The Role of Paddy Fields in the Sediment of a Small Agricultural Catchment in the Three Gorges Reservoir Region by the Sediment Fingerprinting Method

Abstract

Identifying sediment sources is a prerequisite for developing sediment management strategies. Erosion sediment derived from a small agriculture catchment is an important component of sediment inflow in the Three Gorges Reservoir Area. Paddy fields are one of the major land-use types in this region and can have both positive and negative effects on sediment. In this study, two different source group classification schemes were used to analyze the effect of paddy fields on the sediment in a typical small agriculture catchment in the Three Gorges Reservoir Region. A total of 32 soil source samples were collected from four kinds of land-use types (13 from dry land, 5 from orchards, 8 from paddy fields, and 6 from forest) in the Shipanqiu catchment. Moreover, the properties consisted of 41 elements and 12 element ratios were analyzed. Composite fingerprinting methodology was applied to discriminate and quantify the sediment source contributions. Additionally, element ratio was used as the fingerprint property in the fingerprinting application. The results showed that the element ratio was verified as an effective fingerprint property. Additionally, the relative sediment contributions of the potential land-use sources were 55.25% of dry land, 32.69% of orchards, and 12.06% of forest. Paddy fields played a role of sink rather than of source in this study. Accordingly, both forest and paddy fields are effective sediment management strategies. Particularly, paddy fields are a preferred choice for soil erosion control in mountainous and hilly areas. Furthermore, the proper management of paddy fields can help promote sediment retention and reduce soil erosion, which have positive effects on both the environment and agricultural productivity.

1. Introduction

Soil plays an essential role in terrestrial ecosystems, providing a variety of ecosystem services [1,2]. Soil erosion leads to continuous loss of fertile topsoil, nutrients, and organic matter in situ, thereby reducing the ability of soil to sustain valuable ecosystem services [3]. The resulting land degradation is a major challenge to sustainable agricultural production globally [4]. Sediment, as a product of soil erosion, is crucial in structuring landscapes, creating ecological habitats, and transporting nutrients [5,6]. However, overmuch sediment loads may bring about an apparent change of channel, deterioration of water quality, and degradation of aquatic ecosystems [7]. Therefore, the key prerequisite for implementing effective measures to limit excessive sediment transport is to obtain the main sources of sediment and related nutrients and pollution to rivers and reservoir areas. At present, the sediment source fingerprinting process is one of the most commonly used direct methods to investigate sediment provenance [8,9,10,11].

The sediment source fingerprinting process is a powerful tool used to identify and quantify the sources of sediment in a particular area. This process is increasingly applied to evaluate sources of suspended sediment [12,13], deposited sediment of rivers, lakes, and dams [14,15], erosion sediment [16], and aeolian sediment [17], etc. Providing basic information on soil and sediment dynamics, which contributes to learning about landscape evolution and optimizing watershed management and river restoration, is responsible for this increasing trend [4,18]. Physical and biogeochemical properties of sediment were regarded as the reflection of its soil sources in the fingerprinting approach. Accordingly, it is feasible to use these differences in properties to identify the sources of the target sediment [8,19,20,21].

A wide variety of fingerprinting properties have been verified and applied in a growing number of fingerprint studies [8,22]. These common fingerprints include color parameters [23], particle shapes [24], fallout radionuclides [14], mineral magnetics [25], spectroscopy properties [26,27], geochemical properties [28], stable isotopes [16], plant pollens [29], and biological enzymes [30]. The fingerprint properties are supposed to be conservative, that is to say, the properties of fingerprints do not change with sediment transport and environmental changes or have predictable and measurable changes within a certain range [31]. The spatial variability of properties varied with regions. However, some specific element ratios were used for a paleosediment provenance analysis due to their relative stability over large spatiotemporal scales [32]. For example, the sediments of Lake Xingkai came from broadly similar desert sources with the loess–paleosol in north China due to similar ratios of Ti/Al and K/Al [33]. However, few studies have applied the element ratio as the fingerprint factor to composite fingerprinting, especially on a small catchment scale.

The classification of source groups is a key step in the application of fingerprinting techniques, as they are used to construct field samples of potential sources [34]. Summarizing existing studies, it is common practice to selected priority source groups based on spatial sources subdivided by geological units [35,36,37] or tributary subbasins [38,39,40,41] or source types consisting of surface and subsurface sources [18,42,43,44]. In practical applications, the sources of different classification methods can be combined. On a small catchment scale, the one classified by land-use pattern is the most widely used. Preselected source groups may be perfectly acceptable, but trying to explore multiple source group partitions may improve robustness [34].

