This study combines the observed riverine DIN (dissolved inorganic nitrogen) export and the
controlling factors (land-use, population and discharge) to inversely estimate the effective DIN
yield factors for individual land-use and DIN per capita loading. A total of 16 sub-catchments,
with different land-use compositions on the Danshui River of Taiwan, were used in this study.
Observed riverine DIN concentrations and yields varied from 20–450
Anthropogenic nitrogen, e.g. fertilizer and excrement, elevated by increasing population and food
production, cause serious water and land pollution (Fitzpatrick et al., 2004; Brown et al., 2009;
Tu, 2009; Jiang and Yan, 2010). In addition, it causes nitrogen cycle imbalances that already exceed
the safe operating parameters for humankind (Downing et al., 1999; Peterson et al., 2001; Rockstrom
et al., 2009). This increase of human-related N emission is unlikely to stop and has resulted in
eutrophication (Tsai et al., 2013), exacerbation of dissolved oxygen consumption in water bodies
(Parr and Mason, 2003), and has also been the cause of human health issues such as
Previous global studies on dissolved inorganic nitrogen yields (DIN includes nitrite, nitrate, and
ammonium) varies over 3 orders of magnitude from 0.58 to
The riverine DIN output from a watershed is commonly described by the flux or load (a product of the concentration and discharge volume) which is a mixture of all kinds of physical and biological processes interacting with topography, deforestation, urbanization, and hydrodynamics (Fitzpatrick et al., 2005; Bouwman et al., 2013). Thus, previous studies applied multiple regression analysis to estimate riverine DIN concentration and export through the use of dominant factors (e.g. land-use, population and runoff) (Smith et al., 2005, Lee et al., 2014). Other modeling attempts, like PLOAD, Global NEWs, and NANI, compiled export coefficients and parameterizations for individual sources (including point and non-point) to straightforwardly simulate or estimate the riverine DIN concentration and export (EPA, 2001; Seitzinger et al., 2005; Harworth et al., 1998). Undoubtedly, determination of the export coefficient plays a key role in such relevant applications. Our previous study combined the network of riverine DIN loads to inversely deduce the DIN yield for different land-use categories (Huang et al., 2012). This result is only applicable in the mean state of discharge condition. Moreover, population was not taken into account due to limited population in the study area. We therefore further advanced this concept of inverse estimation onto a larger watershed with a significantly different degree of urbanization and parameterized the stream discharge variability, which has not been considered before, to derive DIN yield factors.
This study monitored 16 sub-watersheds in the Danshui River watershed during 2002–2004. Those sub-watersheds have different mosaic land-use patterns from pristine to the intensively urbanized. Initially, the watershed landscape characteristics, including land-use composition and population density, were delineated. Secondly, the DIN concentration and stream discharge were compiled to calculate the riverine DIN load and yield. Thirdly, the DIN relationships between watershed landscape characteristics were quantitatively determined. Finally, the mosaic land-use pattern and DIN export for those sub-watersheds were superimposed to derive the DIN yield factor for each land-use and the per capita loading. This method, applied to the nested sub-watersheds with different urbanization gradients, can not only aid in understanding the process and the impacts of urbanization on DIN yield but also can be used to quantify the runoff-independent controlling factors.
