The use of the self-organizing map (SOM) methodology to study the dissolved organic matter (DOM) in different water body types

absolute principal component-multiple linear regression, dissolved organic matter, fluorescence spectroscopy, machine learning, non-point source pollution, self-organizing map

This study investigated the characteristics of dissolved organic matter (DOM) in two distinct water bodies, through the utilization of three-dimensional fluorescence spectroscopy coupled with self-organizing map (SOM) methodology. Specifically, this analysis concentrated on neurons 3, 14, and 17 within the SOM model, identifying notable differences in the DOM compositions of a coal subsidence water body (TX) and the MaChang Reservoir (MC). The humic substance content of DOM TX exceeded that of MC. The origin of DOM in TX was primarily linked to agricultural inputs and rainfall runoff, whereas the DOM in MC was associated with human activities, displaying distinctive autochthonous features and heightened biological activity. Principal component analysis revealed that humic substances dominated the DOM in TX, while the natural DOM in MC was primarily autochthonous. Furthermore, a multiple linear regression model (MLR) determined that external pollution was responsible for 99.11% of variation in the humification index (HIX) of water bodies.

HIGHLIGHTS

Listen

Study of the variation in DOM spectral characteristics of different waterbodies.
Neural networks were applied to spectral resolution.
Source analysis was performed to identify the origin of water pollution.

INTRODUCTION

Listen

Dissolved organic matter (DOM) is a soluble fraction of organic matter, consisting of humic-like substances, proteins, hydrophilic organic acids, and polysaccharides, with complex chemical properties and environmental effects (Zhang et al. 2020). DOM exists widely in aquatic environments, such as lakes, rivers, and soils, typically originating from rainfall runoff, soil leaching, surface runoff, and sewage discharge (Inamdar et al. 2012). DOM rich in elements (such as C, N, and P) affects carbon and nutrient cycles, participating in geochemical and photochemical reactions (Fellman et al. 2011; Shatrughan et al. 2013). Previous studies have shown that the chemical composition and properties of DOM in aquatic environments are closely related to their source, while also being affected by the soil–water transport of DOM and release processes from submerged silt (Sulzberger & Durisch-Kaiser 2009; Yang et al. 2014). Therefore, understanding the sources and characteristic properties of DOM in aquatic environments is essential to better understand the role of this complex fraction in different ecosystems.

A considerable body of research is available on DOM in water (Zhang et al. 2010). Studies indicated that the dynamic changes in DOM and the release of endogenous nutrients can contribute to eutrophication, such as in the Chaohu Lake (China), causing the deterioration of lake ecosystems (Bao et al. 2023). In coal-mining areas, high groundwater levels can lead to land subsidence, with only about half of the subsidence areas in Huainan (China) being actually submerged in water (Hu et al. 2014; Xiao et al. 2014). Large-scale ground subsidence is an important issue and it alters the local topography and brings about fundamental changes in the structure and function of local ecosystems, significantly affecting the surrounding environment. Currently, some coal-mining subsidence areas have been designated as water source protection zones, fishing areas, and wetland ecological restoration zones, exhibiting considerable variations in their nutrient status. With the continuous increase in coal-mining subsidence areas and the ongoing deterioration of mining area ecological environments, some subsidence areas are experiencing aquatic eutrophication or severe level of pollution.

The self-organizing map (SOM) methodology is an unsupervised artificial neural network (ANN) technique that exhibits powerful clustering capabilities when dealing with nonlinear data, constructing a competitive learning network by Santos et al. (2020) reducing high-dimensional data matrices into one- or two-dimensional topological structures. The SOM model is composed of interconnected neurons forming an adaptive system. During the learning phase, these neurons adjust their structure and weight vectors between neurons based on the specific input features, minimizing the overall error (Kohonen 2013). As an unsupervised clustering algorithm, SOM classifies all participating objects, without specifying the name of the classified object. Compared to parallel factor analysis (PARAFAC), which has been widely used for the analysis of fluorescence data, the SOM approach demonstrates higher noise tolerance, without the need for a cumbersome data preprocessing stage. Additionally, SOM has excels in the analysis of short-excitation wavelength fluorescence data (Jin et al. 2023). By employing SOM models to predict fluorescence data from two different types of water body, the distribution characteristics of DOM can be effectively established.

This study utilized the three-dimensional fluorescence spectroscopy (3D-EEMs) technology combined with an SOM model to investigate samples collected from the MaChang Reservoir (MC) (Kongxiangdian, Huainan City, China) and a coal subsidence waterbody in Suntuan (Huaibei City, China). The relationships between DOM content and structure were investigated in both the coal subsidence waterbody and reservoir samples, while also determining other water quality parameters, for establishing distinct nutrient patterns for different waterbodies.

