A Modified Location-Weighted Landscape Index to Evaluate Nutrient Retention in Agricultural Wetlands: A Case Study of the Honghe Hani Rice Terraces World Heritage Site

Abstract

Understanding the influence of landscape patterns on the water quality of agricultural wetlands is critically important for their management and related decision-making. However, the question of how to quantify this objectively remains a challenge in the relevant scientific fields. In this study, the location-weighted landscape index (LWLI), a process-oriented indicator that integrates ecological processes with landscape patterns based on the source and sink theory, was modified into the SLWLI by assigning nutrient-based weights in the Honghe Hani Rice Terraces World Heritage Site (HHRT). The results indicate that the five watersheds are dominated by sink landscapes, representing 64 percent of the total area. Rice terraced fields were a composite “source–sink” landscape, and their areas in the five watersheds ranged from 4.82 to 20.40%. The nutrient retention function of the sink landscapes of total nitrogen (TN) ranged from 0.64 to 0.86, whereas the total phosphorus (TP) ranged from 0.72 to 0.82, showing good retention function in regard to both nutrients. The contribution rates of forest land and rice terraces to TN and TP retention were greater than 47.07 and 17.07%, respectively, which indicates their key regulation of the nutrient retention function, reducing the risk of water eutrophication and leading to optimized conservation. The vertical pattern of the HHRT plays an important role in nutrient retention function. The SLWLI is an effective index that can be used to assess nutrient retention function and to identify sink landscapes for regulating water pollution in agricultural wetlands.

1. Introduction

Nitrogen (N) and phosphorus (P) are essential nutrient elements for organisms and agricultural production. However, excessive nitrogen and phosphorus entering freshwater from non-point sources (NPS) always causes pollution, leading to water eutrophication [1] and seriously threatening water quality and aquatic biodiversity throughout the world [2,3,4,5,6]. In Denmark, 94% of the N load and 52% of the P load are caused by NPS pollution in 270 rivers [7]. According to another study, 30–50% of the Earth’s surface water has been affected by NPS pollution [8]. In China, compared with point pollution, NPS pollution has become the key problem affecting and restricting the overall development of the country [9]. Studies show that at least 50% of China’s water pollution load is contributed by NPS in the Songhua River, the Huaihe River, Taihu Lake, Chaohu Lake and Dianchi Lake [10,11]. Agricultural production, atmospheric subsidence and urban rainstorm water are the primary causes of surface water NPS pollution [12,13]. Among these, excessive fertilizer use on agricultural land is one of the greatest contributors to water NPS pollution [14]. The global total amounts of nitrogen (TN) and phosphorus (TP) released into freshwater from agricultural production are 31 and 2.9 million tons per year, respectively [15]. Therefore, the control of NPS pollution from agriculture has become a key task in water environment conservation, management and restoration.

Landscape patterns, including landscape composition and configuration, are an important factor controlling NPS pollution because of their impact on the nutrient transfer process [16,17]. If the landscape pattern is poorly designed, it may amplify the transfer of nutrients to water bodies, leading to the deterioration of water quality [18,19]. Effectively assessing the interactions between the landscape pattern and its nutrient retention function for spatial optimization, with the aim of controlling NPS pollution of agricultural landscapes, is a challenging area of study in landscape ecology, water pollution, agricultural ecology and heritage conservation, as well as in sustainability science. Since the effects of landscape patterns on the transfer of nutrients vary at different temporal and spatial scales [20,21,22,23], the question of how to quantify the relationships between the landscape patterns and the nutrient retention functions of agricultural landscapes remains a key scientific question [24,25,26].

Many studies have presented mathematical and statistical indices to quantify landscape patterns [27,28,29], such as edge density, the landscape diversity index, the landscape shape index and patch density [30,31,32,33]. These indices are calculated using the popular Fragstats software, based on the geometry of patches and their spatial relationships [34,35,36]. Unfortunately, many of them are correlated with each other, causing redundancy and interpretive difficulties in their application [37]. Most of these approaches only consider the size and shape of the landscape structure and do not examine the ecological processes, leading to their unclear ecological interpretation [38]. Therefore, more attention should be paid to the development of indices that reflect the relationships between landscape patterns and ecological processes [39,40]. Recently, some process-based landscape pattern indices have been put forward, and a widely used one is the location-weighted landscape index (LWLI), which is based on the source–sink theory [39,41]. LWLI describes the spatial heterogeneity of non-point source pollution in relation to the landscape pattern at the watershed scale. This method classifies landscape types into “source” and “sink” classes. A source landscape is a landscape type that contributes positively to the development of an ecological process and thus exhibits a source function. A sink landscape is a landscape type that contributes negatively to the development of the same ecological process and exhibits a sink function. Due to its flexibility, LWLI has been applied to (1) evaluating NPS pollution risk [42], (2) indicating species migration and protection [43,44], (3) determining water conservation function [45], (4) assessing soil erosion [46,47] and (5) identifying the urban heat island effect [37,48]. However, the above studies using LWLI have mainly applied it to source-landscape-dominated regions, but not to sink-dominated ones. The major factors influencing LWLI (e.g., the weight of source and sink landscapes) have great effects on the accuracy of results due to their weighting based on expert experience. Using reasonable and subjective weights for source or sink landscape parameters to modify LWLI will help in accurately delineating the impact of specific environmental landscape patterns on pollution processes [38].

