Magnitudes and sources of dissolved inorganic phosphorus inputs to surface fresh waters and the coastal zone: A new global model

Authors


Abstract

[1] As a limiting nutrient in aquatic systems, phosphorus (P) plays an important role in controlling freshwater and coastal primary productivity and ecosystem dynamics, increasing frequency and severity of harmful and nuisance algae blooms and hypoxia, as well as contributing to loss of biodiversity. Although dissolved inorganic P (DIP) often constitutes a relatively small fraction of the total P pool in aquatic systems, its bioavailability makes it an important determinant of ecosystem function. Here we describe, apply, evaluate, and interpret an enhanced version of the Global Nutrient Export from Watersheds (NEWS)–DIP model: NEWS-DIP–Half Degree (NEWS-DIP-HD). Improvements to NEWS-DIP-HD over the original NEWS DIP model include (1) the preservation of spatial resolution of input data sets at the 0.5 degree level and (2) explicit downstream routing of water and DIP from half-degree cell to half-degree cell using a global flow-direction representation. NEWS-DIP explains 78% and 62% of the variability in per-basin DIP export (DIP load) for U.S. Geological Survey (USGS) and global stations, respectively, similar to the original NEWS-DIP model and somewhat more than other global models of DIP loading and export. NEWS-DIP-HD output suggests that hot spots for DIP loading tend to occur in urban centers, with the highest per-area rate of DIP loading predicted for the half-degree grid cell containing Tokyo (6366 kg P km−2 yr−1). Furthermore, cities with populations >100,000 accounted for 35% of global surface water DIP loading while covering less than 2% of global land surface area. NEWS-DIP-HD also indicates that humans supply more DIP to surface waters than natural weathering over the majority (53%) of the Earth's land surface, with a much larger area dominated by DIP point sources than nonpoint sources (52% versus 1% of the global land surface, respectively). NEWS-DIP-HD also suggests that while humans had increased DIP input to surface waters more than fourfold globally by the year 2000, human activities such as dam construction and consumptive water use have somewhat moderated the effect of humans on P transport by preventing (conservatively) 0.35 Tg P yr−1 (∼20% of P inputs to surface waters) from reaching coastal zones globally.

1. Introduction

[2] Global budgeting efforts suggest that P mining and subsequent use as fertilizer has more than doubled P inputs to the environment over natural, background P from weathering [Mackenzie et al., 1998; Bennett et al., 2001; Fixen and West, 2002]. Often a limiting nutrient in lakes and other freshwater systems, P is also thought to play an important role in controlling coastal primary productivity and ecosystem dynamics, increasing frequency and severity of harmful and nuisance algae blooms [Anderson et al., 2002] and hypoxia [Diaz and Rosenberg, 2008], among other effects. Though coastal systems are typically thought of as nitrogen (N) limited, there are several coastal systems where P limitation has been demonstrated for at least part of the year [Harrison et al., 1990; Jensen et al., 1998; Fisher et al., 1999; Murrell et al., 2002; Sylvan et al., 2006; Conley et al., 2009], and coastal P limitation may well become more prevalent if anthropogenic N mobilization increases faster than P mobilization, with projected increases in food and energy production [Justic et al., 1995; Turner et al., 2003].

[3] In many systems, dissolved inorganic P (DIP) (also called soluble reactive phosphorus (SRP) or orthophosphate (PO43−)) constitutes a relatively small portion of the phosphorus in rivers (∼1.5 Tg P yr−1 transported as DIP globally versus ∼20 Tg P yr−1 as total P (TP) globally [Meybeck, 1982; Melack, 1995]). However, whereas all of the DIP pool is generally thought to be bioavailable in rivers, lakes, and coastal waters, significant portions of the particulate and organic P pools are not available for use by organisms [Bradford and Peters, 1987; Ekholm, 1994; Fox, 1989]. Therefore, even accounting for desorption of sorbed P in estuaries [Froelich, 1988; Howarth et al., 1996], DIP plays an important role in controlling the biology of such systems. As such, the development of a DIP loading and river transport model constitutes a critical first step toward a synthetic understanding of coastal P delivery, which must eventually also include models for delivery of particulate and dissolved organic P.

[4] There have been a number of attempts to model within-basin P dynamics at scales ranging from a single catchment [Baffaut and Benson, 2009] to a large river basin (U.S. Geological Survey (USGS)–SPARROW; Alexander et al. [2008]), to studies that include several midsized basins [Thieu et al., 2009]. However, until now, global-scale P transport models have been limited to predicting P export at the mouths of large river basins as a function of basin-averaged characteristics [e.g., Caraco, 1995; Smith et al., 2003; Harrison et al., 2005] or in a nonspatially explicit manner altogether [e.g., Mackenzie et al., 1998]. In the sections that follow, we describe, apply, evaluate, and interpret output from an enhanced version of the Global Nutrient Export from Watersheds (NEWS)–DIP model [Harrison et al., 2005]: NEWS-DIP–Half Degree (NEWS-DIP-HD).

2. Methods

2.1. NEWS-DIP-HD Description

[5] NEWS-DIP-HD constitutes an update and modification of the original NEWS-DIP model, which is described in detail by Harrison et al. [2005]. As with NEWS-DIP, NEWS-DIP-HD predicts average annual DIP export values, and the central equation of NEWS-DIP-HD is identical to that of the updated NEWS-DIP model (NEWS-DIP-II; see E. Mayorga et al., Global Nutrient Export from WaterSheds 2 (NEWS 2): Model development and implementation, submitted to Environmental Modelling and Software, 2009). The primary difference between NEWS-DIP-HD and NEWS-DIP-II lies in the fact that NEWS-DIP-HD is applied on a per-half-degree grid cell basis, rather than at the large- basin scale. The central equation for NEWS-DIP-HD is as follows:

