Introduction

Nutrient inputs have dramatically increased in recent decades, and have degraded water quality in many rivers, lakes, and oceans worldwide (Nixon 1995; Schindler and Vallentyne 2008; Vitousek and others 1997; Vollenweider 1982). Eutrophication caused by surplus nitrogen (N) and phosphorus (P) has thus become one of the most rapidly growing environmental crises of surface waters. China faces the same pollution challenges as many other parts of the world, and inland lakes are the most severely affected aquatic systems. The Tai Lake Basin in the Yangtze River Delta, which has witnessed dramatic eutrophication over the past three decades, is widely regarded as a mandatory target of water pollution control (Liu and Diamond 2005; Qin and others 2007; Yang 1996). Excessive anthropogenic nitrogen and phosphorus inputs have been implicated in the rapid rate of eutrophication of Tai Lake (Paerl and others 2011; Xu and others 2010). Nutrient discharges come from various point and nonpoint sources in most watersheds (Kaushal and others 2011). Distinct sources such as industrial or municipal wastewater discharge and runoff from farms or arable land have contributed to nutrient enrichment in the basin (Paerl and others 2011; Qin and others 2007, 2010; Xu and others 2010).

Since the early 1990s, water pollution control has been prioritized to protect the ecosystem of the Tai Lake Basin (Ellis and Wang 1997; Zhang and others 2008). During the 9th Five-Year Plan period (1995–2000), the central government and local authorities invested CNY 10 billion for environmental protection of the basin. Numerous measures and policies have continually increased pollution control of point sources, but concerning the mitigation threshold, the effects appear to have been limited (Chen and others 2003; JSEPD 2009). Water in the basin is still nutrient enriched, and the concentration of nitrogen alone has increased by more than 50 % in the past 10 years (Fig. 1). Owing to this fact, attention has gradually shifted to nonpoint pollution from agriculture and other rural activities. To improve water quality and relieve eutrophication in the entire watershed, detailed understanding of nutrient loads and sources is essential. Therefore, a systematic modeling tool capable of depicting the entire picture of basin nutrient release should be proposed (Kennedy and others 2007; Liu and Qiu 2007; Puckett 1994).

Fig. 1
figure 1

Nutrients concentration in Tai Lake during 1992–2008 (based on monthly values). Adapted from Qin and others (2006), Zhang and others (2008)

Full size image

Excessive nutrient surplus in the Tai Lake Basin comes from various sources, among which industrial point source discharge can be easily quantified via monitoring or statistical sources, making nonpoint pollution estimation the primary challenge in assessing nutrient input. Various models and methods have been developed to tackle this mission. White box physical hydrologic models, such as ANSWERS (Beasley and others 1980), SWAT (Arnold and others 1998), AGNPS (Young and others 1989), INCA (Whitehead and others 1998) and HSPF (Donigian and others 1995) all provide reliable results after long-term continuous modeling. Unfortunately, strict input requirements significantly prevent their widespread application, especially for data scarce watersheds like the Tai Lake Basin (Parsons and others 2001; Zhao and others 2011).

Since extensive nutrient sampling and modeling expertise are often absent or unattainable, simple alternative methods like the export coefficient (Johnes 1996) or the hydrograph separation method (Sloto and Crouse 1996) remain attractive options. These methods describe most pollution processes and yield credible results (Ding and others 2010). In China, researchers have also developed applicable native models (Guo and others 2004; Hong and others 2010; Li and others 2009), most of which are constrained to specific areas and lack extensive testing (Shen and others 2011). For the Tai Lake Basin, there are also regional estimation models that show certain accuracy and potential (Lai and others 2006; Zhang and others 2010; Zhao and others 2011). However, because of data scarcity and difficulty of source identification, a more practical accounting method that requires fewer data and achieves acceptable accuracy is needed for watershed management.

