Alarming nutrient pollution of Chinese rivers as a result of agricultural transitions

Our study is based on two models: Global NEWS-2 (Nutrient Export from WaterSheds) [16, 17] and NUtrient flows in Food chains, Environment and Resources use (NUFER) [18]. Global NEWS-2 is a spatially explicit model that has been applied to analyze nutrient-related problems worldwide [19–30] including China [17, 31–33]. This model quantifies river export of different nutrients (nitrogen, phosphorus, carbon, silica) in different forms (dissolved inorganic, dissolved organic, particulate) as a function of human activities on land (e.g., agriculture, sewage) and basin characteristics (e.g., hydrology, land use). This basin-scale model has been used to analyze past (1970, 2000) and future (2030, 2050) trends [16, 34]. NUFER was developed for China to quantify efficiencies of nutrients in the food chain at national [13, 35–37] and provincial [18] scales over time (e.g., the period of 1980–2010 [18, 37]). This model is also used to assess management options for efficient nutrient management in the food chain [38, 39].

We developed a sub-basin version of the Global NEWS-2 model in which we included information from NUFER (see extended materials and methods in supporting information). Novel aspects of our modeling approach include (i) the sub-basin scale at which we run Global NEWS-2 to quantify total dissolved nitrogen (TDN) and phosphorus (TDP) inputs to Chinese rivers by source, and (ii) the coupling of this sub-basin version of Global NEWS-2 with NUFER. The latter we did to include direct discharge of animal manure as a point source of nutrients in rivers in our modeling to analyze the consequences of agriculture transitions on river quality in China. This has not been done before. TDN and TDP include both dissolved inorganic and organic forms of N and P.

To assess the consequences of agricultural transition on water quality, we modeled TDN and TDP inputs to rivers for 1970 and 2000. We selected 1970 to reflect the pre-transition period and 2000 to reflect the ongoing transitions in Chinese agriculture. We also modeled TDN and TDP inputs to rivers for 2050 as an illustrative example to show the risks for future river pollution.

Our model includes point (RSpnt_F.total.j, kg, see equation (1)) and diffuse (RSdif_F.total.j, kg, see equation (2)) sources of nutrients in rivers. Point sources of the nutrients (F: TDN and TDP) include direct discharges of animal manure to rivers (RSpnt_F.ma.j, kg) and human sewage (RSpnt_F.sew.j, kg). Diffuse sources of the nutrients (F: TDN and TDP) include manure (RSdif_F.ma.j, kg) and synthetic fertilizers (RSdif_F.fe.j, kg) used in croplands, atmospheric N deposition (only for TDN; RSdif_F.dep.j, kg), biological N fixation (only for TDN; RSdif_F.fix.j, kg), leaching of organic matter (for dissolved organic N and P as a function of runoff) (RSdif_F.lch.j, kg), and P-weathering (only for TDP; RSdif_F.wth.j, kg)

$$\begin{eqnarray}&&{{\rm{RSpnt}}}_{{\rm{F}}{\rm{.total}}{\rm{.}}j}={{\rm{RSpnt}}}_{{\rm{F}}{\rm{.ma}}{\rm{.}}j}+{{\rm{RSpnt}}}_{{\rm{F}}{\rm{.sew}}{\rm{.}}j},\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&{{\rm{RSdif}}}_{{\rm{F}}{\rm{.total}}{\rm{.}}\bar{j}}={{\rm{RSdif}}}_{{\rm{F}}{\rm{.ma}}{\rm{.}}j}+{{\rm{RSdif}}}_{{\rm{F}}{\rm{.fe}}{\rm{.}}j}+{{\rm{RSdif}}}_{{\rm{F}}{\rm{.dep}}{\rm{.}}j}\\ &&\quad +{{\rm{RSdif}}}_{{\rm{F}}{\rm{.fix}}{\rm{.}}j}+{{\rm{RSdif}}}_{{\rm{F}}{\rm{.lch}}{\rm{.}}j}+{{\rm{RSdif}}}_{{\rm{F}}{\rm{.wth}}{\rm{.}}j}.\end{eqnarray} \tag{ 2 }$$