equation image

where DIP is the local DIP yield (kg P km−2 yr−1) to surface waters computed for each half-degree cell globally (as opposed to DIP load (kg P basin−1 yr−1) or DIP concentration (mg P L−1)). DIP is calculated as a function of within-cell P sources, which include both point sources and diffuse sources (Table 1). Point sources are calculated as the sum of P from human sewage (Psew) and P from P-based detergents (Pdet). Diffuse sources are calculated as a function of runoff (R) (m yr−1), fertilizer P inputs (Pfert) (kg P km−2 yr−1), animal manure P inputs (Pam), P removal by harvest and animal grazing (Pexp) (kg P km−2 yr−1), and four calibrated coefficients defining the shape of the runoff response curve for weathering and nonpoint DIP sources (a, b, Wmax, and Lmax, as in the work of Harrison et al. [2005]). Diffuse sources were treated as a sigmoid function of runoff, increasing slowly with runoff at low runoff values, more rapidly with runoff at higher runoffs, and topping out at a threshold level in high-runoff systems. This sigmoid relationship between runoff and diffuse sources is responsible for the term (1/(1 + (R/a)b)) in NEWS-DIP-HD's central equation. Input variables consisted of spatially explicit, 0.5° × 0.5° resolution gridded data sets (Table 2). Calibrated coefficients a, b, Wmax, and Lmax were taken directly from the calibration of the NEWS-DIP model (E. Mayorga et al., submitted manuscript, 2009, Table 1) and were not recalibrated at the half-degree scale. These coefficients were set to 0.85, 2, 26, and 0.04, respectively. This calibration was achieved using approximately half of the basins (56 rivers) in the original NEWS-DIP calibration and validation data set [Harrison et al., 2005]. Calibration was achieved by optimizing the model to attain the highest model efficiency (R2) while maintaining coefficients within observation-based ranges (as in the work of Harrison et al. [2005]). Model efficiency (capital R2, not the coefficient of determination (r2)) is a metric ranging from 0 to 1 reflecting the degree of fit between measured and modeled values [Nash and Sutcliffe, 1970]. When R2 = 1, all points fall on the 1:1 line. When R2 is 0, model error is equal to the variability in the data. Coefficients relating to point source inputs or reservoir retention were not calibrated.

Table 1. Input Data Sets for NEWS-DIP-HD
Data SetResolutionYearSource(s)
Basin delineations0.5°1960–1994STN30 [Vörösmarty et al., 2000a, 2000b]
River networks0.5°1960–1994STN30 [Vörösmarty et al., 2000a, 2000b]
Water runoff and discharge0.5°1960–1994Fekete et al. [2002]
Population density0.5°2000Bouwman et al. [2005]
Fertilizer P inputs0.5°2000Bouwman et al. [2009]
Manure P inputs0.5°2000Bouwman et al. [2009]
Crop harvest P export0.5°2000Bouwman et al. [2009]
Sewage P inputs0.5°2000Bouwman et al. [2005]
Detergent P inputs0.5°2000Van Drecht et al. [2009]
Dam locations and volumes0.5°2000International Commission on Large Dams (ICOLD) [2003]
Table 2. Metrics of Model Performance for NEWS-DIP-HD and Other Modelsa
ModelDIP Yield (kg P km−2 yr−1)DIP Load (Ton P basin−1)IQRbPrediction Errors (%)
r2R2r2R2min25th75thmax
  • a

    Validated with data set described in section 2.2 unless otherwise noted. R2 is model efficiency as defined in section 2.5, and r2 is the coefficient of determination. Errors are computed as the difference between the predicted and measured values of stream phosphorus yield (kg km−2 yr−1) expressed as a percentage of the measured export (section 2.5).

  • b

    Interquartile range (IQR) (difference between the 25th and 75th percentiles of the distribution of errors).

  • c

    Harrison et al. [2005] model using validation basins only.

  • d

    NEWS-DIP-HD with USGS-WQN data.

  • e

    NEWS-DIP-HD with data from Harrison et al. [2005].

  • f

    NEWS-DIP-HD using all validation data.

  • g

    Smith et al. [2003] model using validation data from Harrison et al. [2005].

  • h

    Caraco [1995] model using validation data from Harrison et al. [2005].

NEWS-DIPc0.560.510.600.47247−90−92392,542
NEWS-DIP-HDd0.600.510.780.7782−90−50324,811
NEWS-DIP-HDe0.450.290.520.52221−99.6−461754,574
NEWS-DIP-HDf0.500.390.620.61204−99.6−471574,811
LOICZ-DIPg0.460.170.520.46502−78−1648713,672
CARACOh0.410.340.540.43692−963072219,982

[6] To calculate DIP contributions from individual cells to coastal margins, local DIP yields were multiplied by a cumulative transfer efficiency factor. The cumulative transfer efficiency factor for each cell was calculated as the product of all local downstream transfer efficiencies (1 D for each half-degree, downstream grid cell, where D is the fraction of DIP removed by reservoirs, calculated for each cell according to Harrison et al. [2005] on the basis of the water residence time of the reservoir(s) in that cell) and the ratio Qact:Qnat for the entire basin (where Qact is water discharge accounting for human activities such as water extraction and Qnat is water discharge without the influence of humans). This makes it possible to use NEWS-DIP-HD to predict coastal DIP yield for every half-degree cell. Values for Qact:Qnat were the same as those used by Harrison et al. [2005], and when not available, were assumed to equal one.

[7] The magnitudes of individual source contributions to local half-degree cells were calculated as follows:

equation image
equation image
equation image
equation image
equation image

where H is population density, Psw is per capita delivery of human P effluent to surface waters via sewage, Pdet is per capita delivery of P to surface waters via sewage, and other symbols and coefficient values are the same as in equation (1) and Appendix A. To calculate coastal contribution, each source was multiplied by the cumulative transfer efficiency from each cell to the coast.

[8] Improvements to NEWS-DIP-HD over the original NEWS DIP model include (1) the preservation of spatial resolution of input data sets at the 0.5 degree level, (2) explicit downstream routing of water and DIP from half-degree cell to half-degree cell using a global flow-direction representation [Vörösmarty et al., 2000a, 2000b], (3) the inclusion of detergent P as a potential DIP point source, (4) the explicit use of the surface P balance concept to calculate nonpoint DIP sources, and (5) the incorporation of more recent input data sets as model drivers (2000 instead of 1995). Several of these enhancements (3–5) occurred as a result of updating the NEWS-DIP model to run it with Millennium Assessment scenarios [Seitzinger et al., 2009] and are described more completely by E. Mayorga et al. (submitted manuscript, 2009), Bouwman et al. [2009], and Van Drecht et al. [2009], respectively.