MATERIALS AND METHODS

Listen

Study area

Listen

The study area used for this research encompassed the coal subsidence water body (TX) in the Suntuan Mining Area (Suixi County, Huibei City, China) and the MC in Kongxiangdian (Huainan City, China), with TX dominantly affected by human activities as it is surrounded by villages and agricultural lands, while MC is situated at a higher elevation and is primarily influenced by natural factors. Water samples were collected from MC in December 2022 and from TX in May 2023. A total of 24 water samples were collected during the two sampling campaigns, consisting of seven TX samples (TX1–TX7) and 17 MC samples (MC1–MC17), as illustrated in Figure 1.

Figure 1

Map of the study areas and sample collection point locations.

Sample analysis

Listen

Temperature (T), conductivity (EC), salinity, pH, turbidity, total dissolved solids (TDS), and dissolved oxygen (DO) were measured in the field, while total nitrogen (TN), ammonia nitrogen (⁠⁠), nitrate nitrogen (⁠⁠), total phosphorus (TP), and chemical oxygen demand (COD) were determined in the laboratory, as shown in Table 1. Water samples were filtered through 0.45 μm glass fiber membranes for three-dimensional fluorescence spectra (EEMs) and ultraviolet-visible absorption spectra (UV-vis) analysis.

Table 1

Water indices and the experimental methods applied

Parameter	Experimental method
TN	Water quality – determination of the total nitrogen-alkaline potassium persulfate digestion UV spectrophotometric method (HJ 636-2012)
	Water quality – determination of ammonia nitrogen-Nessler's reagent spectrophotometry (HJ 535-2009)
	Water quality – determination of nitrate-nitrogen-ultraviolet spectrophotometry (HJ/T 346-2007)
TP	Water quality –determination the of total phosphorus-ammonium molybdate spectrophotometric method (GB 11893-89)
COD	Water quality – determination of the chemical oxygen demand-dichromate method (HJ 828-2017)

Parameter	Experimental method
TN	Water quality – determination of the total nitrogen-alkaline potassium persulfate digestion UV spectrophotometric method (HJ 636-2012)
	Water quality – determination of ammonia nitrogen-Nessler's reagent spectrophotometry (HJ 535-2009)
	Water quality – determination of nitrate-nitrogen-ultraviolet spectrophotometry (HJ/T 346-2007)
TP	Water quality –determination the of total phosphorus-ammonium molybdate spectrophotometric method (GB 11893-89)
COD	Water quality – determination of the chemical oxygen demand-dichromate method (HJ 828-2017)

DOM measurement and characterization

Listen

The absorbance of water samples was measured (200–800 nm) using a UV-Vis photometer (N5000PLUS, Yoke, Shanghai, China). Milli-Q water was applied as the blank sample for determinations, which were performed using a quartz cuvette with a 1 cm optical range. The scanning interval was 1 nm and spectral calibration was performed using the 680–800 nm band. EEMs’ data were obtained for water samples (F-7000, Hitachi, Japan) using an excitation wavelength (Ex) from 200 to 500 nm, an emission wavelength (Em) from 200 to 550 nm, a scanning speed of 1,200 nm/min, and a scanning interval of 5 nm. Measurements were done using a 1 cm quartz cuvette and EEMs data for ultrapure water were simultaneously recorded as a quality control reference.

Data analysis and processing tools

Listen

This study employed Excel and IBM SPSS Statistics v.27 for statistical correlation analysis, while principal component analysis (PCA) was conducted using the ‘ggplot2’ and ‘ggrepel’ packages of R v.4.2.3. SOM processing of EEM data, determination of the K-means clustering algorithm, DBI index analysis, and the establishment of SOM models were performed using the ‘dream’, ‘statisticstoolbox’, and ‘somtoolbox’ tools in Matlab v.2022b (Agoubi 2018; Chen et al. 2022). The PARAFAC model was established using the ‘stardom’ v.1.1.21 package in R (Krylov et al. 2020), while graphical representations were prepared using ArcMap v.10.8.1, Bigemap GIS Office, Origin v.2022b (educational version), Matlab, and the ggplot2 package in R.

RESULTS AND DISCUSSION

Listen

SOM model analysis

Listen

The SOM model was used to cluster the EEMs dataset, with the clustering results then used as the output for the data mapping method. The SOM model algorithm consisted of two steps: self-organizing training and output feature mapping, requiring the adjustment of weights and neural network parameters in the input layer to connect output layer neurons with input layer neurons. The specific steps involved in this process have been described in detail in previously reported literature (Bieroza et al. 2011).

The SOM model was trained using Matlab v.2022b, in which fluorescence data were assigned to best matching units (BMUs). The final model results were presented in a grid, with each unit representing an original EEM spectrum, allowing the spectral features of these units to gradually be adjusted based on the input data characteristics. By the training of 24 samples, a neural network grid (8 × 3) was ultimately determined as the mapping size, with additional mapping units in the U-matrix showing the distance between different neurons based on color depth (Figure 2). To better determine the number of clusters, a combination of the K-means algorithm and Davies–Bouldin Index (DBI) was employed. The results showed that when the cluster number was 2, the DBI showed its lowest value, suggesting that the fluorescence spectral region could be effectively divided into two clusters.