TDN (RSpnt_TDN.ma.j, kg) and TDP (RSpnt_TDP.ma.j, kg) inputs to rivers of sub-basin j from the manure point source were calculated as:

$$\begin{eqnarray}&&{{\rm{RSpnt}}}_{{\rm{TDN}}{\rm{.ma}}{\rm{.}}j}={{\rm{Nexc}}}_{j}\cdot {{\rm{frN}}}_{{\rm{sw}}{\rm{.}}j},\end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray}&&{{\rm{RSpnt}}}_{{\rm{TDP}}{\rm{.ma}}{\rm{.}}j}={{\rm{Pexc}}}_{j}\cdot {{\rm{frP}}}_{{\rm{sw}}{\rm{.}}j}.\end{eqnarray} \tag{ 4 }$$

Here Nexc_j and Pexc_j are the N and P animal excretion in each sub-basin (j) (kg), respectively. We used gridded Global NEWS-2 information: 0.5 longitude by 0.5 latitude; i.e. cell areas ranging from 2135 to 2868 km² for the study area. Gridded information includes animal manure excretion corrected for losses of N during storage and housing (see figure S7) [16]. These values were originally derived from the integrated model to assess the global environment (IMAGE) model [40]. Bouwman et al [41] prepared these inputs for Global NEWS-2. IMAGE first calculated country-based manure production based on animal stocks and excretion rates, and then spatially allocated this manure over grids using land use maps (see details in [40, 41]). We used this gridded information to calculate animal manure for sub-basins (see figure S8 for values) using ArcGIS functions. We used these values for P excretion (Pexc_j). We calculated N excretion (Nexc_j) from the available N animal manure from Global NEWS-2 while accounting for N losses to the air (these losses were derived from NUFER, see supplementary methods).

frN_sw.j and frP_sw.j are the fractions (0–1) of N and P in animal excretion that are discharged directly to surface waters (sw) of sub-basin (j), respectively. These fractions were calculated for each sub-basin from provincial data by NUFER [18] (figure S2) using ArcGIS (area-weighted averages). Thus the fraction of manure that is directly discharged to surface waters of a sub-basin is Σ_i (fraction of area covered by province i · fraction of manure directly discharged for province i). Fractions of manure discharged for provinces were taken from NUFER for 1980 and 2005 because NUFER includes those years. We assume that 1980 is close to 1970 and 2005 is close to 2000 in our study. In NUFER discharges of N and P from animal manure to surface waters were calculated from manure excretion by correcting for losses during storage and housing, and for applications to cropland and to grassland (see details in [18]). Manure excretion was calculated based on animal feed from crop and kitchen residues, animal stocks (e.g., pigs, layer and broiler poultry, milk and beef cattle, sheep and goat), and excretion rates for N and P (see 'Model performance' for sources of information in the supplementary methods). In NUFER animal production (see details in [18]) includes two intensities of animal breeding, reflecting traditional and industrial farming systems. The high intensity types of animal breeding are defined for pig systems (>50 heads), dairy cattle (>5 heads), beef cattle (>50 heads), layer poultry (>500 heads), and meat poultry (>2000 heads) (figure S10 as an example).

