Impacts of Landscapes on Water Quality in A Typical Headwater Catchment, Southeastern China

: The relationship between land landscape and water quality has been a hot topic, especially for researchers in headwater catchment, because of drinking water safety and ecological protection. In this study, Lita Watershed, a typical headwater catchment of Southeast China, was selected as the study area. During 2015 and 2016, water samples were collected from 18 sampling points every month, and 19 water quality parameters were tested such as nutrients and heavy metals. Through multistatistics analysis, the results show that the most sensitive water quality parameters are Cr, NO 3 , NO 2 , and COD. The type and scale of water body have direct e ﬀ ects on water quality, while the land-use patterns in the surrounding areas have an indirect impact on the concentration and migration of pollutants. This e ﬀ ect is sensitive to seasonal change because heavy metals are mainly from atmospheric deposition, but nutrients are mainly from agricultural nonpoint source pollution. According to the results, increasing the proportion of forest land and paddy ﬁeld is e ﬀ ective to the reduction of water nutrients. Besides, balancing the conﬁguration of water bodies, especially increasing the capacity of the pond, can signiﬁcantly alleviate the water pollution in the dry season. This study is useful to provide policy suggestion for reﬁned watershed management and water source planning basing on seasons and pollution sources.


Introduction
Degrading of water quality has been a significant problem worldwide [1] (pp. 104-121), [2][3][4][5]. This phenomenon has been more serious in China during the last two decades, which brings challenges in protection of water quality [6,7]. Therefore, keeping water resources clean and purifying the water for supply are the practicable ways [6,7]. Generally, among different kinds of water resources, headwater is widely used for drinking water supply and cropland irrigation [8][9][10]. As a result, the contaminated water sources are risks to public health [11,12]. Several factors can cause the degrading of water concentration of water quality parameters in the water source. However, studies on the small-scaled catchments were lacking.
Therefore, this study tries to find out the effect of landscape among different small-scaled water bodies in a typical headwater catchment in Southeastern China. In this study, multivariate statistical methods were introduced in the Lita Watershed based on the water quality data [42,46,50] in order (1) to identify the main sources of water pollutants in the study area, and (2) to examine how land-use patterns and water body configurations affect water quality.  (Figure 1). Located in the monsoon controlled region and in the subtropical climate zone, the annual precipitation in the watershed is around 1320 mm, and the annual average temperature 15.7 • C. There are two distinct seasons in Lita Watershed, which are the dry season, from October to March next year, and the wet season, from April to September. More than two-thirds of rainfalls are concentrated in the wet season. The main types of soils are paddy soil, sandy soil, meadow soil, and swamp soil. Generally, the soils are weak acid and oligotrophic or mesotrophic [9].

Study Area
The outlet of Lita Watershed, Lita Reservoir (capacity 5.9 million m 3 ), is one of the drinking water sources of Jurong City. Lita Reservoir also provides irrigation supply for surrounding villages scattering in an area of 8.1 km 2 . Furthermore, Lita Watershed is the Water Conservation Zone of Qinhuai River Watershed (Figure 1a), which flows through Nanjing City, the provincial capital with a population over 8.4 million.
As a mountainous rural watershed of Maoshan Mountain, main land-use types in Lita Watershed are forest, plantation, paddy, construction, and water body. The study area is divided into three regions, the upper, the lower, and the outlet (Figure 1b). In the upper regions, forest occupies the area of around half, and in the lower parts, the percentage of forest, plantation, and paddy are 25%, 28%, and 22%, separately. The area of water bodies is around 12% in whole watershed, and for construction, the percentage is less than 8%. Also, paddies are distributed near the stream to take advantage of the irrigation, and close to ponds, construction and poultry culture are common. Forest and plantation are often located in the sites which are far away from water. In Lita Watershed, the main species in plantation is beeches, which are less than five years, but the beech trees need fertilization.
Inside Lita Watershed, two subcatchments (Laorenshan and Chenzhuang) were chosen for comparison as they had different landscape and human activities. Two subcatchments were chosen for detailed and compared study, the Laorenshan Catchment ( Figure 1b) and the Chenzhang Catchment ( Figure 1b and enlarged in Figure 1c). Laorenshan Catchment has low development intensity, since less than 15% of lands are plantations and only about 5% are constructions. In contrast, in the Chenzhuang Catchment, the proportions of paddies, plantations, waters, and constructions are 22%, 21%, 13%, and 12%, separately. Meanwhile, the water bodies are various in Chenzhuang Catchment, including streams, ponds, and well waters, while in Laorenshan Catchment, only stream water was studied.