Figure 2

The self-organizing map (left) and the U-matrix (right).

Due to the tendency for the SOM model to label similar EEMs and retain labels with the highest frequency of neuron hits, significant differences between the two water samples are shown in Figure 3. TX sample data were dominantly distributed on the left side of the SOM map, while MC sample data were situated on the right side, with a more concentrated distribution of neurons. The click counts of neurons are shown in Figure 3, with the neurons at the map edges having the highest hit frequencies. Among them, neuron 3 has a higher hit rate for TX and therefore, and represents the fluorescence characteristics of TX, while neurons 14 and 17 represent the fluorescence characteristics of MC.

Figure 3

Histogram of SOM neuron hits in different water samples.

The visualization results for neurons 3, 14, and 17 are presented in Figure 4, showing that neurons 3 and 17 (a) exhibited similar color differentiations in corresponding components, indicating a certain degree of correlation between these neurons.

Figure 4

SOM-based visualization diagram of the neuron 3, neuron 14, and neuron 17.

Three representative neurons derived from EEMs data were analyzed, revealing distinctive features at different excitation/emission wavelengths. Neuron 3 exhibited primary excitation and emission peaks at Ex/Em = 250/380 nm. Neuron 14 (a) exhibited primary excitation and emission peaks at Ex/Em = 280/330 nm, while neuron 14 (b) displayed secondary excitation and emission peaks at Ex/Em = 280/360 nm. The primary excitation and emission peaks of Neuron 17 (a) were present at Ex/Em = 240/350 nm, while neuron 17 (b) exhibited secondary excitation and emission peaks at Ex/Em = 260/400 nm (Figure 5). Both neuron 3 and neuron 17 (b) had emission peaks in the short-wave ultraviolet region, indicating UV-type humic substances, with fluorescence signals resembling ‘A’ peaks (Dalmagro et al. 2019). The fluorescence signal of neuron 17 (a) was similar to that of the ‘T2’ peak, while the emission signal of neuron 14 was similar to the ‘T1’ peak, both of which belong to tryptophan-like substances. The secondary peaks of neuron 3 and neuron 17 were similar, indicating humic substances originating from terrestrial inputs, microbial activity within the waterbody and photochemical oxidation (Fellman et al. 2008). Neuron 3, representing TX, was influenced by surrounding agriculture and aquaculture activities. Neuron 17 (a), representing MC, was associated with humic substance components related to anthropogenic activities. Neuron 14 and neuron 17 (b), representing MC, were associated with proteins and linked to the growth and reproduction of autochthonous planktonic organisms and microbial activities (Artifon et al. 2019).

Figure 5

Three-dimensional fluorescence spectral information or neuron 3, neuron 14, and neuron 17.

Under natural conditions, the composition of DOM is complex and susceptible to the effects of biogeochemical processes and human activities (Du et al. 2021). TX was located near residential villages and agricultural land, resulting in a significant influence from external inputs, with humic substances being dominant. Both MC and TX contained humic substance signals in the visible light range, which are primarily generated associated with forested streams, agricultural discharge, and wetlands, with a combined influence from terrestrial inputs, water microbial activity, and photochemical oxidation processes (Fellman et al. 2008). In the case of MC, humic substances were likely to be produced by aquatic plants, other organisms and biological processes within the waterbody, while also being influenced by sewage discharge from surrounding anthropogenic activities although these inputs have a relatively low humic substance content (Yu et al. 2020). Additionally, the poor mobility of water and the long hydraulic retention time, coupled with the significant influence of bioavailable small molecules (anthropogenic origin), contributed to the high humic substance content in the water (Zhang et al. 2006; Solomon et al. 2015; Gao & Wang 2019).

Fluorescence parameters

Listen

The fluorescence indices, such as the fluorescence index (FI), biological index (BIX), and humification index (HIX), were analyzed to determine the spectral characteristics of DOM in water (Huguet et al. 2009; Lavonen et al. 2015), with the results presented in Table 2. The fluorescence indices FI and BIX exhibited no significant differences between the two waterbody types. The FI values for both waterbodies ranged from 1.4 to 1.9, indicating both external and internal DOM sources. Similarly, analysis of the biological index BIX showed that the average BIX values for DOM ranged between 0.8 and 1.0, suggesting that DOM originated from both terrestrial and internal sources, which is consistent with the FI index results.

Table 2

Water fluorescence index

Water	FI	BIX	HIX**
MC	1.73 ± 0.11	0.98 ± 0.056	0.66 ± 0.059
TX	1.69 ± 0.13	0.95 ± 0.063	0.85 ± 0.034

Note: Numerical values are presented as mean ± standard deviation (SD).