2.2. Model Validation Data

[9] Concentration and water discharge data from 205 globally distributed sites were used to evaluate the predictive power of the NEWS-DIP-HD model (Figures 1 and 2). These data were derived from four primary sources, including data used for the calibration and validation of the original NEWS-DIP model (109 sites; Harrison et al. [2005]), the United Nations Global Environment Monitoring System (GEMS) Water (http://www.gemswater.org/; 33 sites), the USGS WQN (53 sites; Alexander et al. [1996]), and the CAMREX study of the Amazon River (10 sites: see Devol et al. [1995], Figure 1, and Appendix B). These sites encompassed a broad range of basin sizes (5896–6,112,000 km2), a variety of climate types (ranging from tropical to boreal), and both coastal and inland sites (Figure 1). When available, reported basin surface area was used in calculations. However, in certain cases (e.g., with CAMREX and GEMS data) it was necessary to calculate watershed surface area on the basis of STN 6 hydrography [Vörösmarty et al., 2000a, 2000b]. All sites were georeferenced to the STN 6 hydrographic network for ease of comparison between measured and modeled P export. Together, the coastal sites included as validation data accounted for 49% of global runoff. DIP load at each site was calculated as the product of mean annual water discharge and flow-weighted mean [DIP] when sufficient data were available (17 cases). When this was not possible, median [DIP] (90 cases) or mean concentrations (94 cases) were used. DIP yield for each site (kg P km−2 yr−1) was calculated as load divided by watershed contributing area. For USGS data, only 5th order or larger streams were used.

Figure 1.

Spatial distribution of basins used to validate the NEWS-DIP-0.5 model (n = 205 overall). Crosses represent sampling stations within the Amazon River Basin (n = 10) [Devol et al., 1995]. Plus signs represent sampling stations monitored by the Global Environment Monitoring System (GEMS) Water program not used by Harrison et al. [2005] (n = 33; http://www.gemswater.org/index.html). Hollow circles represent sampling stations within the Mississippi, Sacramento, and San Joaquin River basins (n = 53; data source: Alexander et al. [1996]). Black diamonds represent stations used in calibration and validation of the original NEWS-DIP model (n = 109; Harrison et al. [2005]). See Appendix B for data, model output, and station names.

Figure 2.

Measured versus modeled DIP (a) load (kg P basin−1 yr−1) and (b) yield (kg P km−2 yr−1) for Mississippi River Basin stations (hollow circles), global coastal stations included by Harrison et al. [2005] (black diamonds), global stations from the United Nations Global Environmental Monitoring System (GEMS; plus signs), and data from the Amazon Basin [Devol et al., 1995] (crosses). See Appendix B for data, model output, and basin names. The 1:1 line is also shown. Symbols are the same as in Figure 1.

2.3. Model Input Data

[10] Sewage point source P was calculated in a manner similar to Van Drecht et al. [2009]. In this method, per capita excretion of P is calculated at the national level as a function of per capita income (higher PPP results in higher per capita P excretion), per capita P detergent use, P removal efficiency (by sewage treatment), and sewage connectivity. Country per capita P excretion rates were spatially disaggregated by multiplying them by population density, averaged at the half-degree scale. One subtle but important difference between the estimate of point source P inputs used in this analysis and the estimate generated by Van Drecht et al. [2009] is that while Van Drecht et al. [2009] assume that nonurban half-degree cells contribute no point source P to surface waters, here we assume that an urban fraction of each half-degree cell contributes point source P (where the urban fraction of each cell is equal to the fraction of a country's population that is urban). Anthropogenic nonpoint source P inputs included fertilizer and manure, but not septic P, and were calculated according to Bouwman et al. [2009]. In this method, fertilizer and manure inputs were disaggregated spatially using land use information and national fertilizer use statistics [FAO, 2008]. Then a surface P balance was calculated by subtracting the P exported in crop harvest and animal grazing from fertilizer and manure P inputs to avoid double counting of P inputs.

2.4. Postprocessing of Model Output

[11] Regional and global totals of DIP export were calculated as the sum of all grid cells within continents and ocean drainages as defined by the STN6 global half-degree hydrography data set [Vörösmarty et al., 2000a]. P retention was calculated as locally emitted DIP minus DIP delivered to the coast. This estimate is likely quite conservative as only large dams were used to estimate reservoir storage of DIP, and where no consumptive water used data were available, this loss pathway was assumed to be negligible. Also, floodplain and wetland storage are not included explicitly in the model, and these loss pathways could be significant.

2.5. Model Evaluation (Model Uncertainty, Sensitivity, and Efficiency)

[12] Model uncertainty was evaluated by comparing measured and predicted DIP load and yield. Metrics of model error included root mean squared error (RMSE), model efficiency [Nash and Sutcliffe, 1970], and interquartile range (IQR). Model efficiency was calculated using log-transformed model predictions, load, and yield data. RMSE and IQR were determined using untransformed data and predictions. Model error was compared with interannual variability in several rivers for which multiple years of DIP export data were available. Model sensitivity to inputs and coefficients was examined by increasing each model input by 10% and examining the response in model output. Model sensitivities are expressed as percent change in output (100 × new value–original value)/original. In addition, to evaluate how critical each model component was to model predictive capacity, individual model components were removed and Nash-Sutcliffe model efficiency was recalculated. Components with little impact on the model had little impact on Nash-Sutcliffe efficiency, whereas removal of critical model components had a large impact on Nash-Sutcliffe efficiency.

[13] Because DIP load and yield data collected by the USGS in the Mississippi River are based on samples collected at least seasonally over multiple years and analyzed using a consistent analytical approach, we viewed these data as potentially higher quality than data from the diverse array of sources contained within the other global data sets included in this analysis and thus likely a better test for the NEWS-DIP-HD model than data collected from other sources. Because of this, we present a comparison between NEWS-DIP-HD predictions and USGS data as well as a comparison between NEWS-DIP-HD predictions and all available P export data. For ease of comparison with the original NEWS-DIP model, we also present a comparison between NEWS-DIP-HD predictions and DIP export measurements used in the original NEWS-DIP paper [Harrison et al., 2005].

3. Results and Discussion

3.1. Model Performance

[14] NEWS-DIP explains 60% and 50% of the variability in per-area DIP export (DIP-yield) for USGS sites and for all validation sites, respectively, similar to the NEWS-DIP model and somewhat more than other global models of DIP loading and export (Figure 2 and Table 2). It explains 78% and 62% of the variability in per-basin DIP export (DIP load) for USGS and global stations, respectively (Table 2).