In the context of this study, a nutrient load estimation framework for the Tai Lake Basin covering five major pollution sources was developed. This framework is presented in the following sections. Nitrogen and phosphorus release in 2008 was then calculated. In addition, considering mitigation potential for major polluters, selected best management practices (BMPs) were analyzed in a cost-effectiveness evaluation. The rest of the paper is organized as follows. The “Methodology” section describes the nutrient load estimation method and data sources, followed by empirical results and discussion in the “Results and Discussion” section. The “Conclusions” section offers major conclusions.

Methodology

Analytical Framework

Nutrient sources and flows in a lake basin are shown in Fig. 2. A portion of nutrient load flows from sources directly into surface waters, and the remaining sources discharge pollutants after treatment. In contrast to unit-area models like the nutrient load model of the PAMOLARE package (Zhang and Jørgensen 2005) or the CMSS system (Broad and Corkrey 2011), diffuse nutrient surplus estimates here assume that nutrient exports depend on pollution sectors, and each sector may have various sources, leading to nutrient generation and losses. However, the hydrology factor impact and nutrient release of fishponds are not explicitly addressed in this paper.

Fig. 2
figure 2

Simplified nutrient sources and flows diagram for load release estimate of the Tai Lake Basin. In the above framework, the five pollution sectors are shown in dashed green boxes. The solid blue boxes refer to the nutrient sources in each sector, while the dashed black arrows indicate nutrient loads into surface waters. Adapted from Li and others (2010), Robertson and Vitousek (2009), and Smil (2000)

Full size image

Considering the complexity and multi-source characteristics of the nutrient cycle in an anthropogenic system, the developed model predicts total loads produced in the watershed by calculating nutrient loads generated from major pollution sectors and summing over all sectors. The five major nutrient release sectors are industrial manufacturing, livestock breeding, crop agriculture, household consumption and atmospheric deposition. Classes of sectors and sources are presented in Table 1. Moreover, aiming to directly control nutrient loss contributors, applicable BMPs were selected for further discussion. The BMP referring to the atmospheric deposition sector is excluded, because it is difficult to directly reduce nutrient inputs from deposition using technical and managerial measures (Yang and others 2007). By way of a quantified cost-effectiveness analysis, the performance of identified BMPs were evaluated, the results of which may provide useful information to support policy making.

Table 1 Detailed classification of pollution sectors and sources
Full size table

Because of differences in discharge pathway, the livestock breeding sector is divided into industrial and family-scale farms. The latter farms refer to husbandry work within the residential boundary of rural families, and solid waste and wastewater discharge are two different pathways of urine and manure loss from livestock farms. Additionally, household solid waste primarily refers to garbage disposed in urban and rural daily life, which might run off into surface waters without treatment from landfills in rural areas or even city outskirts. For existing or planned pollutant treatment facilities, removal efficiency was included in the accounting process. Considering different conditions of solid waste and sewage treatment infrastructures, urban and rural residents are discussed separately.

Study Area

The Tai Lake Basin of Jiangsu Province (colored regions in Fig. 3) includes three major cities (Suzhou, Wuxi and Changzhou) and covers approximately 19,399 km2. The basin is downstream of the Yangtze River in Eastern China. It is one of the most industrialized areas in China, and the high residential density promotes social activity. The value added by agricultural production accounts for a moderate share of regional GDP. High nutrient runoff rates from cropland and farms are closely related to total nutrient loads. Atmospheric deposition has been shown to be a significant nitrogen contributor, because of widespread coal-fired power plants and chemical industries. These characteristics fully represent the five major pollution sectors in the basin. Most importantly, the industrial manufacturing, livestock breeding, crop agriculture, and household consumption sectors are found in the above three cities. Atmospheric nutrient deposition is estimated based on the area of the basin and its 15 inflow rivers.