In this study, we examined five watersheds in the world cultural landscape heritage site of the Honghe Hani Rice Terraces (HHRT). The HHRT is an area of typical mountainous agricultural wetlands in southwest China that has a special vertical landscape pattern of up-slope forests, middle-slope villages, down-slope rice terraces and ditches running through it. Previous studies have shown that the HHRT is dominated by sink landscape types, including rice terraces that have a good nutrient retention function in regard to surface water [49,50]. However, the impacts of vertical landscape patterns on nutrient retention in the HHRT are still unclear and the maintenance of this area’s strong nutrient retention function is essential to ensuring the sustainability of the heritage landscape. The aim of this paper was to explore the usability of the LWLI index in the sink-landscape-dominated agricultural wetlands of the HHRT, to find an objective weight-assigning method for assessing NPS pollution processes and nutrient retention functions and to identify the key sink landscapes and their contribution rates to the nutrient retention function for environmental planning in relation to NPS pollution control.

2. Study Area

Five small watersheds are located in the World Heritage Region of the HHRT (23°06′–23°17′ N, 102°19′–102°44′ E) in Southwest China (Figure 1); these belong to tributaries of the Menglong and Malizhai rivers. The area’s highest altitude is 2375 m and the lowest altitude is 1624 m above sea level. The slope gradient is between 0.0 and 59.2° (

Table 1. Basic information on the five small watersheds.

Watersheds	Area/km2	Elevation/m	Slope	Number of Sample Sites	Landscape Components
a	2.95	1740–2028	0.0–45.4	16	Forest-village-terraces-river
b	0.34	1624–1864	0.0–52.1	12	Forest-terraces-river
c	0.39	1667–1880	0.0–56.2	6	Forest-village-terraces-river
d	1.11	1763–2375	0.5–59.2	4	Forest-village-terraces-river
e	0.39	1810–2010	0.9–35.9	7	Forest-terraces-river

). Precipitation is mainly concentrated in the rainy season, with an average annual precipitation of approximately 1353.8 mm. The territory is deeply cut by the Menglong River and Malizhai River, and a “V”-shaped landform has developed. The vegetation is dominated by subtropical evergreen broad-leaved forests. Generally, in cold and wet areas above 1900 m above sea level, large areas of rice terraces are distributed between 800 m and 1900 m above sea level, whereas below 800 m above sea level, there are dry and hot valleys with savanna vegetation. The soil in the study area is dominated by the yellow-red soil subtype and the submerged paddy soil subtype, and the rice terraces are distributed across a large area. The nutrients are highly vulnerable to losses due to the high mountains, deep valleys, steep slopes and abundant precipitation during the rainy season.

The landscape pattern of the HHRT is unique, and its distribution pattern can be described as “up-slope forests, middle-slope villages and down-slope rice terraces, and a water system running through them” (Table 1). Among these, the b-watershed and e-watershed have no villages, and the areas of the b-watershed, c-watershed and e-watershed are similarly small. In addition, the terraces are vertically distributed; the ponds present a beaded distribution along ditches in the forest. The water in the ditches flows through the forest, villages and terraces, and then into the river, constituting a network distribution from high altitude to low altitude. Therefore, a complex wetland system is formed that includes ditches, ponds, terraces and rivers. This system is useful for intercepting nutrients and is also significant for the protection of the regional ecosystem and the water environment of the HHRT.

3. Data and Methods

In this study, we used the location-weighted landscape index (LWLI) to quantify the landscape pattern based on the source–sink theory [51]. Its principle is to apply Lorenz curves to describe the relationship between landscape patterns and ecological processes. According to the function of the landscape in an ecological process, the landscape is divided into two classes: source and sink. Some landscape types in the watershed play a promoting role and are therefore called source landscapes. Other landscape types play a hindering role and are called sink landscapes. The source and sink landscape patterns are related to their distance to the water outlet, the elevation and the slope gradient. Theoretically, if the elevation of a sink landscape is lower, then the slope gradient is steeper, the distance to the outlet is smaller, the contribution of nutrient or pollutant retention is greater and fewer nutrients and pollutants are produced, and vice versa. Overall, the source–sink theory is mainly used to assess landscape patterns and functions and also to seek suitable landscape spatial patterns and enhance their function in ecological processes [52].