Water Sampling
Surface water and ground water were sampled in Lita Watershed with the following array: Thus, for water body types, there are pond (P1-P5), stream (S1-S6 and S11), reservoir (R1-R2), wetland (W1), and well (G1). The configurations of these water bodies are presented in Table 1, including the depth, width, and flow for streams, and the area and capacity for ponds, wetland, and reservoir.

Water Sampling
Surface water and ground water were sampled in Lita Watershed with the following array: Thus, for water body types, there are pond (P1-P5), stream (S1-S6 and S11), reservoir (R1-R2), wetland (W1), and well (G1). The configurations of these water bodies are presented in Table 1, including the depth, width, and flow for streams, and the area and capacity for ponds, wetland, and reservoir.
Taking locations into concern, there are upper (S1-S4), outlet (R2) and the rest are lower ones, as shown in Figure 1b,c. Water samples were collected every month from 2015 to 2016 by sampling bottles. Taking the pollution sources and conditions, 18 water quality indices are measured including discharge (Dis), temperature (TEM), dissolved oxygen (DO), total nitrogen (TN), nitrate nitrogen (NO 3 ), nitrite nitrogen (NO 2 ), ammonia nitrogen (NH 4 ), total phosphorus (TP), phosphate phosphorus (PO 4 ), chemical oxygen demand (COD), arsenic (As), cadmium (Cd), chromium (Cr), copper (Cu), mercury (Hg), manganese (Mn), nickel (Ni), lead (Pb), and Zinc (Zn). The standard methods [51] are introduced in the laboratory work of testing. All the water quality variables, units, analytical methods, and lowest detected limit are displayed in Table 2. The GIS data, including DEM and land-use types, are derived from Remote Sense Images and other materials. during 2015-2017 to test the real land-use patterns and correct the hydrological catchment division [52]. The meteorological data, including daily average temperature and precipitation, were gained from an automatic observation system located inside the watershed. In addition, the automatic observation tools were also used in recording the flow of outlet section data hourly by Infrared Detectors.

Statistical Analysis Methods
Spatial and temporal analyses took place in order to figure out the general characteristics of water body types and seasonality. To specify, the spatial analyses include the variation analysis and spatial cluster analysis, and the temporal analyses are mainly based on seasonal analysis. Afterwards, multivariate analyses of the water quality parameters were processed by hierarchical cluster analysis (HCA), Pearson's Linear Correlation (PLC), principal component analysis (PCA) and Canonical correlation analysis (CCA).
HCA was employed to describe the relationship through groups basing on the similarities within a class and dissimilarities among different classes. In HCA, clusters are formed stepwise by means of the Ward's method; using squared Euclidean distances cluster significance of 16 sampling sites was determined, the results of which can help explain the data and indicate patterns.
PLC was used and the matrix of correlation efficient was generated. In the system of PCA, firstly, discriminate analysis (DA) was introduced, which is based on discriminant functions (DF) for each group as in Equation (1) [53]): where i is the number of groups (G); K i is the constant inherent to each group; n is the number of parameter sets in the certain group; and W ij is the weight coefficient assigned by DF to a given selected parameter P ij . To standardize, forward stepwise and backward stepwise modes were employed to assess both the spatial and temporal variations of water quality through seasons and water body types. Then, varifactors were generated by principal component through Equation (2) [53]: where Z is the component score; i is the component number; j is the sample number; A is the component loading; X is the measured value of a variable; and n is the total number of variables. Finally, the most influential factors were found and the component scores were added, the basic concept of which is described in Equation (3) [53]: where Z is the component score; i is the component number; j is the sample number; A is the component loading; f is the factor score; X is the measured value of a variable; n is the total number of variables; and E is the residual term accounting for errors or other sources of variation. Canonical correlation analysis (CCA) was also applied in this study as a comparison and a deeper analysis. CCA is a cross-covariance matrix using the correlation between the comprehensive variables to reflect the overall correlation between the two groups of indicators. Its basic principle is: in order to grasp the correlation between the two groups of indicators in general, two representative comprehensive variables, U1 and V1, are extracted from the two groups of variables, which means that the linear combination of each variable in the two groups, respectively, and the correlation between the two comprehensive variables, is used to reflect the overall correlation between the two groups of indicators [54][55][56]. In this study, CCA was introduced to identify the impact power of land-use patterns on water quality parameters. Thus, the land-use patterns were viewed as environment factors, and the sampling sites were viewed as samples. Sensitivity of water quality parameters were analyzed through Pearson's correlation analysis and the principle component analysis Matrix of Pearson correlation coefficients were displayed with the marked significance range ( Figure 2). The coefficients with the absolute values larger than 0.6 are significant (p < 0.05), and the values between 0.3 and 0.6 are slightly significant (p < 0.1). Some parameters tend to be more sensitive and have several correlation linkages with other parameters, which are TEM, DO, TN, NO2, NH4, PO4, Cu, Hg, Pb, and Zn, namely, and the rest are insensitive. The most robust correlation pairs exist in both inside and inter groups, the robust correlation pairs exist such as PbHg and HgNO2. Meanwhile, the numbers of positive and negative coefficients are almost the same, but the most robust linear relationships are the positive ones.