Fig. 3
figure 3

Location and surroundings of the Tai Lake Basin in Jiangsu province, China. In the above figure, black dots represents main point sources (top 750 regarding nutrients discharge) within the study Tai Lake Basin boundary, which include industrial factories and WWTPs

Full size image

Accounting Approach

The nutrient load accounting methods are presented sequentially. In the industrial manufacturing sectors (NSe), only effluent discharge is considered because the nutrient content of industrial solid waste and other types of waste tends to vary greatly in terms of products (Keller and others 1997). Nutrient release through effluent discharge is calculated based on data from a regional industrial census in 2008. This census covered all large enterprises within the study region, which account for almost 90 % of total regional industrial wastewater discharge (JSPSCO 2008; JSSB 2010). Nutrient loads from industrial effluent can be described by the following equation:

$$ {\text{NS}}_{\text{e(N,P)}} = L_{{ ( {\text{N,P)}}}} \times \left( {1 - {\text{RE}}_{{ ( {\text{N,P)}}}} } \right) $$
(1)

Here, NSe(N, P) is the nutrient load into surface waters from the industrial manufacturing sector, L is the total amount of nutrients in industrial effluent into wastewater treatment plants (WWTPs), and RE is the removal efficiency of nutrients. A modified anaerobic–anoxic–oxic (A2/O) process was adopted for WWTPs in the study area. RE was 80.7 % for N and 91.0 % for P according to the pollution census (JSPSCO 2008) and field investigation of two plants within the basin.

Next, because pollutants from these sectors enter recipient water bodies in a diffuse manner and at intermittent intervals, nutrient release from livestock breeding (NSl), crop agriculture (NSa) and household consumption sectors (NSh) are calculated using a similar unit-based method, which is adapted from the improved export coefficient method (Ding and others 2010; Johnes 1996). This method follows a bottom-up calculation process, and builds direct links between pollution sources and nutrient release through empirical data and coefficients, as given by the following equation:

$$ {\text{NS}}_{{ ( {\text{N,P)}}}} = \sum\limits_{\text{i}} {\left[ {{\text{NC}}_{\text{i(N,P)}} \times \sum\limits_{\text{j}} {\left( {{\text{EU}}_{\text{ij(N,P)}} \times {\text{EC}}_{\text{ij(N,P)}} } \right)} } \right]} $$
(2)

Here, nutrient load of a sector (NSl, NSa, and NSh) is total release into surface waters of all relevant specific nutrient classes (NC). Specifically, for the ith nutrient class of the corresponding NS, the load is calculated based on quantity of the nutrient class (NCi), nutrient amount in the corresponding emission unit (EUij), and its emission coefficient (ECij) into surface waters. For urban sewage discharge, areas with or without a wastewater pipeline network (see Table 1) are considered separately. In the crop agriculture sector, \( {\text{EU}}_{\text{i2}}^{\text{a}} \) (organic manure) is calculated by multiplying the nutrient amount in all possible organic manure (excreta) and the ratio of organic manure applied to farmland over total excreta produced. The excreta come from both household consumption (human excreta) and livestock breeding (livestock excreta) sectors. \( {\text{EU}}_{\text{i3}}^{\text{a}} \) (crop residue) is calculated by multiplying the amount of crop residues applied to farmland and the nutrient content of crop residues. Vegetables are excluded since their residue amount is too small to be considered (Li and others 2010; Smil 2000). In the household consumption sector, because the centralized treatment rate varies significantly by city, values of the three major cities are discussed separately. For urban areas not covered by a sewage pipeline network, nutrient release is included in “rural household consumption” instead of “urban household consumption.”