The Three Gorges Reservoir Region has an important status and impact in the social economy, ecological barrier, and water-quality security of the Yangtze River Basin of China. There is a considerable number of small agricultural catchments mainly used for agricultural production in the reservoir area. Additionally, these catchments are often dominated by agricultural land use [20]. The eroded sediment produced by small agricultural catchments in the reservoir area is an important component of sediment inflow. Therefore, it is of great significance to find out the source contribution of the sediment in a typical agricultural catchment for effective measures to control the sediment quantity. Moreover, paddy fields, dry land, orchards, and forest are the four major kinds of land-use types of the small agricultural catchment in this region. On the one hand, paddy fields can slow down the water flow and allow sediment to settle [20]. On the other hand, paddy fields may also contribute sediment through erosion and runoff [45]. Then, the extent to which paddy fields act as sinks or sources of sediment may vary depending on factors, such as the management practices used in the field and the amount and intensity of rainfall in the area. Therefore, it is important to carefully evaluate the role of paddy fields when using sediment fingerprinting to identify sediment sources in catchments.

Consequently, the objectives of this study were (1) to assess the effect of paddy fields on the sediment contributions within the catchment by two source groups (source group 1: paddy fields, dry land, orchards, and forest and source group 2: dry land, orchards, and forest); (2) to verify the validity of the element ratio as the fingerprint factor in the application of composite fingerprinting; (3) to quantify the relative contribution of the potential sources.

2. Materials and Methods

2.1. Study Area

The study area is Shipanqiu catchment, located in Zhong County, Chongqing municipality (107°3′~108°14′ E, 30°03′~30°35′ N) in the middle reaches of the Three Gorges Reservoir area of China (Figure 1). The climate of the study area is classified as subtropical southeast monsoon with distinct seasons. The annual mean temperature of this area is 19.2 °C with an annual frost-free period of 320 days, while the annual average precipitation is 1150 mm. However, precipitation is unevenly distributed throughout the seasons, and 70% falls from April to September. The total area of the study catchment is about 0.35 km², and the main types of land use in the studied area include dry land (Figure 2a), paddy fields (Figure 2b), orchards (Figure 2c), forest (Figure 2a), residential areas (villages and market towns), etc. In the study catchment, the largest area is dry land, which accounts for 33.42% of the entire catchment. Paddy fields, forest, and orchards account for 20.79%, 20.05%, and 19.68%, respectively. In addition, the proportions of building areas and ponds were 4.31% and 1.75%, respectively. Commonly planted crops are rice, corn, potatoes, citrus, vegetables, and so on.

2.2. Field Sampling

The collection of source soil samples was carried out separately and simultaneously according to different land-use types. The depth of 0–2 cm is the main layer of soil erosion, and 2 cm depth is frequently used in collecting sediment source samples of fingerprinting studies [46]. Hence, a sample depth of 2 cm was selected for this study. Due to the rugged terrain and fragmented landscape morphology, it was difficult to implement a standard grid sampling tactic. As a result, sampling points were unevenly distributed across the study catchment. The amounts of samples mainly depended on the area of different land-use types. Five samples were collected from orchards, six samples from forest, thirteen samples from dry land, and eight samples from paddy fields. A multipoint sampling method was applied to collect about 500 g of massive topsoil from each sampling point with a stainless steel shovel. Each sample was a mixture of 10 duplicate sites around each sampling point.

Simple time-integrating sediment trap was deployed to collect samples of sediment. Such a sampler has been successfully used in previous studies to collect composite sediment samples continuously [47,48]. A time-integrating trap was installed at the outlet of the catchment on 14 May 2020. Subsequently, the sediment deposited in the trap was collected after each rainfall, and then the trap was cleared and refixed in place immediately. Five rainfall events were captured across the whole rainy season in 2020.

2.3. Laboratory Analysis

Particle size effects caused by selective delivery process can lead to deviations of properties between sources and sediments [49,50]. Such deviations may result in inaccuracies of the direct comparisons between source samples and target sediment samples. So far, particle size fractionation is considered to be the most common approach to address this problem. The fraction of <63 μm has been widely used in previous studies around the world [15,18,51,52]. Consequently, the <63 μm fraction was selected for the analysis and comparison of properties in the study. Additionally, dry sieving was used for fractionation.

