Spatiotemporal Dynamics of Nitrogen Transport in the Qiandao Lake Basin, a Large Hilly Monsoon Basin of Southeastern China

: The Qiandao Lake Basin (QLB), which occupies low hilly terrain in the monsoon region of southeastern China, is facing serious environmental challenges due to human activities and climate change. Here, we investigated source attribution, transport processes, and the spatiotemporal dynamics of nitrogen (N) movement in the QLB using the Soil and Water Assessment Tool (SWAT), a physical-based model. The goal was to generate key localized vegetative parameters and agronomic variables to serve as credible information on N sources and as a reference for basin management. The simulation indicated that the basin’s annual average total nitrogen (TN) load between 2007 and 2016 was 11,474 tons. Steep slopes with low vegetation coverage signiﬁcantly inﬂuenced the spatiotemporal distribution of N and its transport process. Monthly average TN loads peaked in June due to intensive fertilization of tea plantations and other agricultural areas and then dropped rapidly in July. Subsurface ﬂow is the key transport pathway, with approximately 70% of N loads originating within Anhui Province, which occupies just 58% of the basin area. The TN yields of sub-basins vary considerably and have strong spatial e ﬀ ects on incremental loads entering the basin’ major stream, the Xin’anjiang River. The largest contributor to N loads was domestic sewage (21.8%), followed by livestock production (20.8%), cropland (18.6%), tea land (15.5%), forest land (10.9%), atmospheric deposition (5.6%), orchards (4.6%), industry (1.4%), and other land (0.8%). Our simulation underscores the urgency of increasing the e ﬃ ciency of the wastewater treatment, conserving slope land, and optimizing agricultural management as components of a comprehensive policy to control N pollution in the basin.


Introduction
The nitrogen (N) cycle is dynamic and strongly influenced by both human activities and physical conditions, especially in environmentally sensitive areas that are prone to nutrient pollution [1,2]. Agricultural practices and domestic sewage account for a substantial proportion of released N, which subsequently enters waterways via pathways that are diverse and poorly understood. Climatic

Data Collection
As shown in Figure 1, the entire QLB was delineated into 96 sub-basins based on the 25-meter digital elevation model. High resolution, 2.1-meter land use data were derived from ZY-3 satellite imagery for 2015, which was purchased from the Chinese Geographical Monitoring Cloud Platform (http://www.dsac.cn/Data/Product/Detail/1009). Whenever possible, e.g., for forest land and built land, default parameters from the land use database were used in SWAT. eCognition Developer, an object-based tool for automatic analysis of remotely sensed data (http://www.ecognition.com/suite/ecognition-developer), was employed to identify land use types that are typical for the area, e.g., tea land and orchard land. Multiresolution segmentation was

Data Collection
As shown in Figure 1, the entire QLB was delineated into 96 sub-basins based on the 25-meter digital elevation model. High resolution, 2.1-meter land use data were derived from ZY-3 satellite imagery for 2015, which was purchased from the Chinese Geographical Monitoring Cloud Platform (http://www.dsac.cn/Data/Product/Detail/1009). Whenever possible, e.g., for forest land and built land, default parameters from the land use database were used in SWAT. eCognition Developer, an Water 2020, 12, 1075 4 of 20 object-based tool for automatic analysis of remotely sensed data (http://www.ecognition.com/suite/ ecognition-developer), was employed to identify land use types that are typical for the area, e.g., tea land and orchard land. Multiresolution segmentation was adopted for this purpose, with scope of classification algorithms from sample-based nearest neighbor analysis. Every imagery object was checked by visual interpretation, with verification by field observations at 100 separate sites in the basin. The final land use classification map is shown in Figure S1 (see Supplementary Materials for Figure S1).
Soil types and their physical properties were extracted from databases of the Nanjing Institute of Soil Science and, when necessary, verified by field sampling to generate the map shown in Figure  S2 (see Supplementary Materials for Figure S2). Distributed rainfall data were collected from 54 rainfall gauges within the QLB (Figure 1). Daily streamflow data from two hydrological stations (Tunxi and Yuliang, see Figure 1) were obtained from the Huangshan Hydrology Bureau and published in Chinese Hydrological Yearbooks [32]. Monthly water quality data were from records of the Chun'an Environmental Protection Bureau. TN data, expressed as the sum of different forms of nitrogen including ammonia, nitrate, nitrite and organic N, were measured according to the National Surface Water Quality Standard (GB 3838-2002) and obtained from the Chun'an Environmental Protection Bureau, which also supplied precipitation and dry deposition data used to determine atmospheric deposition of TN.
The Chun'an Environmental Protection Bureau and Chun'an County Statistical Yearbooks [33,34] provided additional information for the model, including the locations of industrial and domestic wastewater outlets and discharge from them. The spatial distribution of livestock production facilities, their scales of production, and discharge from them was based on governmental monitoring of wastewater streams. Supplemental Information about cropping systems, agronomic practices, management decisions, and rural sewage treatment was obtained from detailed governmental interviews with 487 farmers from 44 villages in the study region. Sewage in the QLB is collected into centralized treatment facilities that discharge wastewater into nearby rivers, and thus for the purpose of SWAT modeling, domestic, industrial, and animal agricultural pollution sources could be generalized as point sources that discharge into point outlets.