Atmospheric deposition into surface waters also contributes a significant share of watershed nutrient loads, especially in terms of nitrogen (Yang and others 2007; Zhai and others 2009). However, aquatic systems have a pollutant reduction effect, such as denitrification. Only deposition into Tai Lake and its 15 inflow rivers was considered; deposition onto other land use types (such as forest) was excluded. Nutrient loads can be calculated using the following equation:

$$ {\text{NS}}_{\text{d(N,P)}} = {\text{SA}} \times {\text{AD}}_{{ ( {\text{N,P)}}}} $$
(3)

Here, nutrient input via atmospheric deposition (NSd) is the total nitrogen or phosphorus load from both wet and dry deposition, which is calculated by multiplying total aquatic surface areas of Tai Lake and its 15 inflow rivers (SA) by the annual atmospheric nutrient deposition rate (AD) in the region.

Load accounting units, such as the multiple nutrient classes and emission units in the unit-based method or effluent volume in the industrial manufacturing sector, are presented in Table 2. Importantly, if treatment facilities exist or are planned (such as municipal wastewater treatment and manure recycling of industrial livestock farms), a centralized treatment rate and nutrient removal efficiency of adopted processes should also be considered.

Table 2 Accounting classes and units of the nutrient loads from the five major pollution sectors
Full size table

In summary, total nutrient loads can be calculated by summing the loads of all five sectors in the study region, as given by the following:

$$ {\text{TNL}}{}_{{ ( {\text{N,P)}}}} = {\text{NS}}_{\text{e(N,P)}} + {\text{NS}}_{\text{l(N,P)}} + {\text{NS}}_{\text{a(N,P)}} + {\text{NS}}_{\text{h(N,P)}} + {\text{NS}}_{\text{d(N,P)}} $$
(4)

Here, total nutrient load into surface waters (TNL) is the sum of nitrogen or phosphorus loss from the industrial manufacturing (NSe), livestock breeding (NSl), crop agriculture (NSa), household consumption (NSh) and atmospheric deposition (NSd) sectors.

For surplus nutrient reduction, managerial measures and engineering facilities, also known as BMPs (Cherry and others 2008; Robertson and Vitousek 2009), have been widely discussed. Using a quantified cost-effectiveness analysis, we evaluate the performance of BMPs targeting the pollution sectors. Typically, a BMP is designed to change nutrient discharge or runoff rates into surface waters. This is done by way of improving the removal efficiency of WWTPs, building retention facilities for runoff, or even modifying the deposition rate from the atmosphere. Specifically, the reduction potential [RP(N,P)] of a potential BMP for each sector can be calculated by the following equations separately:

$$ {\text{RP}}_{\text{e(N,P)}} = {\text{L}}_{{ ( {\text{N,P)}}}} \times \left| {\Updelta {\text{RE}}_{{ ( {\text{N,P)}}}} } \right| $$
(5)
$$ {\text{RP}}_{\text{l(a,h)(N,P)}} = \sum\limits_{\text{i}} {\left[ {{\text{NC}}_{\text{i(N,P)}} \times \sum\limits_{\text{j}} {\left( {{\text{EU}}_{\text{j(N,P)}} \times \left| {\Updelta {\text{EC}}_{\text{j(N,P)}} } \right|} \right)} } \right]} $$
(6)
$$ {\text{RP}}_{\text{d(N,P)}} = {\text{SA}} \times \left| {\Updelta {\text{AD}}} \right| $$
(7)

Here, nutrient reduction potential of a BMP [RPe(N,P), RPl(N,P), RPa(N,P), RPh(N,P), or RPd(N,P)) refers to the maximum amount of reduced nutrient load, and \( \Updelta \) is the result of subtracting corresponding coefficients with and without a BMP. Since this might be positive or negative value in different situations, the absolute value is used during accounting processes.

As another crucial indicator of a BMP, the average mitigation cost (AC), should cover all expenditures on construction, operation, maintenance, and management. Here, we directly obtained the AC of BMPs from governmental documents, technical guides, interview data, or literature sources.