All samples were initially air-dried at room temperature and sieved to 63 μm to remove any dead leaves, plant roots, and coarse gravel, which could distort further measurements. A total of 41 geochemical fingerprints were measured, and 12 element ratios were calculated as fingerprints for further analysis (

Table 1. Summary of properties analyzed.

Classification	Properties
Elements	Na, Mg, K, Al, Ca, Fe, Ti, Cr, V, Mn, Ni, Co, Be, Li, Rb, Y, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, Sr, Cd, Ga, In, Tl, Se, Cu, Pb, Zn, P, S, TN, SOC
Element ratios	K/Al, Al/Ti, Fe/Cu, Cr/V, V/Ni, V/Co, Co/Cu, Li/Be, Pb/In, Zn/Cu, Zn/Cd, Zn/Pb

). In the process of sediment production and transport, the relative content of elements with similar properties, strong correlations, and similar enrichment degrees can basically remain unchanged [53]. Based on this premise and previous studies [53], several element ratios presented in Table 1 were selected.

After acid digestion with HNO₃ and HF, the concentrations of Ni, Pb, Cu, Cd, Sr, Y, Co, Be, Li, Tl, V, Cr, Zn, Se, In, Ga, Rb, Pr, Nd, Sm, Gd, Tb, Dy, Ho, Er, Tm, Yb, and Lu were determined by ICP-MS (Perkin Elmer, Waltham, MA, USA), and the contents of Al, Fe, Na, S, Mn, Ti, K, P, Ca, and Mg were examined using ICP-OES (Perkin Elmer, Waltham, MA, USA). SOC and TN contents were analyzed using Vario Macro Cube (Elementar, Langenselbold, Germany).

2.4. Screening of Optimal Composite Fingerprints

The classification of potential sediment sources and the differences of properties between different source samples are expected to influence the selection of fingerprints and the deduction of optimum composite fingerprints [54]. Forecasting errors of potential sources could create problems in source identification. Significant differences were supposed in different sources. If the properties of soil samples from different sources differ insignificantly, these properties can be considered unable to effectively identify the source type. In other words, the fewer the significant fingerprint properties represented, the fewer the differences determined between the compared sources. Therefore, it is necessary to reconsider whether the default source could act as an individual potential source.

Prior to the fingerprinting procedure, a range test was applied to identify the conservative behavior of tracers, whereby the fingerprint concentrations of the sediment samples were compared with the corresponding ranges associated with the potential source samples. This test has been used worldwide as part of the sediment source fingerprinting technology [18,48,52,55,56]. Fingerprint properties whose concentrations of sediment fell outside the corresponding value ranges of sources were removed from further analysis. The range test is incapable of precisely recognizing all properties that behaved nonconservatively, but it can exclude the properties with the greatest differences, which can also estimate the existence of unavailable samples. Accordingly, the range test is a screening tool used to screen out properties that experienced significant change during delivery.

Several statistical analysis steps were employed to ascertain which fingerprint properties are most important in source discrimination. A two-step statistical procedure including Kruskal–Wallis H test (KW-H test) and stepwise discriminant function analysis (DFA), was selected in this study, which has been validated in many fingerprinting studies [47]. The former test was applied as a primary filter of properties, whereas the latter affirmed the optimal composite fingerprints during stepwise selection. The results of DFA were used to check the percentage of samples properly classified into the correct source group.

2.5. Sediment Source Apportionment

A numerical mass balance mixing model was adopted to quantify the relative contribution of each source type. The expression of the model is as follows:

$$C_{si}={{\displaystyle \sum }}_{i=1}^k{P}_j{C}_{ji}$$

(1)

where $C_{si}$ is the concentration of fingerprint property ($i$) in sediment samples; $P_j$ is the percentage of the contribution of source ($j$) to the sediment samples; $C_{ji}$ is the concentration of fingerprint property ($i$) of soil samples of source ($j$); $k$ is the number of fingerprint properties.

In practical applications, researchers optimized the estimation of relative source contributions by minimizing the sum of squares of residuals. The improved mixing model function is as follows [10]:

$$R_{es}={{\displaystyle \sum }}_{i=1}^k{\left(\frac{C_{si}-{{\displaystyle \sum }}_{j=1}^m{P}_j{C}_{ji}}{C_{si}}\right)}^2$$

(2)

where $R_{es}$ is the sum of the squares of the residuals; $m$ is the number of potential sources.