The Danshui River flows through Taipei City and is one of the main rivers in Taiwan. The drainage
area is 2726
The land-use dataset in this watershed has been provided by the National Land Surveying and Mapping
Center (MLSC, 2008). MLSC introduced the aerial photos, satellite imagery, and field survey to
identify the 103 land-uses for the entire island and distributed the dataset in 2006. Since this
study focuses on the DIN yield, the original land-use types were reclassified into 4 main categories
(forest, buildings, agriculture, and water body). The forest class includes natural forest,
secondary forest, bamboo land, and shrubs. The agriculture class includes paddies (0.5 % of the
whole Danshui basin), dry crops (2.5 %) and tea farms and orchards (2.3 %). Tea farms and
orchards mainly occupy the hillside (1000–1200
The stream discharges for the 16 DIN sampling sites are illustrated in Table 2. Nine of the 16 sites
have discharge gauges maintained by the Water Resource Agency. For some of the missing measurements,
the simulated discharge derived by TOPMODEL was used instead (Huang et al., 2011). For those
sampling sites without discharge gauges, their daily discharges were estimated by an aerial ratio
method referring to the adjacent discharge gauges (Kao et al., 2004). In general, the annual
discharge during 2002 to 2004 for the Danshui River is 1032 to 3229
For DIN monitoring, 5 sampling sites were sampled weekly and 5 sites were sampled monthly due to
limited manpower. In addition to our own sites, there are an additional 6 sites monitored by the
Taiwanese EPA that are monthly sampled and are supplementary to this study (EPA, 2014). For our own
sites, the water samples taken from streams in situ were immediately filtered through
Whatman
The riverine DIN yield is defined as the total DIN flux or load normalized by drainage area. The flux or load is the product of substance concentration multiplied by total discharge volume during a specific period. However, the frequency of substance concentration, compared to the discharge, is much lower than stream discharge. Therefore, different methods have been proposed to supplement the concentration data for flux calculation. In this study, three commonly used methods, linear interpolation (LI), global mean (GM), and flow weighted (FW) have been applied to estimate the DIN flux (Fig. 2).
The LI method linearly interpolates the unmeasured days by the adjacent measurements, and then multiplies by the consecutive daily discharge (Fig. 2a and b). This method is most suitable when the sampling frequency is relatively high. In other words, this method does not explicitly consider the stream-flow effect that occurs when the sampling frequency is low. In these cases, the GM and FW methods are more suited. The GM method multiplies the mean of all sampled substance concentration with the total discharge within the study period to obtain flux. This simple method which does not consider any interaction between concentration and discharge may be the last method in priority, particularly for the lower sampling frequency. The FW method weighs the sampled concentration by discharge. The flux equals the total discharge volume multiplied by the flow-weighted mean of concentration. Comparing the three method (Fig. 2c and d), the GM method seems to under- and over-estimate the flux for enhancement and dilution condition, respectively. Meanwhile, the LI- and FG-derived fluxes are comparable and more accurate than GM-derived result. The advantages and disadvantages of the three methods have been widely discussed (Ferguson, 1987; Preston et al., 1989; Moatar and Meybeck, 2005; Birgand et al., 2010a,b; Lee et al., 2009), but it may not be easy to judge which one is universally suitable for any watershed. The method-choice depends on sampling frequency, hydrological conditions, and substance characteristics. Therefore, we used the average of all the three method-derived fluxes in this study.
As previously mentioned, the observed riverine DIN flux is made up of individual land-use and human
emissions. Previous studies addressed the compilation of the dataset (e.g. land-use and population)
to straightforwardly estimate the riverine DIN concentration and export, like PLOAD. By contrast,
our previous study revealed that this is a feasible way to inversely estimate the export coefficient
for individual land-use when the case of observed riverine DIN export is sufficient (Huang et al.,
2012). This simplified method shows that riverine DIN flux, in an annual basis, is mainly from
land-uses in a mountainous watershed, as following:
Since uncertainties inevitably exist in the riverine DIN yields, the uncertainties likely propagate
to the yield factors and human emission. Thus, the Monte Carlo approach is adopted instead of
a linear algebra one. A total of 100 000 random parameter sets were generated by a uniform
distribution generator to fit the observed 16 riverine DIN yields. The yield in dry season, wet
season and the whole 2003 data were separately used as three dependent variable sets for
consideration of seasonality. The RMSE (Eq. 3) and
The mean and SD of DIN concentrations among the 16 sites during 2002–2004 are shown
in Table 3. The DIN concentrations and stream discharge of three sites (K06, S07 and EPA1908) from