SWAT Model Configuration
Soil data ( Figure S2) were transformed into a modeling format by means of standardized definition and interpretation of SWAT parameters, and discharge and associated TN loads from domestic, industrial, and animal agriculture from each sub-basin were configured into the model as described above. Source attribution of atmospheric N deposition was generalized as the single direct source of pollution into water bodies as follows: deposition of ammonium (0.97 mg/L), deposition of nitrate (0.73 mg/L), dry deposition of ammonium (0.434 kg/ha/yr), and dry deposition of nitrate (0.61 kg/ha/yr).
Information from field surveys of agricultural practices in the QLB was used to guide the scheduling of cropland, tea land, and orchard management for the model. Cropland and orchards were operated with heat unit scheduling, but because of its unique growth and management characteristics, tea land was operated with date scheduling in Table S1 (see Supplementary Materials for Table S1). Localized parameters for tea land were optimized on the basis of multi-year field measurements during the growth season, as well as information obtained in similar areas of southeastern China, and then manually assembled into the final input suite for the SWAT database (Table S1).
The previously used SUFI-2 algorithm was employed to analyze the sensitivity, calibration, validation, and uncertainty of the SWAT model [35,36]. Three statistical indicators were utilized to evaluate the model's performance during calibration and validation: the coefficient of determination (R 2 ), which is an indicator of the goodness of fit between the observed and simulated results; the Nash-Sutcliffe simulation efficiency (ENS), which determines the ability of the model to predict the 1:1 line of correspondence between the observed and simulated values; and PBIAS (Percent bias), which measures the average tendency of simulated values to be larger or smaller than corresponding observed values [37,38]. Previously defined metrics of model performance were adopted. These assign satisfactory, good, and very good performance to ENS values of 0.5, 0.65, and 0.75, and R 2 values of 0.6, 0.7, and 0.8, respectively [39,40]. PBIAS values of ±25% and ±25% ≤ ±40% are considered to indicate very good and good model performance for N, respectively. The corresponding values for streamflow are ±10% and ±10% ≤ ±15%, respectively.
The SWAT model was used for simulations during the interval from 2003 to 2016. The first four-year interval, from 2003 to 2006, was employed as a warm-up period to allow the streamflow variables to reach a limited range of initial values (see Table S2 for details in Supplementary Materials). Observed daily data from 2007 to 2012 were subsequently used for calibration and those from 2013 to 2016 for validation. N load simulations at the two sampling sites, e.g., Jiekou and Baqian, were executed simultaneously, and iterations of the entire model were performed until satisfactory results were obtained.

Calibration and Sensitivity Analysis of the SWAT Model
Parameters such as cropping systems, fertilization protocols, and soil conditions can vary significantly in different basins, with substantial impacts on modeling outcomes [15,18,41]. Appropriate calibration with parameters that are optimized for the QLB is consequently important, especially for tea, which is intensively managed, known to have undergone considerable expansion during the study period, and recognized to be an important source of nutrient pollution [42,43]. Twenty-one optimized SWAT parameters for tea land in the QLB are given in Table S2. Three crucial parameters (Table 1) diverge significantly from those provided by the SWAT model for forest land and land used for cultivation of other crops. This indicates that both maximum root depth and harvest index are comparatively low for tea, but demand for N in seed is relatively high. The most important and sensitive parameters during the calibration of the SWAT model were groundwater (.gw), soil (.sol), basins (.bsn), and management (.mgt). In agreement with previous studies, the moisture condition II curve of the soil conservation service (SCS) curve number method (CN2) was the most sensitive parameter for streamflow [18,36]. This was also the key parameter for runoff yield, as would be expected in a basin characterized by a hilly terrain, agricultural land, and abundant rainfall. Other important parameters representing land-cover features were OV_N, CANMX, HRU_SLP, GWQMN, SOL_AWC, TRNSRCH, and ALPHA_BNK. The most sensitive parameters during calibration of TN loads included N source, transport process, and transformation. The most sensitive nitrogen source parameters were RCN, CMN, SOL_NO3, and FRT_SURFACE, and the most sensitive N transformation and transport process parameters were NPERCO, LAT_ORGN, ERORGN, and BIOMIX.
Overall attribution patterns of TN pollution sources are quite variable across the QLB (Table 2). CN2, a function of land use and soil permeability and thus runoff yield, was the most sensitive hydrological parameter for streamflow and TN output in the basin. Low calibrated values for the initial nitrate concentration in the soil layer and for the N percolation coefficient provide evidence that lateral flow in the shallow soil layer covering much of the basin's steep topography is the main pathway of TN loads. These observations are consistent with an earlier SWAT modeling effort that was confined to the upper half of the basin in Anhui Province [44]. However, a small Manning's 'n' value and low transmission losses (Table 2) indicate that, in contrast to findings of the earlier study, surface and lateral runoff significantly affect TN transport and loss across the basin. We attribute these differences to the larger, basin-wide scope of the current study area, which alters fundamental topographic ( Figure 1) and land use ( Figure S1) relationships and includes soil types not prevalent in the upper basin ( Figure S2).