3.1. Water Sampling and Experimental Analysis

The water quality data were mainly collected through field monitoring. According to the vertical distribution characteristics of the landscape types and the water flow conditions in the HHRT, we set the sampling sites at the outlets of forests, villages, high-altitude terraces, low-altitude terraces, ditches and rivers, both upstream and downstream (Figure 1). Among these, three water samples were collected as a mixed sample horizontally in terraced rice fields due to the large area. The collected water samples were placed in polyethylene bottles (500 mL), sealed and stored at a low temperature immediately. A total of 90 water samples from five small watersheds in the study area were taken in June and December, 2017. Finally, the data of TN and TP were quickly obtained by the WTW portable pHotoFlex (Munich, Germany) to ensure the accuracy of the data. All sample processing was completed within 24 h after collection.

3.2. Classification of Source and Sink Landscapes

In this study, a digital elevation model (DEM) with a resolution of 30 m × 30 m was used to extract the five watershed boundaries and to obtain the slope gradient using ArcGIS 10.4 software. Land use data were provided by the National Geographic Information Center at a 0.6 m spatial resolution and were classified into eight types according to the Chinese Land Use Classification Standard (GB/T 21010-2007) and the study aims (Figure 1).

The eight land use types were divided into source and sink landscapes according to the source–sink landscape theory. Among them, forest land was classed as a sink landscape because no nitrogen or phosphorus material is transferred downstream from these areas even when they are located upstream of a source area in the watershed. The other sink landscapes included three types: ponds, ditches and grassland. The source landscapes included three types, namely, rural residential land, bare land and dry land. The terraced rice fields were classified as a source landscape in the farming period (due to fertilization) and a sink landscape similar to wetlands in the fallow period (

Table 2. Classification and weight of source- and sink-type landscape areas in the Hani rice terrace region.

Landscape Type	Land Use Type	Weight of TN	Weight of TP	Reference
Source landscape	Rural residential land	1.00	1.00	[10]
	Bare land	0.50	0.40	[48]
	Dry land	0.40	0.30
	Terraced rice field	0.10	0.18	This work, retention rate
Sink landscape	Terraced rice field	0.90	0.82	This work, retention rate
	Ditch	1.00	1.00
	Pond	1.00	0.81
	Grassland	0.30	0.20
	Forest land	0.50	0.40

).

3.3. Weight of Source and Sink Landscapes by Nutrient Retention Rate

Usually, the determination of the rate of the contribution of source- and sink-type landscapes to nutrient losses are based on field measurements of specific ecological process and expert knowledge [10,47]. According to the nutrient transfer process in the Hani rice terrace landscape, we defined the weights as follows (Table 2). For the source-type landscape areas, because the TN and TP values of water from rural residential land were at the maximum level, they were assigned a weight of 1, which was the standard. Compared with rural residential land, bare land and dry land were assigned weights according to their relative contributions to nutrient losses based on expert knowledge.

Hani rice terraces are always full of water throughout the whole year, and rice is only planted in summer when farmers add chemical fertilizer into the soil. That is, terraced rice fields play both source and sink roles in the nutrient transfer process at planting and hollowing time, respectively. Therefore, terraced rice fields were first assigned a sink weight based on their nutrient retention rate, and their source weight was assigned based on their loss rate, which is the adverse of the retention rate.

Accordingly, the weight of the other sink landscape areas, including ditches, ponds, grassland and forest land, were assigned based on their nutrient retention rates. The formula used to calculate the nutrient retention rate for the weight of a sink landscape is expressed in (1), where n is the retention rate of the landscape, C1 is the concentration of TN or TP in the upstream landscape inlet and C2 is the concentration of TN or TP in the downstream landscape outlet. In this way, the TN and TP weight of each sink landscape type was defined (Table 2).

$$n=\frac{C1-C2}{C1}\ast 100$$

(1)

3.4. Normalization of the Three Factors and Construction of Lorenz Curves of Source and Sink Landscapes

The Lorenz curve approach was used to describe the spatial features of source and sink landscapes related to three factors—the relative distance, relative elevation and slope gradient from the landscape unit to the watershed outlet. In a watershed, if the outlet is used as the reference point, the spatial distribution of the source and sink landscapes can be identified by comparing them to calculate the accumulated area percentage of different landscapes related to the relative distance, relative elevation and slope gradient. The Lorenz curves for different source- and sink-type landscape areas were constructed and analyzed. The values of relative distance, relative elevation and slope gradient in the total watershed were normalized to the range of [0, 1] using the spatial analysis tool in ArcGIS. The accumulated area percentages were then calculated according to the area arrangement to construct the Lorenz curves of these landscape factors for each source and sink landscape type. To conveniently analyze the spatial distribution of landscape types, we set the evaluation standard of normalized values in the range of [0, 1], as follows (

Table 3. Evaluation standard of source and sink landscapes according to three factors.

