reported for this study area in May (525–540 h per month) compared to September (420–435 h per month) (http://www.greenmap.go.kr). The observed seasonal differences in the spectral slope ratios were much pronounced for the lake (i.e., 128.4 % difference) than for the upstream catchments (41.6 % difference) and the estuary (43.1 % difference), presumably due to longer residence time and photobleaching effects on DOM (i.e., lower molecular weight) (Helms et al. 2008).

For this study, all the samples showed similar FI values ranging from 1.6 to 1.8. However, the average HIX values of the upstream catchments exhibited higher levels compared to those of the lake, and higher HIX values were observed for the lake in September than in May (t test, p < 0.05). The HIX information implies that, for the rainy season, more humified and condensed DOM structures from the upstream catchments may be accumulated in the lake, subsequently affecting the composition of the lake DOM.

Fluorescence EEM

Four peaks were identified from the EEM plots of all the samples (Fig. 2). The peak maxima were located at the excitation/emission wavelengths of <250/410–430 nm, 310–330/410–430 nm, 280–300/410–430 nm, and 270–280/330–360 nm. Following the definitions of Coble (2007), they can be categorized as humic-like peak A, humic-like peak C, marine humic-like peak M, and tryptophan-like peak T, respectively. Seasonal differences were found in the EEM peaks. Except for the lake, all the peak maxima were higher in May than in September (Table 2). The differences were even more pronounced compared to those of the DOC concentrations (Table 1), suggesting that seasonal variability may be much greater for fluorescent DOM compounds compared to the total pool of DOC. Peak M was also observed for the samples from the catchments, not only from the lake, implying that the marine humic-like fluorescence features could be present for the areas where biological activity is much involved (i.e., anthropogenic and/or agricultural sources (Hudson et al. 2007).

Fig. 2
figure 2

EEM plots of DOM samples in May (left side) and in September (right side). a, b Surface samples from L2 site in the lake. c, d Samples from E1 site in the estuary. e, f Samples from I4 site in the industrial channels. g, h Samples from U1 site in the upper streams

Full size image

EEM-PARAFAC components

Three fluorescence components (C1, C2, and C3) were decomposed from the EEM measurements for all 38 samples from the two seasons by the PARAFAC modeling (Fig. 3). The overlap of the excitation and the emission loadings of the two halves on the whole dataset confirmed that the three components were appropriate to represent the measured fluorescence EEM dataset (Stedmon et al. 2003). No outliers were found. C1 exhibited a primary (and secondary) fluorescence maximum intensity at an excitation/emission wavelength of <250 (280)/382 nm. This component was analogous to the peak M in the previous EEM plots, which was reportedly associated with the primary production of phytoplankton in oceans and freshwaters (Coble 2007). Similar PARAFAC components were reported for other DOM samples, and they were assigned as microbial humic-like components in prior studies (Stedmon et al. 2003; Yamashita et al. 2011). The appearance of the peak may also be attributed to the production of new CDOM due to microbial degradation of labile organic substitutes (Hur et al. 2009; Zhang et al. 2011). From incubation experiments using axenic cultures of phytoplankton, for example, (Romera-Castillo et al. 2010) reported that the microbial humic-like components were possibly generated by the biotransformation of algal-derived organic matters. For our study, however, no significant correlation was observed between Chl a and the PARAFAC component. Instead, the highest correlation was found between C1 and DOC concentrations (r = 0.793, p < 0.0001) among the three PARAFAC components, suggesting that anthropogenic organic matter may be largely involved for the presence of the component. Similarly, Zhang et al. (2011) claimed that this component was highly associated with the anthropogenic factors such as land use and human activities of the upstream catchments for the downstream Tianmuhu lake, China.

Fig. 3
figure 3

PARAFAC model output showing the three fluorescent components (left) and the corresponding excitation/emission loadings (right). Excitation/emission loadings consist of two independent halves of the dataset (red and blue dotted lines) and the complete dataset (black lines)

Full size image

C2 presented a fluorescence peak at the excitation/emission wavelengths of 280/320 nm, which resembles Peak T in the EEM plots (Coble 2007). This component is commonly observed for many aquatic samples that are subject to anthropogenic influences in bays, estuaries, coastal areas, as well as the areas of high primary productivity (Stedmon and Markager 2005; Yamashita et al. 2008; Hong et al. 2012). Although many studies implied substantial contribution of phytoplankton to the component of coastal and lake waters (Hanamachi et al. 2008; Zhang et al. 2009), there is still lack of the correlation between Chl a and the component (i.e., C2) (Yamashita et al. 2011). This study also showed no direct relationship between Chl a and C2 (r = 0.076; p = 0.651). Instead, there was a close association between C1 and C2 (r = 0.978, p < 0.0001), suggesting that the two fluorophore groups may be similar to each other in their organic sources.