Solutions of the mixing model were subject to the overall boundary constraint [57]:

$${{\displaystyle \sum }}_{j=1}^m{P}_j=1\cdots \cdots 0\le P_j\le 1$$

(3)

Under the above conditions, the relative contribution percentage of each sediment source can be obtained when $R_{es}$ is the minimum.

Additionally, the goodness-of-fit (GOF) method was used to evaluate the fitting degree of the mixing model to the observed values of samples. The expression of GOF is as follows [58]:

$$GOF=1-\left(\frac{1}{k}{{\displaystyle \sum }}_{i=1}^k\frac{\left|C_{si}-{{\displaystyle \sum }}_{j=1}^m{P}_j{C}_{ji}\right|}{C_{si}}\right)$$

(4)

In general, when the value of GOF > 0.8, the calculation result of the model was considered acceptable.

All the mentioned analyses and calculations were performed with Fingerpro package of R language software.

3. Results

3.1. Paddy Fields as Potential Source

3.1.1. Optimal Composite Fingerprints

The mean concentrations of properties in potential sources and sediments are presented in

Table 2. Mean concentrations of elements of sources and sediments.

Properties	Forest	Dry Land	Paddy Fields	Orchards	Sediment
Na (g kg−1)	14.68	12.49	13.24	12.93	18.42
Mg (g kg−1)	7.68	10.32	10.07	9.82	6.88
K (g kg−1)	20.16	19.45	20.45	23.19	17.54
Al (g kg−1)	82.39	82.88	82.94	80.73	67.05
Ca (g kg−1)	6.35	8.33	8.55	6.77	20.19
Fe (g kg−1)	29.92	33.87	31.51	32.52	25.04
Ti (mg kg−1)	3.94	4.24	4.06	4.19	4.48
Cr (mg kg−1)	57.36	68.81	73.49	65.06	52.27
V (mg kg−1)	71.81	77.17	71.59	72.51	65.45
Mn (mg kg−1)	469.60	550.66	492.70	489.74	661.25
Ni (mg kg−1)	27.06	35.20	39.10	32.07	22.63
Co (mg kg−1)	12.59	13.08	12.92	11.78	10.40
Be (mg kg−1)	2.10	2.36	2.35	2.35	1.85
Li (mg kg−1)	28.32	35.11	33.44	31.22	25.81
Rb (mg kg−1)	88.27	90.63	91.16	87.04	73.87
Y (mg kg−1)	11.22	17.43	14.75	8.95	17.12
Pr (mg kg−1)	6.13	7.18	6.10	3.53	7.88
Nd (mg kg−1)	23.06	27.18	23.04	13.35	28.30
Sm (mg kg−1)	4.27	5.11	4.43	2.82	4.90
Eu (mg kg−1)	0.98	1.17	1.04	0.66	1.15
Gd (mg kg−1)	4.06	4.86	4.07	2.45	3.88
Tb (mg kg−1)	0.55	0.68	0.59	0.37	0.69
Dy (mg kg−1)	3.00	4.24	3.27	2.14	3.39
Ho (mg kg−1)	0.59	0.73	0.63	0.42	0.77
Er (mg kg−1)	1.71	2.11	1.87	1.23	2.01
Tm (mg kg−1)	0.25	0.35	0.29	0.20	0.32
Yb (mg kg−1)	1.82	2.51	1.98	1.36	1.99
Lu (mg kg−1)	0.23	0.37	0.29	0.20	0.32
Sr (mg kg−1)	219.08	204.85	222.48	197.38	235.13
Cd (mg kg−1)	0.33	0.32	0.29	0.25	0.24
Ga (mg kg−1)	19.03	21.06	21.45	20.61	19.19
In (mg kg−1)	0.04	0.05	0.05	0.05	0.04
Tl (mg kg−1)	0.53	0.58	0.55	0.60	0.51
Se (mg kg−1)	0.18	0.18	0.09	0.12	0.20
Cu (mg kg−1)	18.26	25.82	23.70	23.71	21.45
Pb (mg kg−1)	25.50	25.14	24.71	24.63	22.52
Zn (mg kg−1)	71.99	89.40	79.46	92.97	73.87
P (mg kg−1)	434.08	610.83	582.83	1434.46	698.47
S (mg kg−1)	122.76	111.27	304.94	127.27	-
TN (g kg−1)	1.72	1.01	1.09	1.40	0.09
SOC (g kg−1)	17.39	8.37	9.36	12.55	0.83

and

Table 3. Mean values of element ratios of sources and sediments.