[15] Despite the reasonably good fit between measured and modeled DIP yield (or load), error on a basin-by-basin scale is considerable. The standard error of log-transformed predictions is 0.42. DIP yield predictions for 58% of basins are within a factor of 2 of measurement-based estimates, 82% are within a factor of 4, and 90% are within 1 order of magnitude. Error in DIP yield predictions associated with large basins is similar to error associated with relatively small basins. However, absolute error associated with high-yield basins is somewhat greater than error associated with low-yield basins (Figure 2). The range of errors in NEWS-DIP-HD predictions is comparable to or substantially smaller than that for other DIP export models, as indicated by the interquartile range and distribution of prediction errors (Table 2).

[16] The error associated with NEWS-DIP is similar in magnitude to the interannual variability of DIP yields in several U.S. rivers. For example, the difference between minimum and maximum DIP export years is fivefold for the Mississippi River and over an order of magnitude for the Potomac River (data from Alexander et al. [1996]). This suggests that NEWS-DIP-HD predictions are likely to fall within the range of interannual variability for any given river.

[17] In general, the NEWS-DIP-HD model preserves the spatial resolution of DIP sources and sinks at the 0.5 degree level without sacrificing prediction accuracy. In the sections that follow, we use the NEWS-DIP-HD model to gain insight into patterns, controls, and sources of DIP export from watersheds worldwide. We then explore model sensitivities, uncertainties, and potential ways to improve our capacity to model DIP export in future efforts.

3.2. Model Output

3.2.1. Spatial Distribution of DIP Export and Sources

3.2.1.1. Export

[18] Comparison of NEWS-DIP-HD output with NEWS-DIP output reveals several new insights. In general, NEWS-DIP-HD estimates are consistent with a previous global analysis of P loading to surface waters (Figure 3 and Harrison et al. [2005]). However, impressive within-basin spatial heterogeneity is revealed by applying the NEWS-DIP-HD model (compare Figures 3a and 3b). Hot spots for DIP loading tend to occur in urban centers (e.g., in the northeastern United States, the United Kingdom, western Europe, Mexico, Bangladesh, India, Pakistan, China, and Japan). High rates of DIP loading also occur in humid areas with high predicted rates of P weathering (e.g., in the Brazilian Amazon, West Africa, Central America, the northwest United States, and Indonesia). In addition, inland cells appear to be quite important with respect to their contribution to coastal DIP loading, particularly in the midwestern United States, western Europe, southern Asia, and eastern China (Figure 3).

Figure 3.

(a) NEWS-DIP-predicted and (b) NEWS-DIP-HD-predicted DIP yield by half-degree grid cell (kg P km−2 yr−1). White areas are either endoreic (Figure 3a) or have a predicted DIP loading to surface waters equal to zero (Figure 3b).

[19] On a half-degree grid cell basis, predicted DIP yields ranged over several orders of magnitude, from zero in arid zones with little or no population (e.g., Saharan Africa and the Australian interior) to 6366 kg P km−2 yr−1 in the grid cell containing Tokyo, Japan (Figure 3). In general, urban areas are responsible for much of the DIP input to surfaces waters. In our analysis, the half-degree grid cells containing the world's 19 cities with populations greater than 5 million contribute 0.04 Tg of P to surface waters, or roughly 4% of the global total anthropogenic DIP input. The 214 cities with populations greater than 1 million account for 19% of DIP inputs to surface waters globally, and the 1058 cities with populations greater than 100,000 individuals contribute 35% of global anthropogenic surface water DIP loads. Together, all of the cities with populations greater than 100,000 account for less than 2% of the global land surface, highlighting the intensity of DIP production in urban areas.

[20] In general, areas with high predicted rates of DIP loading to surface waters correspond to areas where P-related water quality problems have been reported (e.g., midwestern United States, western Europe, the United Kingdom, Japan, India, and eastern China; see United Nations Environment Programme (GEMS Water Programme, National Water Resources Institute, Environment Canada, 2001, available at http://www.unep.org/dewa/assessments/ecosystems/water/vitalwater/10.htm) and Smith et al. [2003]). However, there are a number of areas globally that have received relatively little attention in terms of focus on eutrophication-related problems, but that are highlighted in this analysis as likely hot spots for DIP loading to surface freshwaters (e.g., Bangladesh, southern Brazil, central Chile, Indonesia, Pakistan, portions of North Africa, eastern Europe and Russia, and Korea).

[21] NEWS-DIP-HD estimates of DIP retention within surface water systems varied widely from cell to cell and from basin to basin. Hot spots for DIP retention are indicated to occur in the St. Lawrence, Volga, Indus, Dnieper, Danube, Yangtze, Yellow, Pearl, and Parana river basins, and in others as well (Figure 4). High rates of DIP retention may also occur in additional river basins, but where information about reservoir locations and sizes and consumptive water use was lacking, zero retention was assumed, making our estimate of DIP retention quite conservative.

Figure 4.

NEWS-DIP-HD-estimated DIP retention (kg P km−2 yr−1) globally by half degree. Estimates without information regarding reservoir locations and consumptive water use were assumed to retain no DIP, making this quite a conservative estimate of DIP retention within watersheds globally.

3.2.1.2. Sources

[22] NEWS-DIP-HD predicts that human activity dominated DIP export in half-degree cells, accounting for over half (53%) of the Earth's land surface, with a much larger area dominated by DIP point sources than nonpoint sources (52% versus 1% of the global land surface, respectively). Point source–dominated areas occur throughout both the north and south temperate zones and in much of the populated, arid tropics and subtropics as well. Of the relatively little surface area dominated by nonpoint sources of DIP, 35% was dominated by manure sources and 65% was dominated by fertilizer sources. Regions where NEWS-DIP-HD suggests fertilizer P inputs are important as a DIP source include much of New Zealand, portions of Vietnam, Japan, Korea, Finland, Norway, parts of Bangladesh, and isolated half-degree grid cells in the Brazilian Amazon and the midwestern United States. Regions where manure P sources are indicated as the single greatest DIP source include Ethiopia, isolated portions of the Brazilian Amazon, China, Nepal, New Zealand, Madagascar, Venezuela, the midwestern United States, and southern Canada. There are also a number of regions where the model predicts no DIP input to surface waters (white regions in Figures 3 and 5) because of a lack of P inputs to the landscape. An examination of Figures 3 and 5 together indicates that many of the high-DIP-yielding areas are areas where human activities (and most often sewage point sources) dominate DIP inputs to surface waters, though interestingly this is not true in all cases (e.g., the Brazilian Amazon, where NEWS-DIP-HD actually somewhat underestimates DIP transport). Despite the undeniable impact of human activity, and especially P point sources on DIP loading to freshwaters, DIP loading across a significant portion of the Earth's land surface is still dominated by nonanthropogenic DIP sources. NEWS-DIP-HD suggests that in 2000 weathering was the dominant source of coastal DIP over 35% of the Earth's land surface (Figure 5). Basins where NEWS-DIP-HD predicts weathering to dominate DIP sources lie across all latitudes and occur in areas with relatively little human influence and in wet tropical systems such as the Amazon, the Congo, northern Australia, and Indonesia.