Spatial and Temporal Variations of Water Quality
The concentration of water quality parameters were tested and are shown in Tables 3 and 4. The character of temporal variation can be simply described as the values are lower in the wet season for major parameters, but there are exceptions since, for TN, TP, PO 4 , and NH 4 , they are higher in the wet season than in the dry season. Interesting, the concentration of Hg and Pb are extremely high in November when compared with other months, and spatially, the outlet and upper regions have a lower index than lower regions. However, TN and COD are higher in the upper region and outlet. While heavy metals usually have alternative character to nutrients, thus, the values of the outlet ones and lower ones are close. The concentrations of Hg and Pb are extremely high in November when compared with other months.

Sensitivity of Water Quality Parameters
Sensitivity of water quality parameters were analyzed through Pearson's correlation analysis and the principle component analysis Matrix of Pearson correlation coefficients were displayed with the marked significance range (Figure 2). The coefficients with the absolute values larger than 0.6 are significant (p < 0.05), and the values between 0.3 and 0.6 are slightly significant (p < 0.1). Some parameters tend to be more sensitive and have several correlation linkages with other parameters, which are TEM, DO, TN, NO 2 , NH 4 , PO 4 , Cu, Hg, Pb, and Zn, namely, and the rest are insensitive. The most robust correlation pairs exist in both inside and inter groups, the robust correlation pairs exist such as PbHg and HgNO 2 . Meanwhile, the numbers of positive and negative coefficients are almost the same, but the most robust linear relationships are the positive ones.
Basing on the correlation analysis and the analysis of PCA, the most changeable parameters were found and compared (Figure 3). For the sensitive parameters, Cr, NO 3 , NO 2 , and COD, namely, shared the absolute values of over 0.75 in PC1 and PC2 through PCA, while the absolute values of others are less than 0.60. Temporally, the values of all the four parameters are higher in the dry season in the boxplots and the simulated normal distribution curves ( Figure 3). The real difference between is diminished due to the using of logarithmic scales. However, the spatial variations are unlikely to be obviously different, which are more scattering distributed. Additionally, COD and NO 2 are widely distributed in certain sites, and Cr and NO 3 are narrowed, and phenomenon is similar through seasons. Basing on the correlation analysis and the analysis of PCA, the most changeable parameters were found and compared (Figure 3). For the sensitive parameters, Cr, NO3, NO2, and COD, namely, shared the absolute values of over 0.75 in PC1 and PC2 through PCA, while the absolute values of others are less than 0.60. Temporally, the values of all the four parameters are higher in the dry season in the boxplots and the simulated normal distribution curves (Figure 3). The real difference between is diminished due to the using of logarithmic scales. However, the spatial variations are unlikely to be obviously different, which are more scattering distributed. Additionally, COD and NO2 are widely distributed in certain sites, and Cr and NO3 are narrowed, and phenomenon is similar through seasons.

Spatial Similarity of Sampling Sites
Effect of landscapes on water quality can be studied in several aspects, including the location and types of water bodies, and land-use patterns. Therefore, hierarchical clustering method was used to assess the similarity of sampling sites, and the dendograms were yielded and displayed in Figure 4. In Figure 4, the clusters indicate the similarity of sampling sites, and the distances are computed by Ward's method, using squared Euclidean distances processes, and the scales illustrate the differences of the coherence among sampling sites.
It is clear that distributions of nutrient parameters are more various than heavy metals. The reason lies in that there are clusters of only one or two members (Cluster 1 in Figure 4c,d) in the graphs of heavy metal parameters, but this phenomenon disappears in the graphs of nutrient parameters (Figure 4a,b). Also inside the clusters of nutrients, the elements are close to each other, but for heavy metal, the distances are usually far. Temporally, nutrients are steadier between wet and dry seasons than heavy metals, which means that the latter ones are more sensitive to seasonal

Spatial Similarity of Sampling Sites
Effect of landscapes on water quality can be studied in several aspects, including the location and types of water bodies, and land-use patterns. Therefore, hierarchical clustering method was used to assess the similarity of sampling sites, and the dendograms were yielded and displayed in Figure 4. In Figure 4, the clusters indicate the similarity of sampling sites, and the distances are computed by Ward's method, using squared Euclidean distances processes, and the scales illustrate the differences of the coherence among sampling sites.