Validation of Streamflow and TN Loads
Distributed rainfall data, which are critical for understanding spatiotemporal dimensions of key nutrient transport processes in hilly areas [45,46], exhibited strong annual variation, ranging from a low of 1375 mm to a high of 2373 mm (see Figure S3 in Supplementary Materials). Significant annual fluctuation with a coefficient of variation above 0.1 was recorded for the entire basin, and monthly precipitation was, as expected, characterized by spatial heterogeneity [47]. Rainfall tended to rise rapidly from May to June and decrease sharply from June to July. Average monthly rainfall peaked in June (389 mm), which accounted for over 20% of the total annual accumulation ( Figure S3).
Rainfall regulates streamflow and TN transport in the QLB, but other driving forces also occur, and monitoring data are limited. Instead of simply relying on the coupling model [12,48], we applied stepwise calibration and validation to the stimulation to optimize model performance in different regions of the basin. TN data are available from the border between the two provinces at Jiekou and from the basin's outlet at Baqian; streamflow data are also available at Baqian and additionally from two upstream stations at Tunxi and Yuliang ( Figure 1). Following the warm-up period, the SWAT model was consequently calibrated and validated for the variables available at each of these sites over a decade-long interval from 2007 to 2016 (Figures 2 and 3).   The correspondence between measured and simulated values for TN and streamflow at Baqian is illustrated in Figure 2. Both variables are characterized by cyclical fluctuations, with annual maxima during the summer rainy season. The TN peaks tend to be displaced to the right compared to corresponding peaks for streamflow, likely because of differences in the frequency of measurement (monthly for TN, daily for streamflow). There is some divergence between observed and simulated TN levels during the peak flow periods of 2015 and 2016, both of which were wet years with extreme rainfall events. R 2 and NSE values (Table 3) are nevertheless indicative of the SWAT model's good and very good performance, respectively, during the calibration and validation periods [35,36]. The SWAT model was also calibrated and validated at sites above the lake and distant from the outlet at Baqian. Observed monthly streamflow at Tunxi and Yuliang was much lower than that at the basin's outlet, and the model accurately simulated this parameter (Figure 3), i.e., the performance of the simulation model at Tunxi and Yuliang was rated as very good according to all three statistical measurements (Table 3). In comparison, performance for streamflow at Baqian was rated as very good by the R 2 and NSE statistics and good by the PBIAS statistic. The model was somewhat less accurate in simulating TN levels at Baqian and Jiekou, especially during the validation period, when TN levels tended to be underestimated ( Figure 3). Nevertheless, with just one exception (the NSE statistic at Jiekou during the validation period), model performance at these two sites was uniformly rated as good to very good, indicating that overall model performance was satisfactory.