**denotes significant differences between the same indicators (P < 0.01).

The degree of DOM humification in water is characterized using the HIX, where a higher HIX value indicates a higher degree of DOM aromaticity (Huguet et al. 2009). The mean HIX value for TX (0.66 ± 0.059) was significantly higher than for MC (0.85 ± 0.034) (P < 0.01). However, both HIX values were <1.5, indicating that the DOM in water exhibited the characteristics of weakly humified substances, suggesting that in the short term, DOM was primarily influenced by algal or bacterial organic matter.

DOM in natural water bodies is primarily influenced by terrestrial inputs and microbial turnover (Johnston et al. 2019). The endogenous fluorescent signals present in MC water samples were pronounced, resembling those reported for artificially regulated water (Jiang et al. 2018; Yingxin et al. 2021). The MC BIX values further confirmed that biogenic DOM production was stimulated by anthropogenic and natural factors. FI values indicated that endogenous and exogenous sources had a combined influence, with anthropogenic inputs having a dual effect on DOM microbial processes. On the one hand, anthropogenic DOM inputs enhance microbial activities in the water, leading to the degradation of external inputs and an increase in the endogenous contribution (Lambert et al. 2017). On the other hand, anthropogenic inputs disrupt the balance of DOM in the aquatic environment, disturbing the aquatic environment, and hindering microbial adaptation to environmental changes, thereby suppressing the growth of endogenous microbial sources within a certain timeframe (Butman et al. 2015).

Anthropogenic influences can alter the composition of DOM in natural water bodies, affecting source determination. Previous studies have indicated that agricultural land can be responsible for a significant influx of terrestrial soil into waterbodies, leading to an increase in exogenous DOM (Wilson & Xenopoulos 2009). The high frequency of agricultural activities around TX is likely to contribute to changes in the nutrient composition and organic content of water, leading to the enrichment of terrestrial and biogenic humic substances. The input of large amounts of bioavailable organic matter alters the metabolism of microbes and algae within the waterbody, increasing DOM degradation and reducing its production, resulting in endogenous sources being dominant. This pattern is similar to those reported previous studies on the contribution of anthropogenic sewage discharge on DOM in natural water bodies (Williams et al. 2016).

Principal component analysis

Listen

PCA was performed to explore the correlation between water quality parameters in the subsidence area and those in natural waterbodies (Table 3) (He et al. 2021). The Kaiser–Meyer–Olkin (KMO) measure and Bartlett's Test of Sphericity yielded a KMO value of 0.707 (Sig < 0.001), indicating that the samples met the requirements for use in PCA. The results of PCA revealed a cumulative contribution rate of 67.3% within the water body (Figure 6).

Table 3

Physical–chemical indicators determined for each waterbody

Water type	EC (μS·cm⁻¹)	TDS (g·L⁻¹)	pH	DO (mg·L⁻¹)	TN** (mg·L⁻¹)	TP** (mg·L⁻¹)	** (mg·L⁻¹)	(mg·L⁻¹)	COD**(mg·L⁻¹)
MC	545.85 ± 184.61	0.35 ± 0.12	8.12 ± 0.29	8.72 ± 0.65	3.40 ± 1.46	0.22 ± 0.15	0.97 ± 0.59	0.39 ± 0.15	19 ± 5
TX	648.97 ± 50.67	0.42 ± 0.034	8.22 ± 0.92	8.85 ± 2.16	0.79 ± 0.14	0.71 ± 0.23	0.11 ± 0.01	0.31 ± 0.13	72 ± 20

Water type	EC (μS·cm⁻¹)	TDS (g·L⁻¹)	pH	DO (mg·L⁻¹)	TN** (mg·L⁻¹)	TP** (mg·L⁻¹)	** (mg·L⁻¹)	(mg·L⁻¹)	COD**(mg·L⁻¹)
MC	545.85 ± 184.61	0.35 ± 0.12	8.12 ± 0.29	8.72 ± 0.65	3.40 ± 1.46	0.22 ± 0.15	0.97 ± 0.59	0.39 ± 0.15	19 ± 5
TX	648.97 ± 50.67	0.42 ± 0.034	8.22 ± 0.92	8.85 ± 2.16	0.79 ± 0.14	0.71 ± 0.23	0.11 ± 0.01	0.31 ± 0.13	72 ± 20

Note: Numerical values are presented as mean ± standard deviation (SD).

**denotes significant differences between the same indicators (P < 0.01).

Figure 6

Principal component analysis of TX and MC.

To simplify the structure of the factor loading matrix and aid in source apportionment, two principal components extracted from PARAFAC were utilized as factors. The contribution rate of the first principal component was 54.4%, which was characterized by a higher loading of factor variables in fluorescence components and TN, this component was primarily influenced by the proteinaceous fraction (Figure 6(a)). The contribution rate of the second principal component was 12.9%, exhibiting a high loading on the BIX. Previous studies showed that humic substances produced by microbial activities exhibit a good correlation with nitrogen-containing elements (Hur & Cho 2012), suggesting that the second principal component primarily reflects the influence of endogenous microbial sources (Figure 6(a)).