Influence of Land-Use Patterns on Water Quality
Canonical correlation analysis (CCA) was introduced here to detect the interaction between land-use types and sampling sites and the biplots are presented in Figure 5. The red arrows represent the land-use patterns and the open circles represent sampling sites. The intersection angles between arrows indicate the relationships. To clarify, the acute angle represents positive correlation, and the smaller angle represents the greater positive correlation coefficient; the obtuse angle represents negative correlation, and the larger angle represents the greater absolute value of negative correlation coefficient; the right angle represents uncorrelation, and the angle close to right angle represents the weak correlation. Also, for points, the distance between the point and the arrow line means the relationship between them; thus, a short distance for a strong relationship. Generally, the graph patterns are more similar from seasons than from parameter groups. In detail, the patterns of nutrients, heavy metals, and sampling sites are illustrated separately. For nutrients, the plantation has different characters from other land-use sorts (Figure 5a,b), and the arrow of water body is close to the opposite direction of plantation. For heavy metal, forest is the opposite one to others, which is obviously different to the distributions of arrows (Figure 5a,b). The distribution of points means the similarity among sampling sites. In the wet season, the points are more scattered, and in the dry season, they are concentrated (Figure 5a,b). This pattern is the same for both nutrients and heavy metals. It is clear that distributions of nutrient parameters are more various than heavy metals. The reason lies in that there are clusters of only one or two members (Cluster 1 in Figure 4c,d) in the graphs of heavy metal parameters, but this phenomenon disappears in the graphs of nutrient parameters (Figure 4a,b). Also inside the clusters of nutrients, the elements are close to each other, but for heavy metal, the distances are usually far. Temporally, nutrients are steadier between wet and dry seasons than heavy metals, which means that the latter ones are more sensitive to seasonal changes.

Influence of Land-Use Patterns on Water Quality
Canonical correlation analysis (CCA) was introduced here to detect the interaction between land-use types and sampling sites and the biplots are presented in Figure 5. The red arrows represent the land-use patterns and the open circles represent sampling sites. The intersection angles between arrows indicate the relationships. To clarify, the acute angle represents positive correlation, and the smaller angle represents the greater positive correlation coefficient; the obtuse angle represents negative correlation, and the larger angle represents the greater absolute value of negative correlation coefficient; the right angle represents uncorrelation, and the angle close to right angle represents the weak correlation. Also, for points, the distance between the point and the arrow line means the relationship between them; thus, a short distance for a strong relationship. Generally, the graph patterns are more similar from seasons than from parameter groups. In detail, the patterns of nutrients, heavy metals, and sampling sites are illustrated separately. For nutrients, the plantation has different characters from other land-use sorts (Figure 5a,b), and the arrow of water body is close to the opposite direction of plantation. For heavy metal, forest is the opposite one to others, which is obviously different to the distributions of arrows (Figure 5a,b). The distribution of points means the similarity among sampling sites. In the wet season, the points are more scattered, and in the dry season, they are concentrated (Figure 5a,b). This pattern is the same for both nutrients and heavy metals.

Source Identification of Aquatic Heavy Metals and Nutrients
The spatial distribution of heavy metals in Lita Watershed showed the mix patterns through wet and dry seasons (Table 4 and Figure 2). For instance, the temporal distribution patterns of heavy metal have the opposite characters to the nitrogen and phosphorus. Also, the seasonal difference is significant, as the peak of pollution often occurred in November and December in 2015 and 2016, while during this period of dry season, the concentration of nutrients remained in the low level. Furthermore, an interesting finding is that, in Figure 4a,c, the upper site S2 and the lower site P5 share the same cluster with short distance. However, the upper regions less affected by anthropogenic aspects are unlikely to have the similar characters of the lower regions. So, there