Nutrient inputs to rivers from sewage and diffuse sources were quantified following the modeling approaches of Global NEWS-2 [16] (summarized in the supplementary methods). Most model inputs were derived from gridded global datasets (0.5 longitude × 0.5 latitude) developed for Global NEWS-2 [41–43] for 1970, 2000 and 2050, and most of model parameters are also from this model [16] (figure S7). Inputs for 2050 were based on earlier studies [41–43] which are based on interpretations of the storylines of the millennium ecosystem assessment scenarios [34, 44]. In this study we used model inputs that reflect a global orchestration scenario assuming globalized trends towards socio-economic development and reactive management of environmental problems. This scenario was implemented in the sub-basin Global NEWS-2 model as a starting point [16, 45]. The total population is assumed to have increased slightly by 2050 because of better education (e.g., low mortality and fertility rates, and high migration). However, urban population is assumed to have increased fast. This reflects migration of people to cities to find better jobs [43]. The demand for animal production will have increased by 2050 and thus agriculture is projected to intensify for food security reasons [41]. This existing scenario assumes that all available manure will be applied on land. However, this may not be likely. Considering the development of industrial farms today and the poor manure management it is more likely that direct discharges of animal manure to rivers will occur in the coming years without efficient manure management. Thus, in this study we assumed that direct discharges of manure to surface waters will still occur in 2050. Therefore, we used for 2050 the same fractions of manure directly discharged to rivers as for 2000 (see details in materials and methods in supporting information).

Model uncertainties are associated with model inputs, parameters and approaches. The original Global NEWS-2 was validated and calibrated for world rivers [16], indicating an acceptable performance according to the Nash-Sutcliffe model efficiency (R_NSE²: 0.54, 0.51, 0.71, 0.90 for DIN, DIP, DON, DOP, respectively). Strokal et al [17] indicated a satisfactory performance of the model for DIN and DIP export by large Chinese rivers for 2000 (Pearson's coefficient of determination (${{R}_{{\rm{p}}}}^{2}):$ 0.96; ${{R}_{{\rm{NSE}}}}^{2}:$ 0.42; moldel error (ME): 18%). Yan et al [46] validated the model for DIN export by the Changjiang River for 1970–2002 (${{R}_{{\rm{p}}}}^{2}=0.93).$ Similar conclusions were drawn for application of Global NEWS-2 to other world regions (see supplementary methods for references). NUFER was developed based on validated statistical data, field surveys, literature, and has been widely applied in many studies [13, 18, 35–37]. See details in supplementary methods on evaluating the original global NEWS-2 and NUFER.

We consider our model appropriate for modeling N and P inputs to the Chinese rivers at the sub-basin scale for three main reasons (see details in supplementary methods for references). First, our modeled nutrient pollution levels in rivers are generally in line with observations. Our model captures the increasing trends in dissolved N and P inputs to the Chinese rivers since 1970, which is in agreement with other studies (e.g., [8, 48–50]). Second, the comparison of model inputs with independent county data (figure S9) convincingly shows that our model inputs for sub-basins are of good quality (e.g., ${{R}_{{\rm{P}}}}^{2}:$ 0.73–0.98 for studied rivers for N and P synthetic fertilizers and animal manure). Third, we verified relevant model parameters with experts and local information (e.g., nutrient removal during treatment in the Dongjiang basin, fractions of direct manure discharges from NUFER; see supplementary methods).

Chinese agriculture has been in transitions since the 1980s [11, 12, 51]. These transitions include changes in animal and crop production systems. Pre-1980, Chinese agriculture was dominated by small traditional farms with combined crop and animal production [12, 35, 52] (figure 1 and figure S3). Synthetic fertilizer was not widely used. Instead, farmers used most animal manure collected in confinements or from other places as a fertilizer [52]. Thus direct discharge of manure to surface waters was minimal [52].

Since the 1980s Chinese agriculture has been industrializing to ensure food security (figure 1) [12, 35, 51, 52]. Crop production gradually separated from animal production in large parts of the country. This happened when the centrally planned economy shifted to a market-oriented economy [12, 52]. The human population increased over this period (figure S4, table S1). With the increasing prosperity and urbanization, the human diet shifted towards more meat consumption. This all led to an increasing demand for agricultural products [18]. Synthetic fertilizers are now preferred over animal manure, because of their relatively low prices and low labor demand [52, 53] (figure 1).