Factor	0	0.5	1
Relative distance	Close	Middle	Far
Relative elevation	Low	Middle	High
Slope gradient	Shallow	Middle	Steep

).

3.5. Modification and Calculation of the Sink Location-Weighted Landscape Index

Sink landscapes play an important role in intercepting nutrients. However, in the a-watershed, there are many types of sink landscapes and it is difficult to determine which type of sink landscape retention function is the best. Therefore, in this paper, we used the retention rate to assign a weight to a sink landscape and modified the LWLI calculation method into Formula (2), which was named the sink location-weighted landscape index (SLWLI). This index was not only used to analyze the landscape pattern; it was also used to evaluate the retention function of a sink landscape:

$$SLWL{I}^{’}={{\displaystyle \sum }}_{j=1}^n{A}_j\ast W_j\ast P_j/\left({{\displaystyle \sum }}_{i=1}^m{A}_i\ast W_i\ast P_i+{{\displaystyle \sum }}_{j=1}^n{A}_j\ast W_j\ast P_j\right)$$

(2)

$$SLWLI=SLWL{I}_{distance}^{’}\ast SLWL{I}_{elevation}^{’}/SLWL{I}_{slope}^{’}$$

(3)

$$R=SLWL{I}_{jn}/\left(SLWL{I}_{j1}+\dots +SLWL{I}_{jn}\right)$$

(4)

In Formula (2), ${SLWLI}^{’}$ is the ${SLWLI}_{distance}$, ${SLWLI}_{elevation}$ and ${SLWLI}_{slope}$, respectively. $m$ and n refer to the number of source and sink landscape types. $A_i$ and $A_j$ refer to the accumulated areas of source and sink landscapes. $W_i$ and $W_j$ refer to the weights of source and sink landscapes and $P_i$ and $P_j$ are the area percentages of source and sink landscapes. The values of ${SLWLI}^{’}$ and $SLWLI$ are both between 0 and 1; when the value is 0.5, the source and sink landscape distribution is balanced in the watershed. When the value is close to 1, the dominant landscape type in the watershed is sink. It is good when the sink landscape can play a retention role; this means that a relatively small amount of nutrients will be lost from the watershed, with a low risk from the TN and TP. In Formula (3), $SLWLI$ indicates the comprehensive SLWLI. Formula (4) can be used to calculate the contribution rate of different sink landscapes to the TN and TP retention function, where $j$ refers to sink landscapes and $n$ refers to the number of sink landscape types.

4. Results and Analysis

4.1. Area Characteristics of Source and Sink Landscapes

The area characteristics of source and sink landscapes were also analyzed (

Table 4. Statistics of source and sink landscape areas in the five watersheds (unit: %).

Source or Sink Landscape	Land-Use Types	Watershed
		a	b	c	d	e
Source	Rural residential land	4.74	0.84	3.61	5.11	-
	Bare land	4.02	1.77	2.87	-	-
	Dry land	3.61	9.21	10.84	15.62	13.74
	SUM	12.37	11.82	17.32	20.73	13.74
Sink	Pond	2.57	-	0.60	0.23	-
	Ditch	0.29	0.10	-	0.18	0.07
	Forest land	67.47	69.73	69.58	69.24	64.59
	Grassland	8.35	-	-	4.80	1.32
	SUM	78.68	69.83	70.18	74.44	65.98
Source–sink integrated landscape	Terraced rice field	8.95	18.35	12.50	4.82	20.28

). For the source landscapes, the area percentages of rural residential land in the a-watershed, c-watershed and d-watershed were large, with values of 4.74%, 3.61% and 5.11%, respectively, showing little difference among them. The value of the b-watershed was the smallest (0.84%), and there was no rural residential land in the e-watershed. The bare land area was small, with only a small amount distributed in the a-watershed, b-watershed and c-watershed. The area percentage of dry land in the d-watershed was the largest (15.62%), and the minimum value was observed in the a-watershed (3.61%). Overall, the area percentage of all source landscapes was less than 21% in each watershed, indicating that there were few source landscape areas in each watershed. For the sink landscapes, the area percentages of ponds, ditches and grassland were relatively low, with values of 0.23–2.57%, 0.07–0.29% and 1.32–8.35%, respectively. The proportion of forest land area was the largest (64.48–69.73%). The area percentage of all sink landscapes was greater than 64% in each watershed. Terraced rice fields were considered a combined source–sink landscape. The area percentage of these fields in the d-watershed was the smallest (4.82%), whereas it was largest in the e-watershed (20.40%). In general, there were more sink than source landscape areas in each watershed.

4.2. Spatial Distribution Characteristics of Source and Sink Landscapes Related to Relative Distance, Relative Elevation and Slope Gradient

The Lorenz curves of relative distance, relative elevation and slope gradient are depicted in Figure 2, and the accumulated area percentages of the different landscape types are shown in

Table 5. Lorenz curve of the accumulated area percentages of source and sink landscapes related to landscape factors in the five watersheds (unit: %).