Properties	Forest	Dry Land	Paddy Fields	Orchards	Sediment
K/Al	0.24	0.24	0.25	0.29	0.26
Al/Ti	20.90	19.54	20.44	19.27	14.99
Fe/Cu	1.64	1.31	1.33	1.37	1.15
Cr/V	0.80	0.89	1.02	0.90	0.81
V/Ni	2.72	2.19	1.87	2.26	3.02
V/Co	5.71	5.91	5.55	6.16	6.36
Co/Cu	0.69	0.51	0.55	0.50	0.48
Li/Be	13.50	14.89	14.25	13.26	13.85
Pb/In	637.55	560.64	544.50	517.55	580.67
Zn/Cu	3.95	3.45	3.35	3.92	3.39
Zn/Cd	216.85	287.71	284.40	377.06	315.64
Zn/Pb	2.82	3.56	3.24	3.78	3.25

. The concentrations of S in the sediment samples were not detected; therefore, the element of S was removed from further analysis. The elements Mg, Ti, Cr, V, Ni, Be, Li, Rb, Y, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, Sr, Cd, Ga, In, Tl, Se, Cu, Pb, Zn, P and the ratios of K/Al, Fe/Cu, Cr/V, V/Ni, Co/Cu, Li/Be, Pb/In, Zn/Cu, Zn/Cd, and Zn/Pb passed the range test, because their concentrations in the sediment fell between the minimum and maximum concentrations of the potential sources. These elements and ratios that passed the range test were considered conservative properties. The conservative properties mentioned above were tested by the KW-H test. The results of the KW-H test (

Table 4. Results of KW-H test for performance of each property of source group 1.

Properties	p-Value	Properties	p-Value	Properties	p-Value	Properties	p-Value	Properties	p-Value
Mg	0.033 *	Pr	0.119	Tm	0.059	Cu	0.007 **	Li/Be	0.001 **
Ti	0.007 **	Nd	0.119	Yb	0.032 *	Pb	0.096	Pb/In	0.035 *
Cr	0.012 *	Sm	0.175	Lu	0.009 **	Zn	0.008 **	Zn/Cu	0.002 **
V	0.076	Eu	0.132	Sr	0.226	P	0.001 **	Zn/Cd	0.003 **
Ni	0.008 **	Gd	0.166	Cd	0.066	K/Al	0.010 *	Zn/Pb	0.012 *
Be	0.222	Tb	0.197	Ga	0.052	Fe/Cu	0.010 *
Li	0.017 *	Dy	0.038 *	In	0.122	Cr/V	0.001 **
Rb	0.831	Ho	0.231	Tl	0.231	V/Ni	0.001 **
Y	0.116	Er	0.232	Se	0.348	Co/Cu	0.004 **

*, ** Statistically significant values at p ≤ 0.05, 0.01.

) indicate that a total of 21 properties (Mg, Ti, Cr, Ni, Li, Dy, Yb, Lu, Cu, Zn, P, K/Al, Fe/Cu, Cr/V, V/Ni, Co/Cu, Li/Be, Pb/In, Zn/Cu, Zn/Cd, and Zn/Pb) were significantly different among the four potential sources at p < 0.05. The remaining properties failed to discriminate the four sources; they were discarded from further analysis.

Subsequently, a stepwise DFA was performed on the properties determined by the KW-H test. The first three typical discriminant functions were used in the analysis (

Table 5. Characteristics of typical discriminant functions of stepwise discriminant function analysis (DFA) of source group 1.

Function	Variance (%)	Cumulative Variance (%)	Canonical Correlation	Sig.
1	72.0	72.0	0.992	0.000
2	18.4	90.4	0.970	0.000
3	9.4	99.8	0.946	0.000

). The three functions explained 72%, 18.4%, and 9.4% of the total variance of sediment sources with corresponding canonical correlations of 0.992, 0.970, and 0.946, respectively. The high value of canonical correlation indicated that the discriminant scores were strongly correlated with the source groups. Furthermore, the Lambda test significance of the three functions were p = 0.000 < 0.05, indicating that the discriminant effects obtained by the three functions were effective. Additionally, the optimal combination of properties for discriminating four source materials in the studied catchment are listed in

Table 6. Summary of the stepwise DFA results of two selected source groups.