Data Collection

Statistical data were largely obtained from regional or local government documents, including statistical yearbooks, industrial statistics, census data, and technical guides. For example, regional effluent nutrient load into surface water was from the pollution source census (JSPSCO 2008). Centralized wastewater treatment rates in urban areas (88.9, 77.4 and 86.2 % for Suzhou, Wuxi and Changzhou, respectively), amounts of NCs of each sector, and aquatic surface areas of Tai Lake and its inflow rivers are found in local statistical yearbooks (CZSB 2009; JSSB 2010; SZSB 2009; WXSB 2009).

As another category of key parameters, emission units have been set to correspond to the emission coefficient (ECs) and efficiency of treatment facilities. Literature reviews, along with expert knowledge and field surveys, are the main sources for coefficients and parameters. For example, annual atmospheric nutrient deposition rates (summing those of dry and wet deposition) were 2,763 kg N km−2 and 70 kg P km−2, adapted from Yang and others (2007) and Zhai and others (2009). Discharge rates of livestock farm excreta into surface waters were set to 3.6 % for family-scale farms and 2.8 % for industrial-scale farms (Chen and others 2008; Li and others 2010). Median values were assumed in cases of reported ranges or multiple sources. Detailed values and sources of data and parameters in load estimation are listed in Table 3.

Table 3 Values and sources of data and parameters in the nutrient load accounting process
Full size table

Interview data was also used as a supplementary source of parameters. We conducted interviews with officials and technicians of administrative authorities. For example, according to local experienced agricultural technicians, the ratio of organic manure applied to farmland over total excreta produced by livestock excreta was set to 75 % (family-scale farms) and 50 % (industrial-scale farms), while that of human excreta was set to 100 %.

Results and Discussion

Annual Nutrient Release

Based on the framework described, nutrient loads of Tai Lake Basin can be systematically calculated. Figure 4 depicts nutrient loads into surface waters and their composition in 2008. Total nutrient loads into aquatic systems in 2008 were estimated at 33043.2 tons N a−1 and 5254.4 tons P a−1. Annual area-specific nutrient loads, which are defined as loads per watershed area, were 1.94 tons N km−2 and 0.31 tons P km−2, respectively. The values agree well with those of Lai and others (2006), Chen and others (2003), and Geng and others (2005), who adopted modeling or monitoring methods for load estimation. Furthermore, household consumption was found to be the major sector with greatest impact on surface waters, contributing 46 and 47 % to the nitrogen and phosphorus loads, respectively. Atmospheric deposition (6246.1 tons a−1 or 18 %) and crop agriculture (5045.6 tons a−1 or 15 %) sectors also represented significant proportions of the nitrogen load, whereas the livestock breeding sector was responsible for the second largest phosphorus load (1683.2 tons a−1 or 32 % in P load). With implementation of the Total Emission Control Policy, which required a 12 % average reduction of the major pollutants every five years after 1996, the industrial manufacturing sector gradually became a less dominant contributor, with less than 10 % proportions of both total N and P release.

Fig. 4
figure 4

Nitrogen and phosphorus loads and the distribution among various sectors (tons a−1)

Full size image

For the livestock breeding and household consumption sectors, the load is directly linked with the quantity of emission units, such as livestock numbers or size of residential population. Although there were treatment facilities on most industrial-scale farms, the tremendous numbers of animals raised created much more pollution than that from family-scale farms (10 vs. 4 % in N load, 23 vs. 9 % in P load). The population distribution between urban and rural areas has a similar effect. The urban population consists primarily of residents, and there is a higher level of infrastructure for waste treatment. As a result, urban and rural household consumption contributed nearly the same amount of nutrients.

Source Identification

A composition analysis of nutrient release further highlights the influence of various sources or activities in Tai Lake Basin. Figure 5 shows that nitrogen and phosphorus release generally shared a similar distribution trend among various sources. The results emphasize that household wastewater discharge (excreta included) was the most significant contributor of nutrient loads in the region (43 % in N load and 42 % in P load). Because human excreta were treated together with urban sewage, the large population size and resultant high nutrient content in the sewage unsurprisingly led to this result. However, because of a low rate of runoff into water bodies, solid waste disposal was not as important as sewage discharge, and accounted for only 4 % in N load and 5 % in P load.