upstream to downstream during 2002–2004 are taken as examples and shown in Fig. 3. Generally, the
mean DIN concentrations vary from 18 to 452
The DIN concentrations against land-use are shown in Fig. 4. For the land-use composition, the
building area varies from 0.2 to 19.9 % (Fig. 4a), the agriculture area ranges from 0.1 to
23.1 % (Fig. 4b), and the forest area spans from 71 to 97 % (Fig. 4c). In those figures, the
DIN concentrations are significantly positively and negatively correlated to building and to forest
proportions, respectively. The Pearson's correlation coefficient (
For concentrations and runoff relations, the DIN concentrations against daily runoff of all
sub-catchments are shown in Fig. 5. Three sites, K06, S07, and EPA1908, with different degrees of
urbanization, are specifically illustrated showing the changes of C–Q relations along the
urbanization gradient. The relatively pristine catchment (K06), characterized by high forest cover
(96.7 %) and low population density (less than 15.8
Based on the observed concentration and runoff, the DIN yield for each sub-catchment was estimated by the three flux estimation methods (LI, GM, and FW). Figure 6 shows that the ratios of the GM- to LI-derived yield and the FW- to LI-derived yield vary with correlation coefficients between DIN and discharge. It revealed that the GM, compared to LI, tends to overestimate DIN yield while the C–Q relation reveals the dilution conditions, but underestimates the enhancement conditions. In addition, the ratio of FW- to LI-derived DIN yield was scattered for both conditions with a somewhat large bias. Since the sampling frequency, hydrological condition, and substance characteristics lead to various results, the average of the three method-derived fluxes was used to mitigate the systematic (GM to LI) and random (FW to LI) biases.
The DIN yields in the wet and dry season during 2002–2004 are shown in Table 4 and Fig. 7. The DIN yield in dry season increased from
83 to 6806
Throughout the method, the DIN yields (including dry-season, wet-season, and 2003 DIN yields) were
used to calibrate the 4 parameters (3 land-use yield factors and 1 per capita loading) for each
specific time frame as shown in Fig. 8. The yield factors of forests in dry season, wet season, and
annual basis are 7.1, 17.0, and 12.7
The calibrated DIN yield factors and per capita loading (the annual basis only) were further
verified by the 2002 and 2004 data. The performances of model calibration and validation for the 16
sub-catchments are shown in Fig. 9. For the calibration, the
DIN concentrations measured at catchment outlets reveal the mixed consequence from mosaic land-use
patterns and population densities within catchments. In general, the DIN concentrations of pristine
sub-catchments (forest
For DIN yield, the background DIN export for pristine sub-catchments are
For seasonality, the DIN yields of the all sub-catchments during the wet season are higher than
those in the dry season, though the DIN concentrations behave differently. The high DIN yield in the
wet season for both hydrological enhancement (pristine) and dilution condition (urbanized) is
interesting. The high DIN yield in the wet season can be expected for condition of hydrological
enhancement. However, even the DIN concentration is diluted by stream discharge, but if the increase
of stream discharge can compensate for the reduction of the DIN concentration, particularly when the
discharge surges more significantly, then the yield (product of discharge and concentration) becomes
higher as well as shown in Fig. 2b and d. Note, that the seasonality of discharge variation in
Keelung River is indistinct due to the north east monsoon in the winter. Thus the seasonality of DIN
yields in Keelung River is insignificant. Nevertheless, the elevated DIN yields in the wet season
confirm the primary role of stream discharge in DIN export (Smith et al., 2005). Meanwhile, it may
also imply that the storage of DIN in a subtropical terrestrial ecosystem is larger than expected,
so even with abundant precipitation (e.g. annual precipitation
The merit of this method is to untangle the individual DIN yield factors from the mosaic land-use
pattern in a catchment. The calibrated yield factors, with some uncertainties, can evaluate the
riverine DIN export promisingly. In this study, the uncertainty in the forest is probably negligible
due to the low deviation of the yield factor. For agriculture, the yield factors for dry and wet
season are 13.6 and 258.4
In urbanized areas, the DIN yield factor of buildings is significantly higher than the other two. This is only the case where the DIN yield factor in the wet season is lower than that in the dry season, indicating hydrological dilution conditions. Lee et al. (2014) revealed that both enhancement and dilution can be found in urban simultaneously. For example, the deposited aerosol (including sulfide, nitrous chemicals and others) on roads and buildings are washed out (Kang et al., 2010). However, the accumulated abundant benthic sludge is transport by high flows (Bernal et al., 1998; Sutton et al., 2000). Although both enhancement and dilution are important in this urban region, the signature of dilution is relatively obvious.