Spatiotemporal Patterns of TN Loads and Yields
TN loads over the interval from 2007 to 2016 at the outlet of the QLB are strongly correlated with variations in average rainfall in the basin, both when measured annually ( Figure 4A) and monthly ( Figure 4B). This indicates that rainfall trends are key drivers of N discharge. TN loads spiked early in the annual fertilization cycle, which begins in March, and reached a maximum in June, when precipitation was also greatest and the risk of erosion highest. Precipitation and TN loads declined thereafter.
The mean annual TN load of the basin was 11,474 t, as calculated from simulations spanning the period 2007-2016. This corresponds to an overall mean TN yield of 1.1 t/km 2 /yr. The TN point sources that were generalized into three categories for simulation with the SWAT model jointly accounted for 44% of the annual TN load in the basin. The contribution of industrial sources was negligible in comparison to that of domestic sewage and livestock production, which accounted for nearly equivalent amounts of TN loads to the basin ( Figure 5A). In addition, nearly 6% of the total could be attributed to atmospheric deposition, with other non-point sources of TN accounting for the remaining half of TN loads in the basin. On a percentage basis, the most important of these sources were cropland and tea land, followed by forest land and orchard land ( Figure 5A). These relationships obscure the fact that intensively fertilized tea land contributes 30 times as much TN as forest land does when expressed Water 2020, 12, 1075 9 of 20 in terms of annual TN yields per km 2 ( Figure 5B). Other agricultural land, including orchards and, to a lesser extent, cropland, also account for disproportionate amounts of non-point source TN pollution when measured this way.
Water 2020, 12, x FOR PEER REVIEW 9 of 20 relationships obscure the fact that intensively fertilized tea land contributes 30 times as much TN as forest land does when expressed in terms of annual TN yields per km 2 ( Figure 5B). Other agricultural land, including orchards and, to a lesser extent, cropland, also account for disproportionate amounts of non-point source TN pollution when measured this way.  The spatial patterns of TN loads and TN yields in each of the 96 sub-basins of the QLB are mapped in Figure 6. Upstream sub-basins in Anhui Province jointly account for 70% of the basin's TN loads, but significant sub-basin to sub-basin heterogeneity is obvious, and several hotspots are apparent ( Figure 6A). The largest of these lies near the Huangshan City urban core. Downstream riparian areas near the lake also emit significant TN loads, but most sub-basins in Zhejiang Province are relatively minor contributors ( Figure 6A). A different pattern emerges when TN yield is considered ( Figure 6B). The highest such yields are from a single cluster of sub-basins, and their magnitude tends to progressively diminish with distance from this hotspot. TN yields in other subbasins, including those near the lake, are generally low. A few additional areas of high TN yield are nevertheless evident, including one near the rapidly urbanizing tourist area in Zhejiang Province's Chun'an County. relationships obscure the fact that intensively fertilized tea land contributes 30 times as much TN as forest land does when expressed in terms of annual TN yields per km 2 ( Figure 5B). Other agricultural land, including orchards and, to a lesser extent, cropland, also account for disproportionate amounts of non-point source TN pollution when measured this way.  The spatial patterns of TN loads and TN yields in each of the 96 sub-basins of the QLB are mapped in Figure 6. Upstream sub-basins in Anhui Province jointly account for 70% of the basin's TN loads, but significant sub-basin to sub-basin heterogeneity is obvious, and several hotspots are apparent ( Figure 6A). The largest of these lies near the Huangshan City urban core. Downstream riparian areas near the lake also emit significant TN loads, but most sub-basins in Zhejiang Province are relatively minor contributors ( Figure 6A). A different pattern emerges when TN yield is considered ( Figure 6B). The highest such yields are from a single cluster of sub-basins, and their magnitude tends to progressively diminish with distance from this hotspot. TN yields in other subbasins, including those near the lake, are generally low. A few additional areas of high TN yield are nevertheless evident, including one near the rapidly urbanizing tourist area in Zhejiang Province's Chun'an County. The spatial patterns of TN loads and TN yields in each of the 96 sub-basins of the QLB are mapped in Figure 6. Upstream sub-basins in Anhui Province jointly account for 70% of the basin's TN loads, but significant sub-basin to sub-basin heterogeneity is obvious, and several hotspots are apparent ( Figure 6A). The largest of these lies near the Huangshan City urban core. Downstream riparian areas near the lake also emit significant TN loads, but most sub-basins in Zhejiang Province are relatively minor contributors ( Figure 6A). A different pattern emerges when TN yield is considered ( Figure 6B). The highest such yields are from a single cluster of sub-basins, and their magnitude tends to progressively diminish with distance from this hotspot. TN yields in other sub-basins, including those near the lake, are generally low. A few additional areas of high TN yield are nevertheless evident, including one near the rapidly urbanizing tourist area in Zhejiang Province's Chun'an County.

Sources and Spatial Dynamics of TN Entering the Main Reach of the Xin'anjiang River
Optimization of pollution management requires as much spatiotemporal understanding as possible about sources and transport of pollutants. We consequently used detailed information about TN loads and yields in the sub-basins of the QLB to simulate sources, transport, and entry points of TN pollution into the main stem of the Xin'anjiang River. Figure 7 relates average annual contributions of different pollution sources of TN to their entry along the river's pathway through the QLB. Almost no TN enters the upper 45% of the river's course, but thereafter, three distinct surges of TN become evident. The first, which accounts for 25% of the total TN pollution entering the river (Figure 7), is at Tunxi (location 154.6 km; see Figure 1). Increased discharges of TN from agriculture, especially livestock production, predominate at this site and account for the majority of the surge. A second, even sharper surge in TN is at location 188.7 km, near a tributary that flows through the municipalities of Shexian and Huizhou (both part of Huangshan City) and nearby highly populated areas before entering the Xin'anjiang River. Significant increases in TN attributable to domestic sewage are evident at this site. Downstream TN pollution from all sources then increases gradually until a third surge appears at location 287.1 km in Zhejiang Province. Elevated TN

Sources and Spatial Dynamics of TN Entering the Main Reach of the Xin'anjiang River
Optimization of pollution management requires as much spatiotemporal understanding as possible about sources and transport of pollutants. We consequently used detailed information about TN loads and yields in the sub-basins of the QLB to simulate sources, transport, and entry points of TN pollution into the main stem of the Xin'anjiang River. Figure 7 relates average annual contributions of different pollution sources of TN to their entry along the river's pathway through the QLB. Almost no TN enters the upper 45% of the river's course, but thereafter, three distinct surges of TN become evident. The first, which accounts for 25% of the total TN pollution entering the river (Figure 7), is at Tunxi (location 154.6 km; see Figure 1). Increased discharges of TN from agriculture, especially livestock production, predominate at this site and account for the majority of the surge.