Figure 5.

Dominant source of DIP by half-degree grid cell. “Dominant source” is defined as the modeled source that NEWS-DIP-HD predicts contributes the largest single fraction of DIP to the coast.

[23] Though a few studies have attempted to attribute sources of TP [Boynton et al., 1995; Baker and Richards, 2002; Moore et al., 2004] and total PO4 (dissolved plus acid-soluble, but undigested particulate) [Jordan et al., 2003] to river P loading, we were able to locate only one study quantifying the relative importance of different land-based sources specifically to river DIP loading. This one study of a relatively rural portion of the Thames River watershed [Cooper et al., 2002] suggests that point sources account for 77–97% of the river DIP inputs, depending on the year (1995–1999). NEWS-DIP-HD predicts that for the whole Thames River watershed, including the more urban portions, point sources account for 99% of the DIP loading, a fairly good agreement with the local study.

[24] Comparison of NEWS-DIP predictions with regional studies that estimate sources of other P forms also suggests that NEWS-DIP-HD predictions are reasonable. Studies attributing TP or total PO4 to point and nonpoint sources have calculated point source inputs on the basis of data from wastewater treatment plants and subtracted that value from total export to calculate contribution from nonpoint sources. Such studies have estimated that point sources contribute 95% of the TP load the Patuxent River [Boynton et al., 1995] and 50% of the TP load to Baltic rivers [Helsinki Commission (HELCOM), 2003]. Assuming TP:DIP ratios of 0.033 and 1 for point and nonpoint sources, respectively (similar to values reported by Cooper et al. [2002] for the Thames), this translates to estimated point source contributions of 97 and 67% for the Patuxent and Baltic, respectively. NEWS-DIP-HD estimates that point sources account for 99 and 92% of the DIP source in the Patuxent and Baltic regions, respectively.

[25] NEWS-DIP-HD suggests that point sources, as opposed to anthropogenic diffuse sources, most often dominate DIP export to coastal regions on global scale. However, in intensively farmed regions, nonpoint sources of DIP can dominate river and coastal DIP loading. On average, DIP export appears to be more often dominated by point sources than river-exported DIN. Whereas global DIN export has been attributed mainly to nonpoint N sources, particularly N fertilizer [Seitzinger and Kroeze, 1998; Caraco and Cole, 1999; Green et al., 2004], global DIP export is influenced mainly by sewage point sources. The dominant role of point sources in controlling DIP export is consistent with previous global analyses [Caraco, 1995; Smith et al., 2003; Harrison et al., 2005].

3.2.2. Global and Regional Analyses

[26] We estimate that 1.45 Tg P yr−1 reached river mouths emptying into major ocean basins as DIP in 2000. This estimate is similar to other recent, measurement-based and model-calculated estimates of global DIP export, which range from 0.8 to 2.4 Tg yr−1 [Pierrou, 1976; Meybeck, 1982; Richey, 1983; Wollast, 1983; Smith et al., 2003; Harrison et al., 2005]. Of the 16.49 Tg of P we calculate are loaded on watersheds by human activity globally, we estimate approximately 8.7% is exported by rivers as DIP. Globally, retention of DIP due to reservoir construction and consumptive water use (i.e., excluding P that is assumed to never reach aquatic systems) is conservatively estimated as 0.34 Tg P yr−1, roughly 20% of the total DIP delivered annually to the coastal ocean globally and essentially equivalent to the amount of naturally weathered DIP (0.35 Tg P yr−1). According to NEWS-DIP-HD, anthropogenic sources account for 76% (1.1 Tg yr−1) of the DIP exported to the coastal zone globally, with the remaining 24% (0.35 Tg yr−1) attributable to natural weathering processes. This predicted rate of weathering-derived P export is virtually identical to the rate predicted by Meybeck [1982] (0.4 Tg yr-1), an estimate based on a study of relatively unimpacted rivers.

[27] According to NEWS-DIP, P point sources alone account for over half (72%) of the total global DIP export via rivers. Inorganic fertilizer (2.0%) and animal manure (2.0%) contribute substantially smaller fractions of the coastal DIP load. On every continent and in every ocean basin, human sewage is the largest source of anthropogenically derived exported DIP, followed by P from P-based detergents.

[28] According to NEWS-DIP predictions, Asia is the largest continental exporter of DIP (Figure 6), contributing 38% (0.72 Tg yr−1) of the total global DIP export from watersheds to the global coastal ocean and inland seas. Export rates for other continents vary substantially, with Europe, South America, North America (including Greenland), Africa, Oceania (including New Zealand), and Australia each exporting 0.27, 0.32, 0.26, 0.21, 0.11, 0.02 Tg of DIP P yr−1, respectively. Of the world's ocean basins, NEWS-DIP predicts that the Atlantic Ocean receives the most DIP from land-based sources (0.59 Tg yr−1), followed by the Pacific and Indian oceans (0.50 and 0.25, respectively), the Mediterranean Sea (0.08 Tg yr−1), and the Arctic Ocean (0.03 Tg yr−1).

Figure 6.

River export of DIP (Tg P yr−1) from continents and to ocean basins. Relative influence of various P sources calculated according to NEWS-DIP-HD.

[29] Continental and ocean basin calculations suggest that humans have equaled or outstripped natural processes as a source of DIP to the coast on all continents (ranging from 57% of DIP inputs in South America to 92% of DIP inputs in Europe) and all ocean basins except for the arctic, where anthropogenically derived P accounts for 48% of P sources (Figure 6). Although it is difficult to compare 1995 and 2000 predictions because of changes in methods and the DIP model between the previous model run which, for example, did not include P-based detergents [Harrison et al., 2005] and this model run which does include such detergents, such a comparison suggests that human influence on DIP export increased substantially between 1995 and 2000. The greatest increases between 1995 and 2000 were observed in Asia and the Pacific Ocean.