C3 exhibited a primary and a secondary fluorescence maxima at the excitation/emission wavelengths of <250/425 nm and 330/425 nm, respectively. The component appears to be the mixture of the two humic-like fluorescence peaks previously observed (i.e., peak A and peak C). Prior studies demonstrated that the component 3 might be either terrestrially derived or possibly produced autochthonously in aquatic environments largely affected by terrestrial input (Stedmon and Markager 2005; Guo et al. 2011; Zhang et al. 2011; Murphy et al. 2008).

The relative distributions of the three PARAFAC components, expressed by the percent component, were compared for all the sampling regions at the two different sampling periods in Fig. 4. Irrespective of the sampling seasons, the highest abundance of C1 over all the three components (i.e., %C1) was observed for the industrial channels, suggesting that the component may be closely related to the anthropogenic organic pollution. Meanwhile, %C3 was the highest among the three components for the upper streams and the estuary, implying a close association between C3 and terrestrial carbon sources (Fig. 4). Except for the industrial channels, C3 appears to be more pronounced for all the samples in September than in May, also suggesting that terrestrial organic input from the forested and agricultural catchments into the lake may be intensified during the rainy season.

Fig. 4
figure 4

Relative distributions (%) of the three PARAFAC components for the four different sampling regions in a May and b September

Full size image

PCA for the interpretation of spatial and seasonal variations in DOM

The basic matrix in PCA was constructed based on 15 different variables of each sample (Fig. 5). The sampling site of I1 was excluded from the PCA because of a concern that the extremely high concentrations of the general water quality parameters may yield bias in the interpretation of the results. For this study, the first two principal components explained over 69.9 % of the total data variance. Correlations between the first two principal components and the measured properties were depicted in a loading plot (Fig. 5a). The first principal component (PC1) was responsible for the 50.3 % variance. Positive high loadings on PC1 were observed for the variables of salinity, %C2, and S275–295, whereas the high negative loadings were observed for a375, HIX, %C3, and nutrients. PC2 represents 19.7 % of the total variability, and it has positive loading with organic-matter-related water quality parameters and %C1 and negative loadings with %C3 and HIX (Fig. 5a). From the overall locations of the variables in the loading plot, it can be postulated that PC1 is an indicator to discriminate between marine organic matter and anthropogenic and/or terrestrial organic matter originating from the upstream catchments, whereas PC2 is related to the concentrations of DOM. The location of %C2 close to salinity suggests that the presence of C2 may be associated with autochthonous production in marine environments.

Fig. 5
figure 5

Principal component analysis of selected water quality parameters and spectroscopic properties of DOM for all the samples with the exception of I1 site (see the text). a Loading plot. b Score plot. The hollow and solid symbols represent the samples collected in May and September, respectively. Note that Shapiro-Wilk test confirmed normality in the distribution of each parameter included in the PCA (p < 0.05)

Full size image

Additional information on seasonal differences was obtained from the score plot (Fig. 5b). The lake DOM showed the unique characteristics distinguished from either the industrial channels or the upper streams, as demonstrated by the distinctive location of the lake samples from the two catchment sources. Furthermore, the lake samples could be easily separated into two seasons in the plot, as the September samples were located toward the left side of the PC1 axis than those of May. This observation suggests that, in September, the lake DOM is likely to be more affected by terrestrial input from the upstream catchments. Such seasonal differences were observed for other sampling regions, as the data points corresponding to the monsoon season were located in the lower positions along the PC2 axis, indicating that more condensed organic carbon components of allochthonous origin may be more dominant for the rainy season irrespective of the sampling sites. Our PCA results provided additional evidence that the dilution effects from the seawater exchange may govern the characteristics of DOM and the water quality in the lake. The variability of the lake DOM appears to be partially affected by seasons as well, exhibiting higher contribution of the organic input from the upper catchments to the lake for the monsoon versus the premonsoon seasons.

Conclusions

Although the upstream catchments were the major sources of DOM and nutrients to the coastal lake, much lower concentration levels were observed in the lake due to the dilution effect from the seawater exchange. The CDOM properties of the lake were distinguished from those dissolved organic matter sources from the upstream catchments. Lower absorption coefficients, HIX values, fluorescence EEM peak intensities, and higher spectral slopes were obtained from the lake samples, in comparison to values of the upper catchments, suggesting that the lake DOM is characterized by lower molecular size and less condensed structures. Seasonal differences were also found for the upper streams and the lake, in which higher absorption coefficients and fluorescence peak intensities but lower spectral slopes and HIX values were exhibited for the premonsoon versus the monsoon season. In contrast, the seasonal differences were much less pronounced for the industrial channels except for fluorescence peaks. PARAFAC modeling revealed that a combination of three fluorophore groups explained the variations in the fluorescence features of the DOM samples. The high abundance of C1 component (microbial humic-like) in the samples from the industrial channels suggests that this component may be associated with anthropogenic organic pollution. C3 component (terrestrial humic-like) was the most enriched for the samples from the upper streams and the estuary, and the relative distribution was greater for the rainy season. Our PCA results based on spectroscopic properties of DOM revealed that although the lake DOM was the most likely to be affected by the dilution effect, the contribution of allochthonous DOM input from the upstream catchments to the lake might be elevated during the rainy season.