Source Group	Optimal Composite Fingerprints	Classified Efficiency
1	Cu, P, Cr/V, V/Ni, Co/Cu, Li/Be, Zn/Cu, Zn/Pb	100%
2	P, Fe/Cu, Li/Be, Pb/In, Zn/Cd	100%

. A total of eight properties (Cu, P, Cr/V, V/Ni, Co/Cu, Li/Be, Zn/Cu, and Zn/Pb) were selected with a predictive power of 99.8% (Table 5). Moreover, the results of the stepwise DFA show that the samples in the initial source groups were classified 100% correctly.

3.1.2. Apportionment of Sediment Sources

The average relative sediment contribution ratios from the four sources are shown in Figure 3a. The results indicate that dry land was the dominating source of sediment, accounting for 52.79%, followed by forest and orchards, constituting 30.62% and 16.54%, respectively. Paddy fields contributed the least. It is worth mentioning that the contribution proportion of paddy fields to sediment was only 0.05%, which was almost negligible. Given this, there were almost no erosion and sediment production in the paddy fields during the period of sediment collection in this study. The paddy fields probably acted mainly as sinks instead of sources. Thus, paddy fields should be excluded from potential sources in the current study. Because the classification of sources will directly affect the fingerprinting results and increase the uncertainty of fingerprinting, the results containing inappropriate source were not analyzed in further detail.

3.2. Paddy Fields as Sinks

3.2.1. Optimal Composite Fingerprints

A total of 39 properties (Mg, Ti, Cr, Ni, Be, Li, Rb, Y, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, Sr, Cd, Ga, In, Tl, Se, Cu, Zn, P, K/Al, Fe/Cu, Cr/V, V/Ni, Co/Cu, Li/Be, Pb/In, Zn/Cu, Zn/Cd, and Zn/Pb) passed the range test. Of these properties, 22 properties (Mg, Ti, Cr, Ni, Li, Dy, Yb, Lu, Cd, Cu, Zn, P, K/Al, Fe/Cu, Cr/V, V/Ni, Co/Cu, Li/Be, Pb/In, Zn/Cu, Zn/Cd, and Zn/Pb) were significantly different among three potential sources at p < 0.05 by the KW-H test (

Table 7. Results of KW-H test for performance of each property of source group 2.

Property	p-Value	Property	p-Value	Property	p-Value	Property	p-Value	Property	p-Value
Mg	0.025 *	Pr	0.101	Er	0.208	Tl	0.217	Cr/V	0.009 **
Ti	0.012 *	Nd	0.101	Tm	0.071	Se	0.538	V/Ni	0.004 **
Cr	0.006 **	Sm	0.151	Yb	0.035 *	Cu	0.006 **	Co/Cu	0.009 **
Ni	0.004 **	Eu	0.125	Lu	0.012 *	Pb	0.054	Li/Be	0.006 **
Be	0.180	Gd	0.134	Sr	0.264	Zn	0.017 *	Pb/In	0.007 **
Li	0.014 *	Tb	0.207	Cd	0.041 *	P	0.002 **	Zn/Cu	0.009 **
Rb	0.733	Dy	0.044 *	Ga	0.134	K/Al	0.008 **	Zn/Cd	0.003 **
Y	0.160	Ho	0.207	In	0.070	Fe/Cu	0.008 **	Zn/Pb	0.012 *

*, ** Statistically significant values at p ≤ 0.05, 0.01.

).

The stepwise DFA was conducted on the above 22 properties passed the KW-H test. Two typical discriminant functions with a predictive power of 100% (79.90% of Function 1 and 20.10% of Function 2) were determined (

Table 8. Characteristics of typical discriminant function of stepwise DFA of source group 2.

Function	Variance (%)	Cumulative Variance (%)	Canonical Correlation	Sig.
1	79.9	79.9	0.992	0.000
2	20.1	100.0	0.968	0.000

). In addition, high values of canonical correlation (0.992 for Function 1 and 0.968 for Function 2) were presented, indicating discriminant scores were highly correlated with the source groups. Moreover, the discriminant effects of both functions were valid because the p values of the Lambda test significance by the two functions were 0.000 (<0.05). Additionally, the optimal fingerprint groups for discriminating three source materials in the Shipanqiu catchment are presented in Table 6. A total of five properties (P, Fe/Cu, Li/Be, Pb/In, and Cu/Cd) were selected. The results show that these properties have excellent discriminative ability because the original source samples were 100% correctly categorized.