The two water samples were subjected to Adonis multivariate analysis of variance (based on Euclidean distance) with 999 permutations (P < 0.001) (Sylvain et al. 2019). As shown in Figure 6(b), TX was distributed along the positive axis of PC1, indicating that TX DOM was dominated by humic substances, while MC DOM was associated with autochthonous sources.

Source apportionment

Listen

The absolute principal component score-multiple linear regression (APCS-MLR) receptor model is a backward tracing method, which standardizes data to calculate the absolute principal component scores (APCS) of different factors. Subsequently, using a multiple linear regression model, the contribution rates of various pollution sources can be quantified, with the predicted results linearly fitted with the measured results. In recent years, this method has been widely used in source apportionment studies in various environments, including soil, groundwater, and surface water (Guowei et al. 2020). Based on the PCA determined composition and distribution characteristics of major pollution sources within the study area, the APCS-MLR model was employed to establish functional relationships between pollution sources and the fluorescence parameters (FI, BIX, and HIX), TN,

⁠, and

(Figure 7), providing further validation of the results of factor analysis results (Table 4). The linear fitting R² between the predicted and actual values of various indicators in the study area ranged from 0.65 to 0.96, indicating good consistency overall. Furthermore, the similarity between model-predicted values and actual measured values suggests that the APCS-MLR model is suitable for performing pollution source apportionment calculations in aquatic environments.

Table 4

Contribution of public factors to indicators

Parameter	Contribution rate %		Measured average concentration	Predicted average concentration	R²
Parameter	F1 exogenous pollution	F2 endogenous pollution	Measured average concentration	Predicted average concentration	R²
FI	87.21%	12.79%	1.72 ± 0.11	1.72 ± 0.074	0.66
HIX	99.11%	0.89%	0.72 ± 0.10	0.72 ± 0.081	0.82
BIX	20.94%	79.06%	0.97 ± 0.063	0.97 ± 0.047	0.74
TN	78.51%	21.49%	2.59 ± 1.61	2.59 ± 1.55	0.97
NH + 4-N	68.85%	31.35%	0.74 ± 0.65	0.74 ± 0.57	0.88
NO- 3-N	43.69%	56.31%	0.36 ± 0.064	0.36 ± 0.053	0.83

Parameter	Contribution rate %		Measured average concentration	Predicted average concentration	R²
Parameter	F1 exogenous pollution	F2 endogenous pollution	Measured average concentration	Predicted average concentration	R²
FI	87.21%	12.79%	1.72 ± 0.11	1.72 ± 0.074	0.66
HIX	99.11%	0.89%	0.72 ± 0.10	0.72 ± 0.081	0.82
BIX	20.94%	79.06%	0.97 ± 0.063	0.97 ± 0.047	0.74
TN	78.51%	21.49%	2.59 ± 1.61	2.59 ± 1.55	0.97
NH + 4-N	68.85%	31.35%	0.74 ± 0.65	0.74 ± 0.57	0.88
NO- 3-N	43.69%	56.31%	0.36 ± 0.064	0.36 ± 0.053	0.83

Figure 7

Assessing hydrothermal groundwater flow path using Kohonen's SOM, geochemical data, and groundwater temperature cooling trend

Comparison of the predicted and measured concentrations for pollution parameters and indices.

Based on the constructed APCS-MLR and pollution source contribution models, the contributions of pollution sources to water parameters were deduced (Table 4), showing that exogenous pollution contributed 99.11% of HIX. Therefore, nutrient-rich nitrogen pollutants in the water originated from agricultural and domestic sewage discharges in the surrounding areas. Autochthonous pollution contributed 79.06% of BIX, indicating that microbial activities were the primary endogenous source. These calculations demonstrate that the sources of DOM were influenced by exogenous and endogenous pollution, in which exogenous pollution was mainly attributed to anthropogenic inputs and agricultural non-point source pollution, while endogenous sources were primarily influenced by the intensification of microbial activities.

CONCLUSIONS

Listen

(1) The content of humic components affected by agricultural non-point source pollution was higher in the DOM from the subsidence waterbody than from natural reservoir water.
(2) The results of the SOM model showed that neurons 14 and 17 (a) representing water samples from the MC exhibited characteristics of protein-like substances, indicating a higher contribution from autochthonous biological activity. Neurons 3 and 17 (b), representing humic-like substances, suggested that the DOM in TX mainly originated from agricultural non-point source pollution, while the humic substances in MC were associated with anthropogenic activities in the surrounding area.
(3) The APCS-MLR model results indicated that humic-like components of DOM in the water primarily originated from agricultural non-point source pollution, with an exogenous pollution contribution rate to HIX of 99.11%. The contribution rate of endogenous sources to BIX was 79.06%, suggesting that proteins in the water originated from plankton and microorganisms.
(4) PCA yielded consistent results in SOM model analysis. The SOM model, based on the original excitation–emission matrix fluorescence spectroscopy (EEMs) data for water samples, effectively analyzed the components of DOM in both waterbodies. Simultaneously, the multiple linear regression model can be used to incorporate monitoring data from various waterbodies to help predict the relative contribution of different pollution sources, providing support for environmental pollution source apportionment studies.