3.3. Sources of Uncertainty and Future Directions

[30] The NEWS-DIP-HD model represents a significant step forward in terms of capacity to model river DIP export at the global scale. However, there is still significant room for model improvement. As global data sets improve, there will be opportunities to greatly improve estimates of river DIP export. In sections 3.3.1 and 3.3.2, we use patterns of model error along with model efficiency and sensitivity analyses (Tables 3 and 4) to infer where improvements to global data sets and model improvements are likely to be most useful in enhancing DIP yield estimates. We also examine the potential implications of assumptions made during the model development process and suggest future directions in the field of global nutrient modeling.

Table 3. Model Efficiencies for Comparison of Log-Transformed Measured and Model-Predicted DIP Yielda
TreatmentModel Efficiency (R2) DIP Yield (kg km−2 yr−1)Percent Change in Model Efficiency Resulting From Component Removal DIP Yield
  • a

    Using NEWS-DIP-HD with various components removed. All available measurement data were used for this analysis.

Complete model0.41 
No weathering P−1.32−422
No fecal point sources−0.39−195
No consumptive water use0.23−44
No reservoir retention0.32−22
No nonpoint sources0.410
No detergent point sources0.447
Table 4. Results of a Sensitivity Analysis Indicating Mean Change in Predicted DIP Yield as a Function of Increasing Input Data Sets and Model Parameters by +10%
TreatmentParameter or InputMean Change in Predicted DIP (kg km−2 yr−1) (%)
+10%Sewage P5.45
+10%Runoff (m)3.43
+10%Wmax3.30
+10%Detergent P0.98
+10%Manure P0.38
+10%Lmax0.27
+10%Fertilizer P0.24
+10%P removal by crop export and animal grazing−0.35
+10%b−1.03
+10%% Retention in reservoirs−1.717
+10%A−2.96
+10%Consumptive water use−10.0

3.3.1. Model Efficiency and Sensitivity

[31] An analysis of model efficiency, wherein model components were removed sequentially to evaluate the contribution of each to model predictive capacity, suggests weathering and sewage point source submodels are particularly important model drivers (Table 3). This analysis suggests that consumptive water use and P retention in reservoirs also play significant roles in the correct determination of DIP yield by NEWS-DIP-HD, but that anthropogenic nonpoint sources, while important in certain regions, are less vital in explaining DIP yield than other model components at the global scale (Table 3). Removal of the detergent point source term from NEWS-DIP-HD actually improved model efficiency slightly.

[32] A sensitivity analysis of NEWS-DIP-HD in which model inputs and coefficients were increased by 10% in order to evaluate model response (Table A1) suggests that the NEWS-DIP-HD model is fairly robust. Ten percent changes in all input parameters, coefficients, and all possible combinations of coefficients result in average changes in predicted DIP yield of 10% or less, and in most cases, substantially less. Of course sensitivities of individual half-degree pixels vary substantially more, depending on the dominant control on DIP transport in a given region. As with the original NEWS-DIP model, NEWS-DIP-HD predictions are somewhat sensitive to small changes in the weathering-related parameters Wmax, R, a, and b (Table A1). NEWS-DIP-HD's sensitivity to changes in the weathering submodel coefficients suggests that any improvement in NEWS-DIP-HD's representation of weathering-derived P is likely to improve model predictive capacity. Finally, NEWS-DIP output is relatively insensitive to removal of its nonpoint P source term (Table 3) and to manipulation of nonpoint source input data sets (Table 4). This suggests that inaccuracies in fertilizer and manure input data sets have relatively minor impacts on regional and global model predictions, especially in comparison with inaccuracies associated with other model inputs. However, the importance of these terms is likely to vary spatially.

3.3.2. Future Directions

[33] Taken together, model efficiency and sensitivity analyses suggest several areas for future improvements in NEWS-DIP-HD. For example, these analyses both suggest that point sources are important in driving predictions of DIP export (Tables 3 and 4). These analyses also suggest that NEWS-DIP predictions are sensitive to estimates of weathering (Tables 3 and 4). In future DIP export models, it may be possible to reduce uncertainty in estimates of weathering rates by refining the NEWS-DIP-HD submodel for predicting weathering-derived P (equation (4)) through the inclusion of factors thought to influence weathering rates such as temperature, soil type, soil parent material, and pH as improved global data sets become available. Finally, efficiency analysis suggests that our characterization of the linkages between human activity, P-based detergent use, and DIP loading of surface waters could bear some improvement.

[34] NEWS-DIP-HD somewhat underestimates DIP export and yield from Amazon subbasins for both the Amazon main stem and its tributaries (mean underestimate 43%; see Figures 2a and 2b). In addition, there is still a significant amount of unexplained variation. However, in general, NEWS-DIP-HD performs as well as the original NEWS-DIP model. The strong performance of NEWS-DIP-HD is encouraging but also curious, given that NEWS-DIP-HD was not recalibrated and there is no explicit in-stream loss pathway for DIP in the NEWS-DIP-HD aside from loss in reservoirs and loss due to consumptive water use. It may be that the half-degree resolution utilized by NEWS-DIP-HD is still coarse enough so that small-scale variation in DIP yield due to in-stream processing of P is averaged out. It may also be that DIP acts relatively conservatively because it is in dynamic equilibrium with particulate P in freshwater aquatic systems (as described by Froelich [1988]).

[35] In future global DIP export models it will be important to improve representation of reservoir retention. Including DIP sinks other than reservoirs and consumptive water use may also improve the model. For example, natural lakes, river-associated wetlands, and floodplains may account for significant levels of DIP retention, but are not treated explicitly by the NEWS-DIP-HD model. In addition, retention on land may also constitute an important DIP sink as reuse of sewage as fertilizer (night soil), conservation tillage practices, and highly weathered, P-deficient soils, and P-limited terrestrial (or aquatic) ecosystems all may lead to DIP retention within watersheds. At present, terrestrial sinks for DIP are represented in the model as sewage treatment, weathering efficiency, and fertilizer and manure transfer efficiency terms. This rather simple treatment of terrestrial P sinks results from a lack of more detailed global-scale input data, and it is possible that future inclusion of such data may improve the NEWS-DIP-HD model significantly. Though our analysis is currently limited to large (>5th order) rivers, this problem will most likely be solved incrementally as increasingly reliable, finer-resolution spatial data sets of model drivers become available.