3.2.2. Apportionment of Sediment Sources

The results of the sediment contributions based on three sources (orchards, forest, and dry land) are shown in Figure 3b. Of the three sources, the relative contribution rate of dry land was the highest (55.25%), followed by orchards (32.69%), and forest was the smallest (12.06%). Furthermore, the GOF value of this result was 0.85, indicating that the relative contributions of each individual source from the mixing model were meaningful.

4. Discussion

4.1. Selection of Potential Source

Sediment source identification is the first step of the fingerprinting methodology. In previous studies, the classification of sediment sources was mostly determined by field investigation and a priori knowledge, which possessed definite subjectivity [4,14]. Additionally, the source classification is diversifying due to the diversity of the study area. There may be omissions or misjudgments in the selection of potential source groups, which may reduce effective source identification [14,34,50]. For example, in the investigation of major sediment sources in a small catchment with high agricultural and mining activities in Nanjing, the authors improved the efficiency of source identification by adjusting the source groups [54]. In this study, it was uncertain whether paddy fields were potential sediment sources in the study area. Additionally, controversies existed about the role of paddy fields. For example, the sediments derived from paddy fields were ignored due to the intercepting effect of paddy fields [20]. However, based on intensive tillage during the flood season, especially harvesting of the first-season rice and sowing of the second-season rice, a large proportion (30.7%) of paddy fields to sediment was also found in the Shouchang River catchment, in southeastern China [45]. Therefore, the paddy fields were considered both as a source or a sink. In general, the process of sediment removal from in situ to migrate to the catchment outlet is not always continuous. Sediments are probably deposited anywhere in the catchment due to topography, vegetation, and even anthropogenic factors and then can also be transported to other locations, including catchment outlets, during subsequent runoff events. Accordingly, paddy fields promote sediment deposition to some extent; now the paddy fields are acting as sinks. However, the sediments deposited in paddy fields may also be transported again, and at this point, the role of paddy fields changes from sinks to sources. The results of the DFA show that both source groups discussed in this study were acceptable, as the provenance samples were classified exactly. According to the fingerprinting results, the sediment contribution of paddy fields was minimal (0.05%) (Figure 3). Based on this result, it was more appropriate to treat paddy fields as sinks rather than sediment sources. A contrary conclusion was reported in [59]. The results show that paddy fields were one of the sources of pond sediment from an agricultural catchment in the Three Gorge Reservoir Region. To sum up, we can infer that paddy fields are likely to be the sources for sediment deposited over a long period of time, while paddy fields may be sources or sinks during short-term, single, or multiple runoff events. Under conditions of intensive tillage practices or prolonged heavy rainfall, the role of paddy fields is likely to be that of the sediment source [45]. During flood events, it is worth noting that paddy fields probably act as sediment sources if they are in a drainage or drying period. Furthermore, the classification of source groups should be deliberated throughout the whole process of fingerprinting. The relative contributions of sediment sources differed from different source groups. Similarly, three kinds of classification methods combined with artificial mixed simulation methods were used in [50], proving that reasonable source classifications can reduce the uncertainty of fingerprinting results and ultimately help inform effective sediment management strategies.

4.2. Discriminant Fingerprint Properties

Fingerprint property concentrations in land-use-based source groups are possible to be governed by a variety of factors, including soil type, colluvium parent material, pedogenesis processes, and anthropogenic impact. [31]. The studied catchment covers a small area of 0.35 km² with similar geological and climatic conditions. Furthermore, the selection of sources has an obvious influence on the ability of fingerprint properties to distinguish sources [14,34]. Geochemical elements widely used in sediment fingerprinting studies are sensitive to the environment, and the optimal fingerprint properties are varied in different study areas. For example, seven elements (Mg, Y, Ti, P, Sc, Co, and Cr) structured the composite fingerprints in a small catchment of the Loess Plateau, China [18]. The elements K-40, Cs-137, Li, Sr, and Ti created a composite fingerprint for a Pyrenean catchment [15]. In fact, the absolute contents of most elements change during the process of erosion, delivery, and deposition. However, the relative contents between some elements could be maintained; that is, the ratio of elements remains relatively stable [53]. Existing studies presented that various element ratios have been widely used in many studies, such as mineral deposit exploration, paleoenvironment analysis, and climate reconstruction [33,60]. Some element ratios have been identified as specific indicators during long time scales or large spatial scales. For example, the K/Al ratio was used to reflect the change of parent material composition in the early stage of chemical weathering [61]. In the current study, the K/Al ratio passed the range test and the KW-H test. Although it was not selected as the optimal fingerprint, it still met the requirements of the fingerprints in fingerprint technology. Not all these ratios meet the demands. For example, the ratio of Ti/Al, which was related to the parent soil or rock type [61], failed the range test. In this study, several element ratios were screened into the optimal fingerprints, demonstrating that element ratios can be used as effective fingerprint properties in fingerprinting. However, element ratios are not always unique to specific sediment sources and can be influenced by factors, such as weathering, transport, and deposition processes. Overall, element ratios were considered to be useful in conjunction with other fingerprints in sediment fingerprinting.