DISCLOSURE STATEMENT

Listen

This study was supported by the Anhui Provincial Key Research and Development Project (202004i07020012).

DATA AVAILABILITY STATEMENT

Listen

All relevant data are included in the paper or its Supplementary Information.

CONFLICT OF INTEREST

Listen

The authors declare there is no conflict.

REFERENCES

Agoubi

B.

2018

.

Environmental Science and Pollution Research International

25

,

13597

–

13610

.

https://doi.org/10.1007/s11356-018-1525-1

.

Aquatic organic matter: Classification and interaction with organic microcontaminants

Artifon

V.

,

Zanardi-Lamardo

E.

&

Fillmann

G.

2019

.

Science of the Total Environment

649

,

1620

–

1635

.

https://doi.org/10.1016/j.scitotenv.2018.08.385

.

Changes of DOM and its correlation with internal nutrient release during cyanobacterial growth and decline in Lake Chaohu, China

Bao

Y.

,

Huang

T.

,

Ning

C.

,

Sun

T.

,

Tao

P.

,

Wang

J.

&

Sun

Q.

2023

.

Journal of Environmental Sciences

124

,

769

–

781

.

https://doi.org/10.1016/j.jes.2022.02.019

.

Exploratory analysis of excitation–emission matrix fluorescence spectra with self-organizing maps – a tutorial

Bieroza

M.

,

Baker

A.

&

Bridgeman

J.

2011

.

Education for Chemical Engineers

7

, 112–116,

e22

–

e31

.

https://doi.org/10.1016/j.ece.2011.10.002

.

Increased mobilization of aged carbon to rivers by human disturbance

Butman

D. E.

,

Wilson

H. F.

,

Barnes

R. T.

,

Xenopoulos

M. A.

&

Raymond

P. A.

2015

.

Nature Geoscience

8

, 112–116.

https://doi.org/10.1038/NGEO2322

.

Spectral characteristic changes of dissolved organic matter in aquatic systems under the influences of agriculture and coal mining

Chen

X.

,

Gao

L.

,

Chen

X.

,

Pang

Z.

,

Ge

J.

,

Zhang

H.

,

Zhang

Z.

,

Mu

M.

,

Qiu

Y.

&

Zhao

X.

2022

.

Polish Journal of Environmental Studies.

31

,

2549

–

2561

.

https://doi.org/10.15244/pjoes/144660

.

Streams with Riparian forest buffers versus impoundments differ in discharge and DOM characteristics for pasture catchments in Southern Amazonia

Dalmagro

H. J.

,

Lathuillière

M. J.

,

Sallo

F. D. S.

,

Guerreiro

M. F.

,

Pinto

O. B.

,

Arruda

P. H. Z. D.

,

Couto

E. G.

&

Johnson

M. S.

2019

.

Water

11

(2), 390.

https://doi.org/10.3390/w11020390

.

Direct versus indirect effects of human activities on dissolved organic matter in highly impacted lakes

Du

Y.

,

Lu

Y.

,

Roebuck

J. A.

,

Liu

D.

,

Chen

F.

,

Zeng

Q.

,

Xiao

K.

,

He

H.

,

Liu

Z.

,

Zhang

Y.

&

Jaffé

R.

2021

.

Science of the Total Environment

752

, 141839.

https://doi.org/10.1016/j.scitotenv.2020.141839

.

Fluorescence characteristics and biodegradability of dissolved organic matter in forest and wetland soils from coastal temperate watersheds in Southeast Alaska

Fellman

J. B.

,

D'amore

D. V.

,

Hood

E.

&

Boone

R. D.

2008

.

Biogeochemistry

88

, 169–184.

https://doi.org/10.1007/s10533-008-9203-x

.

Source, biogeochemical cycling, and fluorescence characteristics of dissolved organic matter in an agro-urban estuary

Fellman

J. B.

,

Petrone

K. C.

&

Grierson

P. F.

2011

.

Limnology and Oceanography

56

(1), 243–256.

https://doi.org/10.4319/lo.2011.56.1.0243

.

Temporal analysis on quantitative attribution of karst soil erosion: A case study of a peak-cluster depression basin in Southwest China

Gao

J.

&

Wang

H.

2019

.

Catena: An Interdisciplinary Journal of Soil Science Hydrology-Geomorphology Focusing on Geoecology and Landscape Evolution

172

,

369

–

377

.

https://doi.org/10.1016/j.catena.2018.08.035

.

Source apportionment of water pollutants in the upstream of Yangtze River using APCS-MLR

Guowei

C.

,

Mingjing

W.

,

Yan

C.

&

Wei

G.

2020

.

Environmental Geochemistry and Health

42

, 3795–3810.

https://doi.org/10.1007/s10653-020-00641-z

.