Animal production has thus shifted from traditional to industrial, land-less farming systems (figure 1, figures S3 and S10). This allows for a higher production needed to meet the demand for meat in China, but also in the world (e.g., around half of the pork, and 18% of the poultry meat produced worldwide are from China in 2005) [11, 12, 54]. Between 1970 and 2000 the total number of pigs, poultry, cattle, sheep and goats increased 1.5- to 8-fold (range for the different animal categories) in China. By 2000 10%–40% of these animals, and by 2010 over two-thirds of the pigs and poultry were grown in industrial farms in China (figure S3) [18, 35]. Pigs and poultry production accounts for over half of the total manure produced (figure S5). Most of the produced manure is considered waste because industrial farms are largely disconnected from crop production, and often lack facilities to treat or recycle animal manure [18, 35, 52]. Treatment is considered as separation of animal manure into solid and liquid fractions. The solid fraction after composting can be transported and applied to vegetable and fruit farms. The liquid fraction is further treated and then discharged ultimately to water systems. However, the separation and treatment of the liquid fraction is not very effective, and therefore large amounts of N and P are discharged to water systems (figure 1, [52]). For example, treatment ratios for animal waste in industrial farms are reported as [52]: 3% for dairy cows, 10% for chicken and 43% for pigs. Both the treated and untreated manure from these systems are point sources rather than diffuse sources of nutrients in Chinese rivers.

Our results indicate that manure point sources have a much larger share in total TDN and TDP inputs to rivers than other nutrient sources (figures 2 and 3, figure S6). In 1970 less than 5% of the animal manure was directly discharged to rivers of the sub-basins as waste. In 2000 this had increased to 30%–70% (figure 2). As a result, the calculated basin area-weighted averaged TDN inputs to rivers from only animal manure (diffuse and point) increased 8-fold, from 156 kg km⁻² of basin area in 1970 to 1316 kg km⁻² in 2000. The TDP inputs increased 22-fold, from 11 kg km⁻² in 1970 to 243 kg km⁻² in 2000. These inputs are mainly from direct discharges of manure to rivers (point source). These manure point sources are responsible for around half of the total TDN and 80% of TDP inputs to rivers in 2000, which is much higher than in 1970 (around 6% for TDN and 37% for TDP) (table 1).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The increases in TDN and TDP inputs to rivers and the shares of manure point sources differ largely among rivers and their sub-basins (figure 3; table 1). Sources of TDN and TDP in rivers include animal manure, synthetic fertilizers, atmospheric N deposition, biological N fixation, leaching of organic matter, weathering of P-minerals and human waste. Nutrient inputs to rivers have increased by a factor 2 to 45 for sub-basins between 1970 and 2000 (figure 3). These increases are particularly large for the northern rivers, including the Huang, Huai, Hai and Liao (figure S6). For example, the total inputs of TDN to these northern rivers increased 11-fold, from 280 to 3000 kton between 1970 and 2000. TDP inputs increased 20-fold, from 27 to 546 kton (figure S6). The sub-basin areas of these rivers are characterized by low precipitation and runoff [55]. As a result, nutrient inputs from diffuse sources are typically low, and direct discharge of manure thus accounts for at least two-thirds of the TDN and over 80% of TDP inputs to these northern rivers (figure 3). The remainder is from synthetic fertilizers or sewage. Animal manure use in cropland accounts for less than 4% of the nutrients in the northern rivers (figure 3, figure S6). More than half of the total nutrients are concentrated in rivers of the downstream Huang delta, Huai and Hai basins (figure 3).

For the central (Changjiang) and southern (Zhujaing) rivers we calculate an increase in TDN inputs from 2500 to about 7000 kton (a 2.8-fold increase), and in TDP from 62 to almost 600 kton (a 9.5-fold increase) between 1970 and 2000 (figure S6). Manure point sources are responsible for 20%–60% of these TDN and 50%–97% of TDP inputs to rivers in 2000 (range for sub-basins; figure 3). The remainder is mainly from synthetic fertilizers (sub-basin range: 12%–46%), and atmospheric N deposition on land (sub-basin range: 12%–40%). Animal manure as a diffuse source contributes less than 10% to TDN and TDP inputs to rivers. Sewage effluents are important sources of both TDN and TDP inputs particularly in river deltas.