Land Use Types	Landscape Factors
	Relative Distance (Min–Max) Mean	Relative Elevation (Min–Max) Mean	Slope Gradient (Min–Max) Mean
Rural residential land	(64.40–80.55) 71.41	(67.71–81.97) 77.28	(54.48–93.66) 78.28
Bare land	(6.59–49.34) 30.56	(6.59–60.61) 37.11	(72.16–81.60) 77.31
Dry land	(42.57–77.73) 56.13	(51.00–71.63) 59.04	(49.98–78.27) 66.72
Rice terraced field	(76.61–91.42) 82.94	(73.19–91.36) 79.71	(60.61–81.05) 68.40
Ditch	(45.89–77.43) 65.60	(53.24–87.59) 68.47	(57.64–81.63) 70.39
Pond	(60.19–92.88) 68.22	(46.89–92.31) 69.53	(81.06–85.14) 83.74
Grassland	(30.30–49.67) 38.13	(26.29–49.67) 40.30	(56.09–88.06) 74.14
Forest land	(37.24–43.61) 39.33	(25.92–46.92) 37.40	(54.37–83.27) 70.33

. The relative distance factor is shown in Figure 2(a1–e1). In the five watersheds, the sink landscapes were distributed within a close distance to one another, including terraced rice fields (82.94%), ditches (65.60%) and ponds (68.22%); they were close to the outlet and contributed greatly to the retention of TN and TP in the watershed. The dry land (56.13%) was distributed in the middle distance, whereas the forest land (39.33%), grassland (38.13%) and bare land (30.56%) were distributed in the far distance. Terraced rice fields were always distributed in positions close to the outlet of each watershed. Forest land was distributed in the far distance, and several other landscape type distributions were irregular. Overall, the distribution pattern can be described as “middle distance for source landscapes; nearest and farthest distances for sink landscapes”. The sink landscape distributed near the outlet of the watershed greatly affects the retention of TN and TP, which is conducive to the retention function of the sink landscape and which reduces the risk of pollution and nutrient losses.

For the relative elevation factor, the smaller the sink landscape relative to the elevation of the watershed outlet, the smaller its contribution to the nutrient loss at the outlet. As shown in Figure 2(a2–e2) and Table 5, some sink landscapes were distributed in low elevations in all five of the watersheds, including terraced rice fields (79.71%) and rural residential land (77.28%). The dry land (59.04%) was distributed in the middle elevation, which was similar to the distribution of the relative distance factors, whereas forest land (37.40%), grassland (40.30%) and bare land (37.11%) were distributed at a high elevation. This means that sink landscapes were located at low relative elevations in the watersheds and they played a crucial role in intercepting nutrients. Overall, the distribution patterns of source and sink landscapes can be described as, “high-elevation sink landscapes, medium-elevation source landscapes, and low-elevation sink landscapes”. The distribution pattern of sink landscapes in relatively low-elevation areas can effectively play an ecological retention function to reduce the risk of nutrient and pollution losses at the water outlet.

For the slope gradient factor Figure 2(a3–e3), the steeper the slope gradient of the sink landscape, the shallower the slope gradient of the source landscape, resulting in low nutrient losses, and vice versa. As can been seen in Table 5, the mean accumulated areas of all the landscapes varied from 66.72% (for dry land) to 83.74% (for ponds), which indicates that the landscapes were distributed with a shallow slope gradient. In relative terms, the rural ponds (83.74%), residential land (78.28%) and bare land (77.31) in the five watersheds had a relatively low slope gradient. Dry land (66.72%) and terraced rice fields (68.40%) had a relatively steep slope gradient, which indicates that some sink landscapes were distributed on steep zones and some were distributed on slow zones; the source landscapes were similar. This means that the distribution of sink and source landscapes was irregular.

4.3. Nutrient Retention Function of Sink Landscapes in the Five Watersheds Analyzed Based on the Sink Location-Weighted Landscape Index

For each of the five watersheds, the SLWLI′distance, SLWLI′elevation, SLWLI′slope gradient and SLWLI were calculated according to Formulas (2) and (3) and are presented Figure 3. The SLWLI′distance, SLWLI′elevation, SLWLI′slope gradient and SLWLI of TN and TP in the five watersheds showed a trend of fluctuation from the a-watershed to the e-watershed. Among them, the distributions of the TN of SLWLI′distance and SLWLI′elevation exhibited the same trend because the distribution landscape patterns of these two indices in all five watersheds were the same. The distribution of the TN of the SLWLI′slope gradient was different in the d-watershed, which shows that the slope gradient played a decisive role in this watershed. Among them, the TN values of SLWLI in the b-watershed, e-watershed and c-watershed were 0.86, 0.83 and 0.82, respectively. These are higher values because their areas were the same and there was no rural residential land in the b-watershed or the e-watershed. The TN values of SLWLI in the a-watershed and d-watershed were 0.77 and 0.64, respectively, which is relatively minor, because the source landscape areas were much greater in those watersheds. The wetland system in the a-watershed was more complex than in the d-watershed, resulting in higher TN values of SLWLI in the a-watershed compared to the d-watershed. The trends of SLWLI′distance, SLWLI′elevation and SLWLI′slope gradient for TP were similar to those for TN.