Fluorescence characteristics of dissolved organic matter in several independent water bodies: Possible sources and land-use effects

He

Q.

,

Gao

L.

,

Wang

Z.

,

Tang

Y.

,

Pan

B.

&

Li

M.

2021

.

Environmental Science and Pollution Research International

28

,

33241

–

33253

.

https://doi.org/10.1007/s11356-021-12972-0

.

Farmland damage and its impact on the overlapped areas of cropland and coal resources in the eastern plains of China

Hu

Z.

,

Yang

G.

,

Xiao

W.

,

Li

J.

,

Yang

Y.

&

Yu

Y.

2014

.

Resources, Conservation & Recycling

86

, 1–8.

https://doi.org/10.1016/j.resconrec.2014.01.002

.

Properties of fluorescent dissolved organic matter in the Gironde Estuary

Huguet

A.

,

Vacher

L.

,

Relexans

S.

,

Saubusse

S.

,

Froidefond

J. M.

&

Parlanti

E.

2009

.

Organic Geochemistry: A Publication of the International Association of Geochemistry and Cosmochemistry

40

,

706

–

719

.

https://doi.org/10.1016/j.orggeochem.2009.03.002

.

Prediction of BOD, COD, and total nitrogen concentrations in a typical urban river using a fluorescence excitation-emission matrix with PARAFAC and UV absorption indices

Hur

J.

&

Cho

J.

2012

.

Sensors

12

(1), 972–986.

https://doi.org/10.3390/s120100972

.

Dissolved organic matter (DOM) concentration and quality in a forested mid-Atlantic watershed, USA

Inamdar

S.

,

Finger

N.

,

Singh

S.

,

Mitchell

M.

,

Levia

D.

,

Bais

H.

,

Scott

D.

&

Mchale

P.

2012

.

Biogeochemistry

108

, 55–76.

https://doi.org/10.1007/s10533-011-9572-4

.

Dynamics of dissolved organic matter (DOM) in a typical inland lake of the three gorges reservoir area: Fluorescent properties and their implications for dissolved mercury species

Jiang

T.

,

Chen

X.

,

Wang

D.

,

Liang

J.

,

Bai

W.

,

Zhang

C.

,

Wang

Q.

&

Wei

S.

2018

.

Journal of Environmental Management

206

, 418–429.

https://doi.org/10.1016/j.jenvman.2017.10.048

.

The spectral resolution of DOM in urban rivers affected by different non-point source intensities using self-organizing maps

Jin

X.

,

Chen

X.

,

Gao

L.

,

Yuan

M.

,

Wu

Y.

,

Lu

H.

,

Cui

J.

&

Wei

F.

2023

.

Water Science and Technology

88

,

266

–

277

.

https://doi.org/10.2166/wst.2023.187

.

Constraining dissolved organic matter sources and temporal variability in a model sub-Arctic lake

Johnston

S. E.

,

Bogard

M. J.

,

Rogers

J. A.

,

Butman

D.

,

Striegl

R. G.

,

Dornblaser

M.

&

Spencer

R. G. M.

2019

.

Biogeochemistry: An International Journal

146

, 271–292.

https://doi.org/10.1007/s10533-019-00619-9

.

Essentials of the self-organizing map

Kohonen

T.

2013

.

Neural Networks

37

,

52

–

65

.

https://doi.org/10.1016/j.neunet.2012.09.018

.

Albatross R package to study PARAFAC components of DOM fluorescence from mixing zones of Arctic shelf seas

Krylov

I. N.

,

Drozdova

A. N.

&

Labutin

T. A.

2020

.

Chemometrics and Intelligent Laboratory Systems

207

, 104176.

https://doi.org/10.1016/j.chemolab.2020.104176

.

Effects of human land use on the terrestrial and aquatic sources of fluvial organic matter in a temperate river basin (The Meuse River, Belgium)

Lambert

T.

,

Bouillon

S.

,

Darchambeau

F.

,

Morana

C.

,

Roland

F. A. E.

,

Descy

J.-P.

&

Borges

A. V.

2017

.

Biogeochemistry

136

, 191–211.

https://doi.org/10.1007/s10533-017-0387-9

.

Tracking changes in the optical properties and molecular composition of dissolved organic matter during drinking water production

Lavonen

E. E.

,

Kothawala

D. N.

,

Tranvik

L. J.

,

Gonsior

M.

,

Schmitt-Kopplin

P.

&

Köhler

S. J.

2015

.

Water Research

85

,

286

–

294

.

https://doi.org/10.1016/j.watres.2015.08.024

.

Hydrogeochemical spatialization and controls of the Serra Geral aquifer system in southern Brazil: A regional approach by self-organizing maps and k-means clustering

Santos

M. R.

,

Roisenberg

A.

,

Iwashita

F.

&

Roisenberg

M.

2020

.