Source Identification of Aquatic Heavy Metals and Nutrients
The spatial distribution of heavy metals in Lita Watershed showed the mix patterns through wet and dry seasons (Table 4 and Figure 2). For instance, the temporal distribution patterns of heavy metal have the opposite characters to the nitrogen and phosphorus. Also, the seasonal difference is significant, as the peak of pollution often occurred in November and December in 2015 and 2016, while during this period of dry season, the concentration of nutrients remained in the low level. Furthermore, an interesting finding is that, in Figure 4a,c, the upper site S2 and the lower site P5 share the same cluster with short distance. However, the upper regions less affected by anthropogenic aspects are unlikely to have the similar characters of the lower regions. So, there should be some sources of pollutants with slight linkages with locations.
In order to compare the results and identify the sources of heavy metals, the results of similar studies were presented in Table 5. The aquatic heavy metal is mainly caused by point pollution, such as energy production, manufacturing, and transport [64][65][66]. Other than the industrial area, there were no obvious industrial activities in Lita Watershed during 2015 and 2016. Some studies demonstrated that the fertilization, fecal discharge, and the incineration and burial of waste can also be the drivers of heavy metal, the process of pollution can be conveyed through soil, not directly [58,67]. However, the amount of these tracing elements by nonpoint sources is at the low level [68] (pp. 657-678), [69]; for instance, the concentration of Hg and Pb in water are unlikely to surpass 2.0 µg·L −1 in surface water samples when the amount of water is ample [57,59]. Therefore, nonpoint source is unlikely to be the main source of heavy metal.
As a result, the wet and dry depositions are the major source of heavy metal in this study, especially for Hg and Pb. Our study area is located in an industrialized region, Yangtze River Delta and surrounded by several large cities, including Shanghai, Suzhou, and Nanjing [67,70]. Moreover, during autumn and winter, the strong monsoon can carry the pollution from Northern China, and several recent studies indicated the linkage between industry and heavy metal pollution in water [50,60]. By consulting similar areas, the concentrations of Hg and Pb are close to those in Lita Watershed [61].
The source identifying and moving mechanism of nutrients among water bodies and land-use patterns can also be introduced to explain the parameters of heavy metals according to the correlation coefficient matrix (Figure 2). The pairs of most significant linear relationships can be grouped by three clues, the inherent properties, seasonality, and sources. For inherent properties, the negative relationship between DO and COD is obvious. For seasonality, in the wet season, the temperature is high, and the concentration of TN and TP is usually relatively high, while in the dry season, the concentration of NO 2 is high in this study, as well as some heavy metals, including Hg and Pb. Thus, the negative correlation pairs (Tem-Hg, Tem-Pb, and Tem-NO 2 ) and positive correlation pairs (Hg-NO 2 and Pb-NO 2 ) can be explained [46,50]. For sources, the NO 3 and NH 4 are different forms of nitrogen element, and can be transformed to each other. In the countryside of China, both fertilization and fecal discharge are contributed to NO 3 and NH 4 , and this phenomenon is also being studied in recent researches [8,9,46]. Another pair with similar sources is Pb and Hg, mainly in power plants [71,72], so the positive correlation is apparent.

Water Body Configurations, Land-Use Patterns and Their Driving Effects on Water Quality
In this study area, Lita Watershed, and many other similar watersheds in the Southern China, the type of water bodies is various, as well as their effects on water quality. For example, differing to lakes and reservoirs, pond water is well mixed, and less protected, and the quality of water is strongly determined by the environment and condition in vicinity, not the whole watershed. Generally, the water quality is degraded gradually from S1 to S6 and from P1 to P5, but for reservoirs, the outlet water is much better than the inlet one, which indicates the effect of purifying, and the clue is revealed by the fact that the outlet is included in the same cluster with the upper sites. Other than locations, this article attempts to trace the linkage between the configurations of water bodies (Table 1) and their influence on water quality.
The scale of water bodies is an important factor to influence water quality. For ponds in the similar region, the larger ones tend to have cleaner water, taking P1 and P2 for instance. However, this phenomenon is not obvious in stream water. The reason can be that the amount of stream water is changeable during seasons, so the downstream water can be less than the upper one in several times. Two special water bodies, the wetland water and ground water, are extremely different; thus, the former is one of the best and the latter is the worse. The plants in wetland can clean up water [73], and the paddies have the similar function. In this study, the site W1 is the upper region perennial wild wetland, so the water quality can be higher. When taking groups of water quality variables into concern, pond water, reservoir water, and ground water are more likely to be affected by heavy metal pollutions than river water. The solution can be found in two aspects, the interaction between air and the soil, pond water, and reservoir water tend to be dominated by the first factor, and the second factor for ground water [74][75][76].
The land-use types and the activities on the certain types (fertilization, irrigation, aquaculture, and heaviest) have the indirect influence on water quality [41,77,78]. The plantation is opposite to the others due to the large amount of fertilizer and less water (Figure 5a,b); as in the study area, the amounts of fertilizer in plantations are large [9]. Paddies are also heavily fertilized, but the effect of water dilution and the quick absorption of rice plant decreases the accumulation of nutrients greatly [27,79]. Also, Figure 5a,b imply that the arrow of water body is close to the opposite direction of plantation, which means that the dilution effect of water body is important, and for sampling sites, the ones with the best water quality are located in the first quadrant. Additionally, after our study period, thus, during 2017 and 2018, there were extreme land-use changes within Chenzhuang Watershed. As a result, influence of certain land-use patterns on water quality needs an alternative focus on the land-use changes.
Several researches demonstrated that the forest can reduce the air deposition of heavy metals [80,81], which is similar to this study, as the forest has the opposite effect to the others (Figure 5c,d). Then for plantation, the leaf area index (LAI) is low, which means that the beech trees are too small to absorb enough heavy metal, and the differences of the ecological functions between plantation and forests are concerned by other studies [27,82]. Additionally, the scattered distribution of points in Figure 5c,d may be evidence to support that the source of heavy metals is mainly from air, not water [83].