Nutrient pollution in Chinese rivers may increase in the coming years if future manure management stays as it is during the ongoing transition. Furthermore, the demand for animal products in China will remain high causing further development of industrial farms. We analyzed river pollution in 2050 as an example to illustrate the environmental consequences of future trends without improved manure management. Figure 4 illustrates to what extent TDN and TDP inputs to Chinese rivers may increase in 2050 compared to 2000 (reflect the ongoing transition). In 2050 the total inputs of TDN and TDP to Chinese rivers are calculated to be 8%–325% higher (range for sub-basins) than in 2000, except for one sub-basin in the Zhujaing River (figure 4). Manure point sources are projected to remain the main contributor to river pollution by nutrients in 2050 (22%–91% of the nutrients in rivers from this source, table S2). In our scenario with a rapid economic development, the total area used for agriculture is projected to decrease between 2000 and 2050 in some sub-basins (up to 46%, figure S11); this results from rapid urbanization (table S1). Rivers may also receive nutrients from other sources. These are, for example, use of synthetic fertilizers to grow more crops, and sewage effluents from urbanized areas [17, 43]. However, we calculate a lower contribution of these sources to nutrients in Chinese rivers (up to 40%) compared to manure point source (table S2). Therefore, the risk for future river pollution by nutrients will largely depend on how food production will develop, in particular for animal production because of the large contribution of manure to water pollution currently.

Zoom In Zoom Out Reset image size

Some studies have already quantified N and P transport to waterbodies of the large Chinese rivers [56], in particular of the Changjiang River [9, 15, 46, 57–59]. A few studies exist on sub-basin analyses of the Changjiang River [9, 15, 59]. However, these studies ignore direct discharges of manure to rivers. Therefore, the current nutrient pollution of rivers may be underestimated. For example, our estimates of the total TDN inputs (from all sub-basins) to the Changjiang are comparable with the results of Liu et al [9] and Bao et al [15] for 1970 when direct discharges of animal manure were small, but higher than the results of Liu et al [9] and Xing et al [56] for 2000 when direct discharges of manure were considerable. Liu et al [9] and Bao et al [15] quantified around 2 Tg N transported to waterbodies of the Changjiang River in 1980, which is close to our estimate of 1.7 Tg TDN in 1970. For 2000 Liu et al [9] quantified approximately 4.5 Tg of N, of which around 0.4 Tg N is from animal manure used in agriculture. We, however, calculated 5.3 Tg TDN, of which 2.2 Tg TDN from direct discharges of animal manure and 0.4 Tg N from animal manure use (figure S6). We argue that the impact of direct discharges of animal manure from livestock production on river quality is much larger than the impact of over-fertilization of soils by synthetic fertilizers.

Few studies consider direct discharges of manure to rivers, but not at the sub-basin scale [13, 18, 60]. For example, Ma et al [13] quantified that 30% of N and 45% of P from manure were discharged to surface waters in 2005 at the national scale. Ti et al [60] adopted a 25% value for livestock waste discharged to waterbodies to quantify N export by the Changjiang, Zhujiang and Huang at the basin scale for the 2000s. We, however, demonstrate that this percentage is much higher for 2000 and ranges between 30% and 45% for N, and 46% and 64% for P among sub-basins (figure 2).

Most existing modeling studies on river export of N and P emphasize the importance of managing diffuse agricultural sources of N (e.g., synthetic fertilizer use) and point sewage sources of P to reduce water pollution in China (e.g., [17, 32, 46]). This is because these studies do not account for manure point sources. We show the importance to manage point manure sources for both N and P in the future. Studies on nutrient inputs to surface waters at the national (e.g., [13, 61]) and provincial scales (e.g., [18]) acknowledge the importance of reducing losses of manure. Our study can contribute to such analyses by providing a better spatially explicit understanding of water pollution at the sub-basin scale.