In summary, the SLWLI values of TN and TP were the lowest in the d-watershed and were the largest in the b-watershed. At the same time, the overall SLWLI, SLWLI′distance, SLWLI′elevation and SLWLI′slope gradient values were higher than 0.50 in the five watersheds, which shows that the landscapes of these watersheds were dominated by sink landscapes. Therefore, this result indicates that the contribution of the sink landscapes in these watersheds to the reduction of nutrient or pollutant losses was significantly higher than that of the source landscapes. Less pollution is exported, resulting in a lower risk of water pollution, and the nutrient retention function works well in these watersheds.

4.4. Contribution Rate of Different Sink Landscapes to Nutrient Retention

The contribution rate of terraced rice fields, ditches, ponds, forest land and grassland to the retention of nutrients was calculated according to Formula (4) (Figure 4). The contribution rate trends of TN and TP were similar. For both TN and TP, the contribution rate of forest land was higher than that of the terraced rice fields in the a-watershed and the d-watershed. The differences between the contribution rates of forest land and terraced rice fields were very small among the b-watershed, c-watershed and e-watershed due to these having the same areas. The contribution rates of forest land to the TN and TP retention function were 50.61–79.38% and 47.07–77.76%, respectively. The contribution rates of the terraced rice fields to TN and TP retention were 17.07–49.31% and 18.95–52.36%, respectively. The contribution rates of the remaining sink landscapes were less than 10% in the five watersheds. Overall, the key sink landscapes were forest land and terraced rice fields; the other sink landscapes (ponds, ditches and grassland) were secondary. Therefore, to reduce the risk of water pollution in the HHRT, the key sink landscapes should be maintained and protected.

5. Discussion

5.1. Effect of Vertical Landscape Patterns on Nutrient Transport and Optimization in the Hani Rice Terraces

The current HHRT landscape pattern (from top to bottom elevation) can be described as “up-slope forests, middle-slope villages and down-slope rice terraces, with a water system running through them”. The respective mean concentrations of TN and TP were 0.01 mg/L and 0.07 mg/L in forest zones, which were the lowest zones. Village zones had the highest concentrations of TN and TP (2.62 mg/L and 0.46 mg/L, respectively). A previous study showed that water belongs to grade IV in village zones [50]. The concentrations in terraced rice fields were the next-highest, because these were wetlands with an effective nutrient retention function; the fields intercept a large amount of nutrients, thereby decreasing the concentration of TN and TP in the river. The water quality of HHRT, as studied by Yao Min et al., showed a vertical characteristic of “good-poor-good” with decreasing altitude [53], and the vertical flow terrace wetland was found to have a better interception effect on TN and TP. Relatively speaking, considering the research conducted by Wang Xiaoling et al. in the plains area on the west bank of Taihu Lake, the interception effects of ecological ditches [54] and ecological ponds [55] on TN and TP were obviously reduced, which was mainly due to the different landscape patterns. The interception function of the advective wetland in the district was good. From the perspective of source–sink landscape types, the vertical landscape types from top to bottom in the HHRT were sink, source and sink landscapes, followed by the river (Figure 5). This pattern is consistent with the idealized source–sink landscape distribution principles of “source→sink→river” from top to bottom elevation in a watershed proposed by Chen et al. [41]. At the same time, the results of this study show that the SLWLI of each watershed was greater than 0.5, indicating that the watersheds were dominated by sink landscapes, and the landscape pattern was relatively reasonable. This finding indicates that the vertical landscape pattern in the watersheds is appropriate and plays an important role in nutrient retention.

In this study, we identified key sink landscape types and the following TN and TP spatial distribution: village zone > terraced rice field zone > river > forest zone. Based on these results, some optimization suggestions can be put forward regarding the landscape pattern in the HHRT. First, the areas of source landscapes should be reduced, and additional sink landscapes should be added near the source landscapes. Second, the functioning of the sink landscapes should be improved. We have formulated the following measures according to the above two principles. The key sink landscape of forest land should be protected and deforestation should be prohibited. Bare land should be planted with trees and grass. In villages, green space should be added around the village. Permeable bricks should be laid on the roads, and drainage should be established on both sides of the road pavement to intercept pollutants. At the same time, multi-pond systems should be established at the water outlets of villages and sewage should be separated; this is more convenient to deal with when the pollutants are concentrated. These measures will reduce the amount of pollutants reaching the rivers. In terraced rice fields, we should maintain the water supply, prohibit terrace drought and maintain the current key sink landscape of the terraced rice fields. It is necessary to regularly inspect the ridges of terraces and repair them in a timely manner when they collapse. The water flow patterns of terrace fields need to be strictly managed. Additionally, hedgerows can be planted around the dry land. In the river area, vegetation buffer zones should be set along the river to intercept pollutants. Ditches and ponds should be dredged regularly to remove soil pollutants and to maintain the migration, flow and purification of waterborne pollutants from the source landscape areas to the sink landscape areas, so that the downstream sink landscapes can adequately intercept and absorb the pollutants.