[36] With improved resolution and quality of input and validation data sets and faster computers it will become possible to improve the spatial resolution of DIP export models even further so that it will be possible to examine P dynamics in even smaller river systems than has been possible in this analysis. The generation of improved input and validation data will likely lead to more accurate model predictions and additional insights. Higher-resolution validation data will facilitate the enhanced representation of DIP retention as well as the inclusion of interactions between elements and elemental forms. Also, as improved temporal resolution data sets of runoff and land use become available, it should be possible to use NEWS-DIP-HD to examine subannual patterns of DIP export. Incorporating such subbasin spatial and subannual temporal variability into global DIP export modeling efforts will constitute significant advances in understanding of the global P cycle and effects. It should also be possible to use the insights and approaches developed in the analysis associated with this effort to further enhance NEWS models that predict the transport of other biogeochemically relevant elements (e.g., N, silica, and carbon) and forms (e.g., dissolved, particulate, organic, and inorganic forms) of these elements. For the present, however, NEWS-DIP-HD represents a significant advancement in its own right as the first spatially explicit, global DIP export model with the capacity to route DIP downstream through watersheds, thereby maintaining within-basin spatial variability in P loading and P sinks.

Appendix A:: Variable Definitions for NEWS-DIP Model

[37] Several symbols are used to represent NEWS-DIP-HD model components in the article text. For the reader's convenience, these abbreviations are defined in Table A1.

Table A1. Variable Definitions for NEWS-DIP-HD Model
VariableDefinition
DIPDIP yield (kg P km−2 yr−1)
QactMeasured discharge after dam construction (km3 H2O yr−1)
QnatMeasured discharge prior to dam construction (km3 H2O yr−1)
DFraction DIP retained in reservoirs (0–1)
RRunoff (m H2O yr−1)
HHuman population density (individuals km−2)
PswPer capita DIP yield (kg P individual−1 yr−1)
AUnitless coefficient defining how nonpoint DIP and weathered DIP respond to runoff; for NEWS-DIP set equal to 0.85
BUnitless coefficient defining how nonpoint DIP and weathered DIP respond to runoff; for NEWS-DIP set equal to 2
WmaxMaximum DIP yield due to weathering alone (kg P km−2 yr−1); for NEWS-DIP set equal to 26
LmaxMaximum fraction of applied manure and fertilizer P lost to coastal zone as DIP; for NEWS-DIP set equal to 0.04
PfertP applied to watersheds as inorganic fertilizer (kg P km−2 yr−1)
PamP applied to watersheds as manure (kg P km−2 yr−1)
PsewDIP entering surface water from sewage (kg P km−2 yr−1)
PdetDIP entering surface water from P-based detergents (kg P km−2 yr−1)
DIPWeatheringDIP entering surface water from nonanthropogenic weathering of minerals (kg P km−2 yr−1)
DIPFertilizerDIP entering surface water from P-based fertilizers (kg P km−2 yr−1)
DIPManureDIP entering surface water as a result of manure production or application (kg P km−2 yr−1)

Appendix B:: Data Used for Model Evaluation

[38] Data used in the evaluation of NEWS-DIP-HD were collected from multiple sources. These sources and several important basin characteristics are characterized in Table B1.