Sources and Spatial Dynamics of TN Entering the Main Reach of the Xin'anjiang River
Optimization of pollution management requires as much spatiotemporal understanding as possible about sources and transport of pollutants. We consequently used detailed information about TN loads and yields in the sub-basins of the QLB to simulate sources, transport, and entry points of TN pollution into the main stem of the Xin'anjiang River. Figure 7 relates average annual contributions of different pollution sources of TN to their entry along the river's pathway through the QLB. Almost no TN enters the upper 45% of the river's course, but thereafter, three distinct surges of TN become evident. The first, which accounts for 25% of the total TN pollution entering the river (Figure 7), is at Tunxi (location 154.6 km; see Figure 1). Increased discharges of TN from agriculture, especially livestock production, predominate at this site and account for the majority of the surge. A second, even sharper surge in TN is at location 188.7 km, near a tributary that flows through the municipalities of Shexian and Huizhou (both part of Huangshan City) and nearby highly populated areas before entering the Xin'anjiang River. Significant increases in TN attributable to domestic sewage are evident at this site. Downstream TN pollution from all sources then increases gradually until a third surge appears at location 287.1 km in Zhejiang Province. Elevated TN A second, even sharper surge in TN is at location 188.7 km, near a tributary that flows through the municipalities of Shexian and Huizhou (both part of Huangshan City) and nearby highly populated areas before entering the Xin'anjiang River. Significant increases in TN attributable to domestic sewage are evident at this site. Downstream TN pollution from all sources then increases gradually until a third surge appears at location 287.1 km in Zhejiang Province. Elevated TN discharge from highly erodible forested slopes [2], and to a lesser extent from domestic sewage and orchard, are major contributors to the increase at this location. Distinct spatial patterns of TN release from plant-based agriculture are clearly evident along the course of the river. Thus, the accumulation of TN from cropland and tea land reaches its near maximum before the Xin'anjiang River flows out of Anhui Province, but most of the TN accumulation from orchard land occurs later, as the river flows through Zhejiang Province.
Delivery of TN from non-point sources to the Xin'anjiang River can follow three general pathways: lateral flow, groundwater, and surface flow. The contributions of these pathways to TN accumulation in the river are visualized in Figure 8, which reveals four significant surges of TN from non-point sources and identifies spatially explicit contributions of lateral flow, groundwater, and surface water. The first surge, at Tunxi (location 154.6 km), is primarily attributable to increases from surface flow and groundwater, but the second, at location 208.3 km, is primarily attributable to increases in lateral and surface flow. The third and fourth surges peak at Jiekou, on the border between the provinces (location 243.7 km) and within Zhejiang Province (location 287.1 km), respectively. Almost all of the incremental increase at both of these locations is contributed by lateral flow, which ultimately accounted for nearly two-thirds of TN transport in the river.
Water 2020, 12, x FOR PEER REVIEW 11 of 20 discharge from highly erodible forested slopes [2], and to a lesser extent from domestic sewage and orchard, are major contributors to the increase at this location. Distinct spatial patterns of TN release from plant-based agriculture are clearly evident along the course of the river. Thus, the accumulation of TN from cropland and tea land reaches its near maximum before the Xin'anjiang River flows out of Anhui Province, but most of the TN accumulation from orchard land occurs later, as the river flows through Zhejiang Province. Delivery of TN from non-point sources to the Xin'anjiang River can follow three general pathways: lateral flow, groundwater, and surface flow. The contributions of these pathways to TN accumulation in the river are visualized in Figure 8, which reveals four significant surges of TN from non-point sources and identifies spatially explicit contributions of lateral flow, groundwater, and surface water. The first surge, at Tunxi (location 154.6 km), is primarily attributable to increases from surface flow and groundwater, but the second, at location 208.3 km, is primarily attributable to increases in lateral and surface flow. The third and fourth surges peak at Jiekou, on the border between the provinces (location 243.7 km) and within Zhejiang Province (location 287.1 km), respectively. Almost all of the incremental increase at both of these locations is contributed by lateral flow, which ultimately accounted for nearly two-thirds of TN transport in the river. The monthly dynamics of TN discharge from non-point sources via groundwater, surface flow, and lateral flow (Figure 9) generally correspond to monthly rainfall (Figure 4). Release of TN is most significant from February through August and sharply reduced during the other months, when TN from groundwater and surface flows become insignificant. Lateral flow and surface flow transport roughly equivalent amounts of TN during the period of peak movement, except for February and March, when lateral flow predominates and exceeds the contribution from surface flow by three to four-fold. Groundwater movement of TN, on the other hand, increases progressively during this same period, but is of minor significance compared to the other two pathways.