Fig. 5
figure 5

Composition of nitrogen and phosphorus loads among various sources (tons a−1)

Full size image

Animal excreta were the second largest nutrient source, especially for phosphorus runoff into surface waters (1683.2 tons a−1 or 32 % in total P load). Industrial and family-scale farms had nearly equal contributions, although the former accounted for slightly more because of a greater number of livestock. In the crop agriculture sector, chemical fertilizer application was responsible for 10 and 9 % of total nitrogen and phosphorus release, respectively, exerting a much greater impact on surface waters than applied organic manure and other additives. Because of tough emission standards in the region, industrial effluent contributed approximately 8 % to total N and P load. However, dry and wet atmospheric deposition jointly contributed significant nutrient inputs, especially for nitrogen, which was 6246.1 tons a−1 or 18 % in total.

Comparison of Results

The findings agree with other studies of Tai Lake Basin (Liu and others 2004; Qin and others 2007), as well as studies in Europe and the United States (Carpenter and others 1998; Gunes 2008). Urban household consumption and industrial-scale livestock breeding sectors were treated as point sources, whereas only industrial discharge was considered in other studies. As shown in Table 4, accounting results agree with this classification since the share of nonpoint sources appeared smaller, for phosphorus in particular (42 % in the total). For nutrient load intensities, defined as loads per watershed area, the results conformed to the estimate of Lai and others (2006) for the entire Tai Lake Basin. Compared with the 3.24 t km−2 of Egirdir Lake Basin in Turkey (Gunes 2008), the estimated nitrogen intensity of 1.94 t km−2 was dramatically lower. These inconsistences might be explained by large temporal span, climate variation, and differences of hydrographic conditions, landscape composition, and leading industries.

Table 4 Comparison of accounting results with reported previous studies (annual values)
Full size table

Urban and rural households jointly contributed 46 % in N load and 47 % in P load, which approach values from previous studies of the same basin (Qin and others 2007; Zhang and others 2008). Phosphorus release per capita was slightly higher than that observed in other studies. This was most likely because we considered not only household sewage (excreta included) but also solid waste emissions in the accounting process.

Meanwhile, the crop agriculture sector accounted for 15 % of N loss and 10 % of P loss, lower than the average value of 30 % in other basins (Drolc and Koncan 2002; Gunes 2008; Jaworski and others 1992). The relatively small contribution of agricultural production (both farming and breeding) in the regional economy most likely explains this difference (CZSB 2009; SZSB 2009; WXSB 2009), and might also be responsible for a lower area-specific nutrient release from farmland than the China average. While the livestock breeding sector contributed only 14 % of total nitrogen load, the same as that of the crop agriculture sector, this was also much lower than the results of previous studies (Chen and others 2010; Zhao and others 2011).

Performance of BMPs

All six BMPs are targeted on a nutrient contributor and should be directly effective. Detailed BMP information and changed values of parameters and results of RP and AC are listed in Table 5. As shown in Fig. 6, there was no BMP that satisfactorily balanced cost and reduction effect. Overall, biogas digester construction for industrial-scale farms reduced the greatest amount of N (1633.5 tons a−1 or 4.9 %) and P (607.5 tons a−1 or 11.6 %) at a relatively acceptable average cost (1.11 × 104 CNY/ton N a−1 and 5.91 × 104 CNY/ton P a−1), thus making it a very satisfying alternative for the basin. On the other hand, though construction of rural wastewater treatment networks has the largest nitrogen reduction potential, the high average cost prevented its widespread application.