As for the human N emission to riverine DIN export, the average yield is
To demonstrate the advantages of this method, four scenarios are proposed. The scenarios contains
two population densities (population densities of 50 and 2000
Due to population growth, an increase in food demand and urbanization expansions is expected (Fan
and Agcaoili-Sombilla, 1997; McCalla, 1998; Foley et al., 2005) and human-induced terrestrial
nitrogen input is expected to be exported into the marine environments, eventually (Galloway et al.,
2004). Ultimately, urbanization causes an imbalance in the marine coastal nitrogen cycle (Downing
et al., 1999) and directly or indirectly affects atmospheric
This study reveals that urbanization associated factors (land-use and population) are the major contributors to the annual mean DIN concentration. Forest and building proportions tightly correlate the DIN concentration negatively and positively, respectively. From the perspective of hydrological control, the transport behaviors transform from enhancement to dilution with the increase of human disturbance and thus hydrology (e.g. discharge volume and seasonal variation) drives DIN exports in this region. According to the inversed estimation of yield factors, the distinctly seasonal yield factor of agriculture land is plantation-dependent and is affected by fertilizer application. This can aid in understanding the nitrogen budget and nitrogen cascade in agricultural lands. Meanwhile, the building yield factor in the wet season is only 50 % less than in the dry season indicating a dominant dilution effect. However, the export can be discharge-dominant as long as the increase of discharge is abundant enough to defeat the concentration reduction. The lower DIN per capita loading, compared to documented human N emissions from GDP or other methods, may be attributed to it being shared by building or treatment efficiency and thus could be regarded as an effective export coefficient. Finally, this method parameterizes the land-use-specific DIN yield factors and the per capita loading. This provides the basis for assessing possible land-use combinations in different scenarios which is a main concern for watershed management.
This study was sponsored by NSC Taiwan grants (NSC 103-2116-M-002-20, 102-2923-M-002-01-MY3, 103-2621-M-002-016). The authors appreciated the reviewers who provided the constructive comments for improving this study.
Watershed characteristics: land-use/pop. density.
Estimated seasonal Discharge for the sampling sites (unit: mm).
Observed mean DIN concentrations for the sampling sites.
Estimated seasonal DIN yield for the sampling sites (unit:
Landscape of the Danshui River watershed, including land-use pattern, population (red dots) and sampling sites. The empty circle and red square represent water sampling sites and discharge gauge.
Sketch of three flux estimation methods for hydrological enhancement conditions
DIN concentrations during 2002–2004. Upstream
DIN concentration against buildings
DIN concentrations against stream discharge for all sites during 2002 to 2004. The upstream (K06), midstream (S07), and downstream (EPA1908) sites are represented by green, brown, and red dots.
Ratio of three DIN yield estimations against correlation coefficient between DIN conc. and discharge.
Seasonal variations of estimated yields for all studied sites during 2002–2004.
Box plot for model derived parameters: land-use mean concentration and per capita load. The median values of the parameters are presented on the top of the box. Dry season, wet season and annual calibrated processes are filled with brown, blue and green color.
Model performance in calibration
The population effect on hydrological control in DIN concentration.
DIN yield with 2 population densities and discharges scenarios. The