Spatio-Temporal Dynamics of TN Transport
The QLB is characterized by hilly topography, diverse, rapidly intensifying agricultural production, and several urban areas where domestic and industrial activities are concentrated [25,31]. It is no wonder, then, that nutrient pollution in the basin is governed by dynamic upstream to downstream transport processes that are spatially and temporally heterogeneous and poorly The monthly dynamics of TN discharge from non-point sources via groundwater, surface flow, and lateral flow (Figure 9) generally correspond to monthly rainfall (Figure 4). Release of TN is most significant from February through August and sharply reduced during the other months, when TN from groundwater and surface flows become insignificant. Lateral flow and surface flow transport roughly equivalent amounts of TN during the period of peak movement, except for February and March, when lateral flow predominates and exceeds the contribution from surface flow by three to four-fold. Groundwater movement of TN, on the other hand, increases progressively during this same period, but is of minor significance compared to the other two pathways.
understood [24,49]. It has been known for several decades that these factors have seriously compromised water quality in the basin [27]. China's first transboundary Payment for Ecosystem Services (PES) plan was consequently established in the QLB in 2012 [50,51]. Although this scheme has generated significant investment to reduce nutrient pollution entering the lower basin from Anhui Province, many fundamental questions persist, and water quality in the Xin'anjiang River and Qiandao Lake remains at risk [31,52]. Here a modeling approach scaled to the entire QLB was employed to understand the spatiotemporal dynamics of TN, a key polluting nutrient in both hilly [44,49,53,54] and lowland basins [55,56]. We establish that TN from point sources, which is almost equally attributable to livestock production and domestic sewage, accounts for nearly half of the aggregate TN loads in the basin. Both of these point sources are longstanding targets for mitigation under the PES plan [51]. On the other hand, seasonal surface and lateral flows from non-point sources account for the majority of TN loads in the QLB, with crop, tea, and orchard lands together comprising nearly 40% of the total. The yield of TN from tea land exceeds that from other crop types, and although crop production is declining in the area [57,58], tea cultivation is rapidly expanding, both in the QLB and in other nearby basins with similar topography [27,31,59,60]. This is important because agricultural fertilizers are commonly applied in excessive amounts to tea and other intensively managed crops in the basin, especially the upper basin in Anhui Province [59][60][61][62][63].
In general, incremental contributions of TN from non-point sources in the lower basin rise modestly and proportionately to one another, but disproportionately high contributions are evident from forest land and orchards in Zhejiang Province [31,59]. The dynamics differ, however, because forest land contributes high TN loads but low TN yields, but the situation is reversed with orchard land, which contributes low TN loads but high TN yields. The significant degree of forest coverage in the QLB contributes to high TN loads [27], but slope and erosion are also major factors. The mean slope of the study area is 24.9°, and the area proportions of slopes < 10°, 10°-20°, 20°-25°, 25°-30°, and >30° are 22%, 13%, 9%, 12%, and 44%, respectively. Sediment loss from such highly erodible forested slopes in the QLB is known to be substantial [44], and the importance of these sediments in the release of TN is documented [57]. In contrast, orchards are currently a minor but growing land use category in the QLB. Our data underscore the need for further study of their role in nutrient pollution in the basin.
Not all sub-basins in a given catchment contribute equally to nutrient loads, even if they are similar [55,56,64]. Nutrient runoff is known to be sensitive to mountainous topography and abundant annual rainfall in eastern China [44], and the differential influence of steeper and flatter lands on seasonal variation of nutrient loads is well documented [55,56]. We found that sub-basin to sub-basin heterogeneity in TN loads and yields was pronounced across the QLB, where four clustered subbasins located upstream in Anhui Province stand out as a hotspot, because of their extensive and intensive contributions to TN pollution. These adjacent sub-basins deliver high annual TN loads