Table 5 Description, coefficient values and performance of BMPs
Full size table
Fig. 6
figure 6

Comparison of nutrients reduction performance of BMPs (annual values)

Full size image

Building decentralized rural wastewater treatment facilities had certain mitigation effects (particularly for N), and the low financial requirement favors this when cost is the constraining factor. When facing strict budget pressures, this approach might be preferred owing to the moderate average cost (0.72 × 104 CNY/ton N a−1 and 4.80 × 104 CNY/ton P a−1). Nonetheless, comparatively weak phosphorus mitigation performance may minimize the total effect, particularly when nitrogen is not the limiting nutrient. Moreover, as a widely implemented BMP, limiting the use of chemical fertilizer was overshadowed because of low mitigation capacity, especially for N (878.3 tons a−1).

On the other hand, some already well-managed nutrient sectors or sources might fall short in nutrient reduction potential. Although enhancing industrial effluent treatment had a low monetary requirement for process innovation, deficiency of its reduction potential made this option not very favorable (259.3 tons a−1 or 0.8 % for N and 89.9 tons a−1 or 1.7 % for P) for watershed nutrient control. The same situation also applies to the BMP of improving the centralized urban wastewater treatment rate in city outskirts. This approach reduced nutrients less effectively than rural treatment network construction, but retained a monetary requirement equal to the average mitigation cost (1.23 × 104 CNY/ton N a−1 and 7.02 × 104 CNY/ton P a−1).

Overall, BMPs with high reduction potential most likely require large monetary investment. Construction of rural treatment networks is an example. However, limited expenditure cannot usually ensure sufficient mitigation effect. However, in cases where phosphorus pressure was minor, building rural decentralized wastewater treatment facilities could be a good choice. Since BMPs mitigate nitrogen and phosphorus with different removal efficiencies, it is important to choose the more pressing indicator of the two eutrophication factors. Considering that point sources in Tai Lake Basin have been somewhat well controlled in recent years, typical diffuse sources appear to have more reduction potential than industrial effluents. This should attract more attention in watershed management.

Conclusions

We proposed an empirical accounting framework to estimate nitrogen and phosphorus loads from five major sectors in Tai Lake Basin. Total nutrient loads were estimated at 33043.2 tons N a−1 and 5254.4 tons P a−1 in 2008, and annual area-specific nutrient loads were 1.94 tons N km−2 and 0.31 tons P km−2. Among the five major sources addressed, the household consumption sector was found to be the major contributor with greatest impact on surface water (46 % in N load and 47 % in P load), whereas household wastewater discharge was the major emission source. Atmospheric deposition and animal excreta loss from livestock farms also contributed a significant share of nitrogen and phosphorus, respectively. Our accounting method uses easily accessible data largely obtained from statistical databases and open publications, and provides a complete picture of nutrient pollution at the watershed level. It may be used to support policy making, which promotes its wide application to nutrient load estimation worldwide.

Considering reduction potential and average monetary cost, six BMPs implemented or under design in Jiangsu Province were selected for evaluation, and targeting proposals were discussed. Overall, biogas digester construction on industrial-scale farms was proven the best alternative for the Tai Lake Basin, whereas the building of rural decentralized wastewater treatment facilities would be a good choice under tight budget restrictions. Compromise is inevitable when facing realistic problems. During the decision-making process, reduction potential, average monetary costs, prior eutrophication indicators and other factors, such as the risk tolerance of policy makers, should all be considered.

It is possible to obtain a reasonable load estimate on a microscale (e.g., onsite soil, field, plot and farm). However, this cannot realistically be achieved on a macroscale such as in a lake basin (Chen and others 2008), which would be subject to estimation uncertainties. One of the causes of these uncertainties is exclusion of minor nutrient sources such as fishponds (or other aquaculture) or natural soil storage (Smil 2000), and the use of statistically median values may also have produced uncertainty in the nutrient load estimation. Moreover, since this work mainly addressed load estimation and source identification, it does not explicitly account for the effects of hydrology, and average BMP costs were mostly obtained directly instead of calculated. These are important issues to tackle in future work.