4.3. Sediment Source Apportionment

Dry land was the dominate source of sediment in the study area at <63 μm (55.25%, Figure 3b). Similar results in the Three Gorges Reservoir Region were reported. For example, the relative sediment contribution of slope land (corresponding to dry land) of the Wangjiagou catchment was 54.11% [20]. Furthermore, in a fingerprinting study of pond sediment in the Lingjiaotang catchment, the relative sediment contribution of dry land reached 82.00% [59]. Dry land is considered to be more susceptible to erosion than forest and orchards. On the one hand, the surface of dry land is exposed with few leafy trees and shrubs. On the other hand, annual herbs are commonly planted on dry land, such as corn, peanut, canola, and potato, with sparse, fine root systems and little surface litter. In addition, orchards contributed about one third to the sediment. As a single-species of planted economic forest, management activities are essential factors affecting soil erosion. In 2019, managers carried out a holistic graft to improve the varieties of orchards, decreasing canopy cover dramatically, which inevitably led to an increase in erosion intensity. However, similar to forest, the sediment from the orchards is thought to decrease gradually as the orchard matures [62]. Meanwhile, orchards are mainly distributed in areas with a large slope. Under the same land-use type, an increasing tendency of soil erosion was noticed with the increase in slope [63]. Moreover, forest contributed the least sediment, but still 12.06%. This is probably because forest is distributed in the western and southwestern areas of the catchment with a steep slope and close to the outlet. A study conducted in the Hujiawan catchment of the Loess Plateau in China found a higher sediment contribution of forest (21.5%) [18]. Factors, such as sparse understory vegetation, bare topsoil, long-term soil desiccation, stunted growth, and small vegetation leaves, caused serious soil erosion in the forest. Owing to varied landforms, soil properties, vegetation covers, and other factors, soil antierodibility of the same land-use type in different regions may differ. However, most of the time, forests were tacitly assumed to contribute little to sediment [20]. As for paddy fields, they can have both positive and negative effects on sediment. On the one hand, they trap sediment by slowing down water flow and creating a large sedimentary surface area. However, if the water flow into the paddy field is too strong, this can lead to soil erosion. Then, the water flow can carry sediment with it to delivery. For example, intense rain, improper management of irrigation, and unreasonable tillage practices could result in soil erosion. As a consequence, during the process of sediment transport, paddy fields can also act as sources or sinks of sediment. In addition, mountains and hills are the main landforms in the Three Gorges Reservoir area [20]. Therefore, the sediment-trapping function of paddy fields is of great significance for sediment management in the catchment. Overall, the relative sediment contribution ratio of one type of land-use source varies greatly in different study areas. The differences between source group selection and the actual situation of the study area are one of the crucial reasons. In addition, based on the sediment contribution results in this study area, we can infer that paddy fields may be a better sediment reduction strategy in the Three Gorges Reservoir area, followed by forest land.

5. Conclusions

This study investigated the role of paddy fields in catchment outlet sediment in a representative agricultural catchment of the Three Gorges Reservoir area by fingerprinting methodology. The results indicated that in this small catchment, the element ratio was verified and can be used as a fingerprint property. Furthermore, the relative sediment contribution ratios of potential sources at particle size of <63 μm were 55.25% for dry land, 32.69% for orchards, and 12.06% for forest. Paddy fields acted as sinks rather than sources during the sampling period. Accordingly, both forest and paddy fields can reduce soil erosion and are effective strategies for sediment management in the Three Gorges Reservoir area. Eminently, paddy fields may be a better-designed strategy of soil erosion control in mountainous and hilly areas. Furthermore, reasonable management practices of paddy fields can contribute to promoting sediment retention and reducing soil erosion, which have positive effects on both the environment and agricultural productivity.