Spatio-Temporal Dynamics of TN Transport
The QLB is characterized by hilly topography, diverse, rapidly intensifying agricultural production, and several urban areas where domestic and industrial activities are concentrated [25,31]. It is no wonder, then, that nutrient pollution in the basin is governed by dynamic upstream to downstream transport processes that are spatially and temporally heterogeneous and poorly understood [24,49]. It has been known for several decades that these factors have seriously compromised water quality in the basin [27]. China's first transboundary Payment for Ecosystem Services (PES) plan was consequently established in the QLB in 2012 [50,51]. Although this scheme has generated significant investment to reduce nutrient pollution entering the lower basin from Anhui Province, many fundamental questions persist, and water quality in the Xin'anjiang River and Qiandao Lake remains at risk [31,52].
Here a modeling approach scaled to the entire QLB was employed to understand the spatiotemporal dynamics of TN, a key polluting nutrient in both hilly [44,49,53,54] and lowland basins [55,56]. We establish that TN from point sources, which is almost equally attributable to livestock production and domestic sewage, accounts for nearly half of the aggregate TN loads in the basin. Both of these point sources are longstanding targets for mitigation under the PES plan [51]. On the other hand, seasonal surface and lateral flows from non-point sources account for the majority of TN loads in the QLB, with crop, tea, and orchard lands together comprising nearly 40% of the total. The yield of TN from tea land exceeds that from other crop types, and although crop production is declining in the area [57,58], tea cultivation is rapidly expanding, both in the QLB and in other nearby basins with similar topography [27,31,59,60]. This is important because agricultural fertilizers are commonly applied in excessive amounts to tea and other intensively managed crops in the basin, especially the upper basin in Anhui Province [59][60][61][62][63].
In general, incremental contributions of TN from non-point sources in the lower basin rise modestly and proportionately to one another, but disproportionately high contributions are evident from forest land and orchards in Zhejiang Province [31,59]. The dynamics differ, however, because forest land contributes high TN loads but low TN yields, but the situation is reversed with orchard land, which contributes low TN loads but high TN yields. The significant degree of forest coverage in the QLB contributes to high TN loads [27], but slope and erosion are also major factors. The mean slope of the study area is 24.9 • , and the area proportions of slopes < 10 • , 10 • -20 • , 20 • -25 • , 25 • -30 • , and >30 • are 22%, 13%, 9%, 12%, and 44%, respectively. Sediment loss from such highly erodible forested slopes in the QLB is known to be substantial [44], and the importance of these sediments in the release of TN is documented [57]. In contrast, orchards are currently a minor but growing land use category in the QLB. Our data underscore the need for further study of their role in nutrient pollution in the basin.
Not all sub-basins in a given catchment contribute equally to nutrient loads, even if they are similar [55,56,64]. Nutrient runoff is known to be sensitive to mountainous topography and abundant annual rainfall in eastern China [44], and the differential influence of steeper and flatter lands on seasonal variation of nutrient loads is well documented [55,56]. We found that sub-basin to sub-basin heterogeneity in TN loads and yields was pronounced across the QLB, where four clustered sub-basins located upstream in Anhui Province stand out as a hotspot, because of their extensive and intensive contributions to TN pollution. These adjacent sub-basins deliver high annual TN loads ranging from 423 to 576 t and correspondingly high TN yields that can be as high as 5.3 t/km 2 /yr. Similar TN loads characterize a few other areas of the QLB, including the large sub-basin surrounding Qiandao Lake, where the annual contribution is 464 t. TN yields in these areas are nevertheless modest (just 0.4 t/km 2 /yr in the case of the sub-basin surrounding Qiandao Lake).
There are two important practical reasons to assign significance to pollution hotspots such as those identified here in the QLB. First, hotspots represent priority areas for detailed analysis of factors that contribute to TN pollution, e.g., soils, land use patterns (including legacy effects), agronomic management procedures, and topographical features such as hillslopes and streams that affect movement of nutrients [65][66][67][68][69][70][71]. Second, hotspots are locations where efforts to control pollution are likely to be most efficient and have the greatest impacts [31,59]. Knowledge of their location and characteristics is especially relevant in the QLB, where significant ongoing investments are underway to reduce transport of TN into Qiandao Lake [72,73].
Although pathways of nutrient transport have been examined in the QLB [10,27] and other hilly areas of eastern China [74][75][76], the proportion of pollution contamination from surface flow, sub-surface flow, and groundwater is difficult to monitor in field experiments [77]. Physical-based modeling with multivariate correlated data as employed here is consequently useful, because it can spatially resolve the relative contributions of TN transport processes. These dynamics are illustrated with the Xin'anjiang River, where, for example, modeling revealed that TN loads from nonpoint sources doubled over a distance of less than 10 km (between locations 146.2 and 154.6 km, see Figure 8). Although groundwater was a relatively significant transport component at this location, surface and lateral flow assumed increasing importance as the river flowed downstream, such that the overall significance of groundwater was eventually marginalized.
The relative contributions of lateral and surface flow to TN also changed as the river flowed downstream, i.e., the incremental contribution of lateral flow rose continuously after the river passed beyond location 154.6 km, but that of surface flow nearly ceased before the river entered Zhejiang Province at location 243.7 km. Thereafter, almost all incremental increase in TN was due to lateral flow, which would be expected to be significant in this area, which is characterized by dense agriculture and disturbed soils capable of enhancing such flows. The above insights from basin-wide analysis underscore the importance of this perspective, which can amplify more detailed information obtained from narrowly focused studies of sub-basins [27]. Figure 10 shows the integrated N cycle for the QLB as simulated by the SWAT model used here. Net N intake is assigned to three non-point sources (in decreasing order of magnitude: fertilizer, N stored in the soil, and atmospheric deposition) and three point sources (in decreasing order of magnitude: domestic sewage, livestock production, and industrial sewage). The most significant N outlet in the terrestrial area consists of processes associated with plant growth, e.g., N removal via harvest of agricultural crops, but processes such as ammonia volatilization, denitrification, and the return of N from plant litter to the soil are also important [78]. Thirteen percent of N from non-point sources in the basin is discharged into water bodies ( Figure 10). This amount exceeds that discharged from point sources in the QLB and on a percentage basis, it is much greater than the 1.8%-4.5% determined earlier in a nearby low hilly basin with humid climate [79]. Entry of N from non-point sources thus represents a major environmental concern, even though nearly one-quarter of the N is removed from the basin's waterways during transport.

The Basin's N Cycle and Implications for Pollution Management
Water 2020, 12, x FOR PEER REVIEW 14 of 20 construction of waste water treatment plants, collection of livestock sewage, and restriction of agricultural development [83]. On the basis of SWAT modeling, a strategy which has previously been employed to identify ways to reduce pollutant loads in other regions [84,85], we suggest a shift from controlling N loads to a targeted and comprehensive strategy emphasizing N distribution and transport processes. Our SWAT analysis draws attention to two key factors for nutrient management in the QLB. The first is temporal and indicates that processes occurring during the period before June, especially those that influence subsurface flows in March, are crucial for annual N output. The second is spatial and emphasizes the importance of hotspots in the basin that contribute disproportionately to N outputs. These factors underscore the importance of restriction of agriculture on steep slopes-especially intensive agriculture that often relies on heavy applications of fertilizers in the spring. They also highlight the need for effective sewage collection-especially strategies that minimize N in tail water discharge, a goal that has proved elusive in the past [2,27,44].
Lakeside buffers [86,87], maintenance of high fractional vegetation cover [88,89], and construction of ponds to slow water movement and intercept nutrients [60] during the rainy season would undoubtedly reduce N loss in the QLB. The health of the basin's riparian and aquatic ecosystem could be further improved by adopting soil and water conservation strategies such as the restoration of vegetation and construction of wetlands. A third option involves the establishment of an ecological interception system to disrupt the N delivery pathway from subsurface flow to the lake and the basin's major waterways. Measures such as vegetative filter strips and ecological interception ditches would exploit chemical and biological mechanisms of self-purification and N retention to promote effective N reduction in the ecosystem [31,59,90].