Journal of Hydrology

591

,

125602

.

https://doi.org/10.1016/j.jhydrol.2020.125602

.

Comparison of Two PARAFAC models of dissolved organic matter fluorescence for a Mid-Atlantic forested watershed in the USA

Shatrughan

S.

,

Shreeram

I.

&

Durelle

S.

2013

.

Journal of Ecosystems

2013

, 532424.

https://doi.org/10.1155/2013/532424

.

Ecosystem consequences of changing inputs of terrestrial dissolved organic matter to lakes: Current knowledge and future challenges

Solomon

C. T.

,

Jones

S. E.

,

Weidel

B. C.

, Buffam, I., Fork, M. L., Karlsson, J., Larsen, S., Lennon, J. T., Read, J. S., Sadro, S. & Saros, J. E.

2015

.

Ecosystems

18

, 376–389.

https://doi.org/10.1007/s10021-015-9848-y

.

Chemical characterization of dissolved organic matter (DOM): A prerequisite for understanding UV-induced changes of DOM absorption properties and bioavailability

Sulzberger

B.

&

Durisch-Kaiser

E.

2009

.

Aquatic Sciences

71

, 104–126.

https://doi.org/10.1007/s00027-008-8082-5

.

A different suite: The assemblage of distinct fungal communities in water-damaged units of a poorly-maintained public housing building

Sylvain

I. A.

,

Adams

R. I.

&

Taylor

J. W.

2019

.

PLoS One

14

,

e0213355

.

https://doi.org/10.1371/journal.pone.0213355

.

Human activities cause distinct dissolved organic matter composition across freshwater ecosystems

Williams

C. J.

,

Frost

P. C.

,

Morales-Williams

A. M.

,

Larson

J. H.

,

Richardson

W. B.

,

Chiandet

A. S.

&

Xenopoulos

M. A.

2016

.

Global Change Biology

22

, 250–263.

https://doi.org/10.1111/gcb.13094

.

Effects of agricultural land use on the composition of fluvial dissolved organic matter

Wilson

H. F.

&

Xenopoulos

M. A.

2009

.

Nature Geoscience

2

, 37–41.

https://doi.org/10.1038/NGEO391

.

Dynamic subsidence simulation and topsoil removal strategy in high groundwater table and underground coal mining area: A case study in Shandong Province

Xiao

W.

,

Hu

Z.

,

Chugh

Y. P.

&

Zhao

Y.

2014

.

International Journal of Mining, Reclamation and Environment

28

, 250–263.

https://doi.org/10.1080/17480930.2013.828457

.

Benthic flux of dissolved organic matter from lake sediment at different redox conditions and the possible effects of biogeochemical processes

Yang

L.

,

Choi

J. H.

&

Hur

J.

2014

.

Water Research

61

, 97–107.

https://doi.org/10.1016/j.watres.2014.05.009

.

Variations in the light absorption coefficients of phytoplankton, non-algal particles and dissolved organic matter in reservoirs across China

Yingxin

S.

,

Pierreandre

J.

,

Lin

L.

,

Zhidan

W.

,

Ge

L.

,

Lili

L.

,

Chong

F.

,

Bai

Z.

,

Junbin

H.

&

Kaishan

S.

2021

.

Environmental Research

201

, 111579.

https://doi.org/10.1016/j.envres.2021.111579

.

Municipal wastewater effluent influences dissolved organic matter quality and microbial community composition in an urbanized stream

Yu

M.

,

Liu

S.

,

Li

G.

,

Zhang

H.

,

Xi

B.

,

Tian

Z.

,

Zhang

Y.

&

He

X.

2020

.

Science of the Total Environment

705

, 135952.

https://doi.org/10.1016/j.scitotenv.2019.135952

.

Effect of hydraulic retention time on biohydrogen production and anaerobic microbial community

Zhang

Z.-P.

,

Show

K.-Y.

,

Tay

J.-H.

,

Liang

D. T.

,

Lee

D.-J.

&

Jiang

W.-J.

2006

.

Process Biochemistry

41

,

2118

–

2123

.

https://doi.org/10.1016/j.procbio.2006.05.021

.

Characteristics and sources of chromophoric dissolved organic matter in lakes of the Yungui Plateau, China, differing in trophic state and altitude

Zhang

Y.

,

Zhang

E.

,

Yin

Y.

,

Dijk

M. A. V.

,

Feng

L.

,

Shi

Z.

,

Liu

M.

&

Qina

B.

2010

.

Limnology and Oceanography

55

, 2645–2659.

https://doi.org/10.4319/lo.2010.55.6.2645

.

Fluorescence characterization and microbial degradation of dissolved organic matter leached from salt marsh plants in the Yellow River Delta

Zhang

T.

,

Feng

Z.

,

Luo

C.

,

Sun

Y.

,

Li

J.

,

Xu

J.

&

Wang

X.

2020

.

Journal of Plant Ecology

13

,

525

–

537

.

https://doi.org/10.1093/jpe/rtaa040

.