5.2. Comparison of the Modified and Original Location-Weighted Landscape Index

Chen et al. proposed the LWLI (LCI) to analyze the relationship between landscape patterns and ecological processes [41,51]. This analysis integrates the area and spatial position of the different landscape types to combine landscape patterns and ecological processes. LWLI thoroughly expresses the influence of landscape patterns on ecological processes and is used to study the relationships among landscape patterns, sediment yield and organic carbon; the results have shown that there is a good correlation among them [42,56]. Jiang et al. identified the distribution of pollution sources in a study area (854 km²) using high-resolution remote sensing. A 100 m × 100 m grid was defined as a unit to study the landscape’s contribution to non-point-source pollution in water [57]. The authors modified the LCI into the GLCI to identify the source of non-point-source pollution within a certain spatial scale. This was a similar approach to the one employed in our study, but the aim was not same. In our study, to conveniently identify the contribution rate of the nutrient retention function of different sink landscape types in the watershed, SLWLI was constructed based on LWLI. This approach considered the factors of landscape type and the distribution of the spatial configuration (the relative distance, relative elevation and slope gradient) that affected the transport process of nutrients in the water. Meanwhile, assigning weights to SLWL was also a challenge due to the complexity of ecological processes. SLWLI uses the nutrient retention rate of sink landscape types to assign a weight to a sink landscape, which enabled the contribution rate of the nutrient retention function of the different sink landscape types to be calculated. SLWLI represents a new method for evaluating nutrient retention function (

Table 6. Differences between LWLI and SLWLI.

	LWLI Value Range: [0, 1]			SLWLI Value Range: [0, 1]
Range	>0.5	<0.5	=0.5	>0.5	<0.5	=0.5
The dominant landscape	Source landscape	Sink landscape	Source–sink is balanced	Sink landscape	Source landscape	Source–sink is balanced
Environment effect	Pollution	Interception effect	Balanced	Interception effect	Pollution	Balanced
Applicability	Evaluate the landscape pattern and the contribution of source landscape load			Evaluate the landscape pattern and the nutrient interception function of sink landscapes

). In this way, the key sink landscape types can be identified. The nutrient retention calculation is not only a single calculation of the nutrient retention rate of a landscape type; it also integrates the areas of the various landscape types and their spatial configurations in order to more accurately calculate the retention function of a watershed sink landscape. The nutrient retention calculation comprehensively considers the factors that influence the nutrient transport process. It is beneficial to optimize and regulate the landscape patterns and the nutrient retention function at the watershed scale.

5.3. Innovation, Limitations and Future Directions

The modification of LWLI presented here represents a new attempt at assessing the impact of landscape types on nutrient retention. Due to the complexity of the ecological process, the weight assignment process in the SLWLI is particularly important. Some studies have assigned the weights according to field measurements and expert knowledge [10,41,47,58]. Some studies have used vegetative coverage, soil erosion management factors [46,56] and N and P correction coefficients [57] to assign the weights. The weight setting methods used in these studies have primarily employed correction coefficients or have applied vegetative coverage or expert knowledge and management factors to the universal soil loss equation (

Table 7. Various weight assignment methods used in different studies.

Landscape Type	Land Use Type	Chen	Zhou	Wang	Li	Sun	Jiang
							N	P
Source	Dry field	0.8	-	0.6	0.3	0.6	0.09	0.06
	Paddy field	0.8	-	0.8		0.8	0.14	0.09
	Rural residential area	1	1	0.4	0.2	1	0.64	0.32
Sink	Woodland	0.8	0.8	-	0.05	0.8	0.38	0.22
	Grassland	0.5	0.6	-	0.15	0.6	0.31	0.18
	Water or river	-	-	0.4	0	-	0.08	0.13

). In our study, to evaluate the nutrient retention of a sink landscape, we assigned weights using the nutrient retention rates. The results showed that the nutrient retention function worked very well in the HHRT, indicating that the weight assignment was reasonable. Therefore, for different research aims, the weight assignment should vary according to the situation and should consider those factors associated with the research aims.