Management Implications
Water quality in headwater area is fundamental for drinking water and food safety. In densely populated and rapid industrialized areas of China, systematic assessment and management of headwater catchment is urgent. According to the results and analysis of this study, policy of headwater protection can be attributed into three steps. First, reservoir plays the key role in water conservation, and the purifying effect is significant (Figure 4). Therefore, the activities upon the reservoir should be forbidden, including poultry culturing and tourist industry. Also, maintaining the water level and capability of the reservoir can help fulfill its function [16].
Second, upstream protection is the effective way to reduce the pollutant discharging. Many studies demonstrated that the construction of forest-soil buffer zones and other engineering actions taken on soils can bring benefits to reducing pollution [27,84]. However, this study points out that making good use of water body systems is also important (Tables 3 and 4). For illustration, the pond systems are used for interception, reservation, and sediment, and the streams systems are used for water supply. In Southeastern China, the paddies and other seasonal water bodies have similar functions as the ponds. Two main factors decide the functions of water bodies, the scales, and the land-use patterns in the subcatchment. So, taking the scales of paddies into concern, in the wet season, the paddies dominate the function of water purifying while, in the dry season, ponds take the role. For land-use patterns, forest, grass, and paddies, having good water capacities themselves, are less reliant on the water bodies for pollution reducing, but construction is opposite. The large subcatchments with big amounts of pollutant need to match the water bodies with high proportions and various types (Figures 1 and 4). Other researches also demonstrated that the upper regions with forest mainly usually can cut off 30% of nitrogen and phosphorus leaking than the lower regions with croplands and urban areas mainly [9,27,78]. Experiences in Lita Watershed can be applied in two situations, the areas where the water body types and shapes are adjustable and where the soil management is hard to realize.
Third, tracing the temporal and spatial variation of water quality parameters can help in finding the shortcut to improve water quality. The management can be described from two aspects. Clues between seasonal distribution of nutrients and heavy metals reveal their sources and movements. Moreover, the relationships among different parameters also contribute to the targeted solutions based on pollution reduction. Also, studies and practices provide evidence that the optimal watershed plan can reduce the risk of flood and other disasters, and benefit the economy and society [85,86].

Conclusions
Basing on the multistatistics analysis on Lita Reservoir, landscapes have strong influence on headwater quality in both groups of nutrient parameters and heavy metal parameters. Results show that the most sensitive water quality parameters are Cr, NO 3 , NO 2 , and COD. The configurations of water bodies (location, types, and scales) are the direct factors to determine the effect of reducing the water pollution, while land-use patterns in the surrounding area have the indirect effect on the movement of pollutants. The influence of landscape on water quality is more sensitive in the aspect of season rather than parameters. This finding means that adjusting the water body configurations and land-use patterns can reduce the loads of contaminants from fertilization and poultry farming significantly, but remain to be useless for heavy metals. These findings and regulations are useful for making the targeted plans on headwater protections in three aspects: source identifying and controlling, water body tracing and adjusting, and predicting and early warning of water pollution events. Additional studies and practical water quality models in monsoon controlled small watersheds are needed to quantify the relationship among water quality parameters in order to make refinement managements.