Conclusions
Complex topographical conditions, varying climatic factors, and intensive anthropogenic activities exacerbate N pollution in hilly basins with monsoon climate. Understanding of source apportionment and spatiotemporal dynamics of N distribution and transportation is consequently Key parameters of streamflow and N transport (Table 1), as well as the relatively small Manning 'n' value, the percolation coefficient, and transmission loss to the deep aquifer (Table 2), highlight the significance of subsurface flow in the QLB. Such flow is facilitated by shallow soils and frequent heavy rainfall events in basins with steep topography and convex hillsides [80], and it represents the main N delivery pathway and thus the principal target for reducing N pollution in the QLB. Consistent with the view that agricultural land and associated N fertilizer inputs are key factors for protecting aquatic environments [10,27,81,82], the control of N pollution in the QLB and similar basins currently emphasizes reduced fertilization, application of controlled-release fertilizers, construction of waste water treatment plants, collection of livestock sewage, and restriction of agricultural development [83]. On the basis of SWAT modeling, a strategy which has previously been employed to identify ways to reduce pollutant loads in other regions [84,85], we suggest a shift from controlling N loads to a targeted and comprehensive strategy emphasizing N distribution and transport processes.
Our SWAT analysis draws attention to two key factors for nutrient management in the QLB. The first is temporal and indicates that processes occurring during the period before June, especially those that influence subsurface flows in March, are crucial for annual N output. The second is spatial and emphasizes the importance of hotspots in the basin that contribute disproportionately to N outputs. These factors underscore the importance of restriction of agriculture on steep slopes-especially intensive agriculture that often relies on heavy applications of fertilizers in the spring. They also highlight the need for effective sewage collection-especially strategies that minimize N in tail water discharge, a goal that has proved elusive in the past [2,27,44].
Lakeside buffers [86,87], maintenance of high fractional vegetation cover [88,89], and construction of ponds to slow water movement and intercept nutrients [60] during the rainy season would undoubtedly reduce N loss in the QLB. The health of the basin's riparian and aquatic ecosystem could be further improved by adopting soil and water conservation strategies such as the restoration of vegetation and construction of wetlands. A third option involves the establishment of an ecological interception system to disrupt the N delivery pathway from subsurface flow to the lake and the basin's major waterways. Measures such as vegetative filter strips and ecological interception ditches would exploit chemical and biological mechanisms of self-purification and N retention to promote effective N reduction in the ecosystem [31,59,90].

Conclusions
Complex topographical conditions, varying climatic factors, and intensive anthropogenic activities exacerbate N pollution in hilly basins with monsoon climate. Understanding of source apportionment and spatiotemporal dynamics of N distribution and transportation is consequently vital for formulating reasonable strategies to ensure comprehensive water quality management in these basins. The physical-based SWAT model was utilized in this study to simulate the drivers and transport process of N, so that effective nutrient management of the QLB can be facilitated. Three main conclusions can be drawn.
First, optimization of localized model parameters suitable for hilly areas with monsoon climate enables the SWAT model to provide reliable estimates of N loads in the basin. The model confirms that, owing to the shallow soil layer and intensive rainfall, N loss from subsurface flow in the period before rainy season is the main driver of the N transport process. Complex terrain and severe rainfall heterogeneity enhance the complexity of the problem.
Second, non-point sources are the major contributors to N loads across the QLB, with cropland, tea land, and the basin's extensive tracts of highly erodible forest land accounting for 18%, 15%, and 10% of the TN loads, respectively. Major point sources include domestic sewage (21%) and livestock production (20%). The basin is characterized by the existence of hotspot sub-basins of high TN yield, non-uniform loading of TN along the reach of the Xin'anjiang River, and spatially determined variability in the relative contributions of different transport processes to TN movement.
Third, physical-based modeling suggests that N pollution control in the QLB can be made more comprehensive and efficient. Upstream to downstream analysis of the N transport pathway highlights the need not only to construct more wastewater treatment plants in areas identified here as sites of high domestic and industrial pollution, but also to increase their operational efficiency. In addition, soil and water conservation practices should be targeted to areas identified here with intensive agriculture and high TN yields. Insights from this study thus are likely to aid regional management of the QLB and provide a framework for similar monsoon basins elsewhere.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4441/12/4/1075/s1. Figure S1: Land use classes used for SWAT analysis of the Qiandao Lake basin, Figure S2: The soil types map of the Qiandao Lake basin that was generated for use in SWAT analysis, Figure S3: Temporal and spatial heterogeneity of precipitation as recorded by individual rainfall stations (dots) compared to that recorded at two meteorological stations (dashed lines), Table S1: Fertilization and other agricultural management operations, Table S2