There are some limitations to this approach. For example, the original location-weighted landscape index (LWLI) was used for small-scale watersheds. Wang et al. [58] used this index to assess the influence of source and sink landscape patterns on non-point-source pollution in 16 two-grade tributaries of the Three Gorges Reservoir area [59,60]. Wu et al. used this index to study the correlations between the landscape indicator and sediment yield in 18 sub-basins of the Yangtze River, and the results showed the landscape pattern to be a principal factor that was significantly related to the sediment delivery ratio [60]. Zhou analyzed the landscape pattern index and hydrological processes in 41 sub-basins of the Yanhe watershed by modifying LWLI to SHLI and found that SHLI was strongly correlated with the sediment [47]. In conclusion, our method has been primarily applied to small-scale basins. When modifying LWLI to SLWLI in our study, we applied it in small basins as well. For different study scales, the landscape patterns and ecological processes will be different, leading to spatial heterogeneity.

Consequently, to obtain a deeper understanding of the influence of landscape patterns on ecological functions, we suggest that researchers consider additional environmental factors to assign the weights and rectify or improve upon the parameters of SLWLI. Furthermore, we should develop SLWLI to enable the calculation and analysis of nutrient retention function at different scales and fully consider the actual regional environment and the applicability of this method. These deeper investigations will be the focus of future research.

6. Conclusions

Landscape patterns can affect the water quality of wetlands. The identification of the spatial patterns of key sink landscapes in agricultural wetlands to assess their nutrient retention function is critically important for environmental management and restoration. In this study, by modifying the original LWLI, the SLWLI index was proposed to assess the nutrient retention function of sink landscapes in five watersheds of the HHRT.

The source and sink landscapes of the HHRT were classified. In the five watersheds, the total area of the source landscapes, including rural residential land, dry land and bare land, was less than 21%, and the area of sink landscapes, including pond areas, ditch areas, grassland and forest land, was greater than 64%. These results indicate that there were more sink than source landscape areas in each watershed. Terraced rice fields were compound “source–sink” landscapes, and their area percentages in the five watersheds ranged from 4.82 to 20.40%.

Using the Lorenz curve approach, the spatial distribution characteristics of the source and sink landscapes were quantified according to three factors: relative distance, relative elevation and slope gradient. For the relative distance factor, the mean accumulated areas of all the landscape types varied from 30.56% (for bare land) to 82.94% (for terraced rice fields), showing that the source landscapes, including bare land, were mainly located at a far distance. The sink landscapes, including terraced rice fields, ponds and ditches, were located near the outlets, indicating that their relative distances were the shortest, and this resulted in less pollution. For the relative elevation factor, the mean accumulated areas of all the landscape types varied from 37.77% (for bare land) to 79.71% (for terraced rice fields). The source landscapes, including bare land, were mainly observed at relatively high elevations. The sink landscapes, including terraced rice fields, were located at relatively low elevations, resulting in in fewer pollutants being produced. For the slope gradient factor, the mean accumulated areas of all the landscape types varied from 66.72% (for dry land) to 83.74% (for ponds) and showed that all of the landscapes were distributed on a slow slope gradient. However, in relative terms regarding their internal slope, some sink landscapes were distributed on steep zones and some were distributed on gentle zones; the source landscapes were similar. These findings indicate that the distribution of sink and source landscapes was irregular.

Using the SLWLI, the nutrient retention function of sink landscapes was calculated. The SLWLI values of TN ranged from 0.64 to 0.86, whereas the SLWLI values of TP ranged from 0.72 to 0.82. The distribution of the TN of SLWLI′distance and elevation exhibited the same trend, but the distribution of the TN of SLWLI′slope gradient was different, which shows that slope gradient plays a decisive role in some watersheds. The TP results were similar. The values of SLWLI′distance, elevation and slope gradient were greater than 0.50 in the five watersheds, which indicates that the sink landscape dominated the landscape pattern, and the retention function for pollutants and nutrients worked quite well.

The contribution rates of sink landscapes to the retention function shows that forest land TN retention ranged from 50.61 to 79.38% and forest land TP retention ranged from 47.07 to 77.76%. The contribution rates of terraced rice fields ranged from 17.07 to 49.31% in regard to TN retention and from 18.95 to 52.36% in regard to TP retention. The contribution rates of the other sink landscapes were less than 10%. The forest land and terraced rice fields were found to be the key sink landscapes in the five watersheds. The other sink landscapes (ponds, ditches and grasslands) constituted secondary sink landscapes. These landscapes interacted and influenced each other and played a large role in nutrient retention to reduce the risk of water pollution.

Our findings contribute to the quantification of the relationship between landscape patterns and ecological process, which has clear ecological significance, as well as for water quality management, for the better maintenance of agriculture environments and for the sustainable conservation of heritage sites and regions with similar nutrient characteristics. Future studies should investigate practical regulation strategies to effectively control water pollution by spatially arranging sink and source landscapes throughout agricultural wetlands.