Table B1. Data Used for NEWS-DIP-HD Evaluation
RiverLatitudeaLongitudeaUpstream Basin Areab (km2)Median Discharge (km3 yr−1)Median DIP (mg P L−1)Sourcea
Alabama30.5−88114,788.869.350.0104
Altamaha31.6544−81.828136,260.011.250.0264
Amazon0.1−496,112,000.06,590.000.0204
Amguema68.1−177.429,600.09.200.0124
Amur53.1140.441,855,000.0344.000.0214
Anabar72114.179,000.013.300.0014
Apalachicola29.67−84.9751,451.329.200.0064
Arkansas35.25−94.25389,914.937.740.0383
Arkansas34.75−92.25409,964.049.540.0343
Arkansas37.25−97.25113,216.11.670.4143
Arkansas38.25−102.2565,811.60.260.0243
Arkansas34.25−91.75415,628.375.780.0313
Arkansas38.25−105.2510,422.10.880.0293
Arkansas36.75−96.75141,063.77.480.0743
Arkansas36.25−96.25193,252.08.040.0883
Balsas17.55−102.1112,000.014.000.0954
Barito−3.32114.2966,000.086.800.0054
Bei Jiang23.25112.25343,513.0111.350.0042
Bug47.3330.4763,700.03.400.0974
Cauweri10.4579.588,000.020.900.1004
Chao Phrya13.44100.3111,435.027.800.0264
Chiang Jiang32.06121.041,808,000.0928.000.0204
Churchill (Hud.)58.47−94.12298,000.025.830.0064
Colorado (Ari)32.44−114.38639,000.00.100.1024
Columbia46.12−123.5669,000.0236.000.0154
Connecticut41.9872−72.605825,019.415.740.0204
Dalalven60.3817.2725,000.09.840.0024
Danube47.2520.2570,222.614.910.0872
Danube47.7519.25182,903.073.700.1952
Danube  805,000.0201.250.1834
Daugava56.5324.0887,900.020.400.0374
Dnepr46.332.18504,000.053.400.0364
Dnestr46.130.1972,100.010.700.0564
Don47.2540.25391,664.030.330.1332
Don47.1539.45422,000.020.700.0424
Drammenselva59.4410.1517,000.08.100.0024
Eastmain52.15−78.0546,400.028.200.0224
Ebro42.25−2.2511,386.42.510.1372
Ebro40.820.5284,000.09.240.1154
Elbe53.59146,000.023.700.3904
Evros40.8826.1755,000.06.800.2804
Fraser49.23−121.27220,000.0112.000.0504
Fuchun Jiang30.18120.0754,349.037.300.0464
Gambia13.31−14.542,000.04.900.0154
Ganges24.0589.021,050,000.0493.000.0754
Ganges-Brahmaputra-Meghna25.2589.75548,939.0401.311.0722
Ganges-Brahmaputra-Meghna24.2590.7559,040.8104.020.7432
Garonne44.250.1455,000.017.200.1044
Glama59.3611.0741,200.019.900.0084
Grijalva17.75−92.755,896.46.000.1082
Grijalva18.36−92.3936,400.023.000.0854
Guadiana38.75−6.7540,779.52.320.2632
Guadiana37.13−7.2472,000.09.000.0574
Hunter−32.9151.821,411.40.470.0624
Huang He37.44118.36752,000.041.000.0204
Hudson  34,700.017.700.0604
Illinois41.25−88.7521,390.77.830.2783
Illinois39.75−90.7569,264.027.790.1323
Indigirka69.6147.5305,000.050.400.0054
Indus31.2573.7534,028.91.070.4222
Indus25.2368.24916,000.057.000.5204
Kamchatka56.14162.2855,900.033.100.0754
Khatanga72.55106364,000.085.300.0064
Klamath41.5144−123.999231,339.014.550.0184
Kolyma68.7158.7526,000.070.800.0064
Kuban45.1637.2457,900.013.400.0304
Kymjoki60.326.5237,200.09.680.0104
La Plata−25.75−54.75931,200.0400.390.0312
La Plata−27.75−58.752,177,480.0557.850.0322
La Plata−26.75−58.25994,251.0120.420.0372
La Plata−32.75−60.752,612,380.0589.280.0522
La Plata−32.25−58.25264,486.0143.030.0452
La Plata−31.25−57.75243,471.0138.150.0142
La Plata−30.25−57.75219,576.0131.150.0072
Lena70.7127.42,430,000.0532.500.0044
Liao40.5121.48219,000.016.200.0534
Loire47.16−2.11112,000.026.000.0904
Luan39.2119.154,000.04.200.0124
Mackenzie68.16−133.41,787,000.0308.000.0044
Magdalena11.06−74.51235,000.0237.000.1204
Mahanadi30.386.7552,094.450.610.0384
Manacapuru−3.25−60.752,235,312.03,195.350.0221
Mekong11.25105.25746,412.0398.600.0222
Meuse51.495.0129,000.010.200.2304
Mezen6545.656,000.020.400.0234
Mino42.25−7.7511,404.75.760.0192
Mississippi29.9508−90.13812,916,081.0527.950.0844
Mississippi33.25−91.252,928,240.4664.840.0703
Mississippi41.75−90.25221,703.068.840.0633
Mississippi40.25−91.25308,208.680.100.1083
Mississippi35.25−90.252,415,940.8471.860.0613
Mississippi44.75−92.7595,829.621.100.1183
Mississippi37.75−89.751,847,179.4239.500.0893
Mississippi32.75−91.252,953,881.3643.380.0543
Mississippi44.25−91.75153,327.360.030.0503
Mississippi38.75−90.25444,182.985.740.0723
Mississippi39.25−90.75443,664.9154.780.0883
Mississippi45.75−94.2530,043.94.760.0133
Mississippi30.75−91.252,924,873.4507.230.0653
Missouri44.25−100.25630,662.119.970.0123
Missouri47.75−110.7564,099.66.260.0113
Missouri43.25−98.75682,461.822.250.0123
Missouri47.25−101.25469,823.817.850.0103
Missouri38.75−91.251,357,671.778.860.0843
Missouri41.25−95.75846,301.932.600.0543
Missouri42.25−96.25825,064.027.790.0323
Missouri39.25−94.751,088,572.037.410.0733
Missouri46.25−111.7537,992.53.830.0133
Missouri47.75−110.2589,041.25.480.0153
Missouri47.75−106.75149,069.38.890.0123
Missouri48.25−104.25237,131.57.680.0163
Missouri47.75−108.75106,155.86.880.0133
Murray−35.22139.221,060,000.07.900.0244
Musi−2.2104.5656,700.080.400.0304
Nadym65.672.748,000.014.600.1284
Narva59.7530.75278,346.069.910.0382
Nelson57.04−92.31,132,000.089.260.0044
Nelson-Saskatchewan53.75−101.25390,198.030.810.0122
Nemanus55.0221.598,200.017.200.0464
Neva59.4830.43282,000.080.400.0304
Ob66.666.62,950,000.0404.100.0734
Odra53.2514.32112,000.016.600.3704
Ohio38.25−82.75160,579.396.740.0103
Ohio38.75−85.25215,409.367.000.0453
Ohio40.25−80.7564,931.038.510.0163
Ohio37.25−89.25526,026.6244.420.0353
Olenek71.8123.6198,000.031.500.0044
Onega63.838.512,000.015.700.0084
Orinocco8.37−62.151,100,000.01135.000.0104
Panuco21.59−98.3466,300.017.300.0164
Paraiba Do Sul−21.45−41.257,000.030.600.0104
Parana−34−58.172,783,000.0568.000.0454
Pechora67.652.2312,000.0135.100.0344
Peel67.37−134.471,000.024.500.0064
Penzhina62.28165.1871,600.022.800.0214
Po44.5311.3970,000.048.900.0754
Potomac38.9294−77.117229,966.318.740.0364
Purari−7.25145.0530,580.084.130.0024
Red31.25−91.75241,291.170.530.0333
Red33.75−96.75102,874.34.380.0203
Red33.75−94.25124,397.115.500.0173
Red33.75−97.2579,725.07.950.0683
Red34.25−98.7553,276.11.690.0173
Rhine47.259.756,334.610.750.0092
Rhine47.758.2533,612.731.820.0392
Rhine47.757.7535,691.133.040.0602
Rhine51.526.02224,000.068.560.4004
Rhone46.256.758,550.712.870.0142
Rhone46.256.2510,688.313.700.0382
Rhone43.924.6795,600.053.900.1014
Rio Grande27.75−105.2532,964.90.540.2532
Rio Grande (United States)25.8764−97.4542456,702.50.530.0304
Rio Ica−2.75−68.25157,576.3183.730.0171
Rio Japura−1.75−65.75222,517.8456.860.0121
Rio Jurua−3.25−66.25193,108.0247.970.0261
Rio Jutai−3.25−67.2561,586.0143.680.0261
Rio Madeira−3.75−59.251,454,335.0860.330.0191
Rio Negro−2.75−60.75608,636.4791.740.0201
Rio Purus−4.25−61.75360,990.0431.920.0341
Rufiji−7.4837.55178,000.035.200.0104
Sacramento38.35−121.370,000.020.500.0304
Sacramento40.75−122.2516,752.06.560.0153
Saint Lawrence  1,025,000.0338.250.0464
Sakarya40.4530.2355,300.05.870.1604
San Joaquin37.75−121.2535,058.12.830.1343
Santo Antonio do Ica−3.25−67.751,193,550.01748.670.0281
Schelde50.753.2511,797.71.690.8812
Scheldt51.224.1511,400.06.000.8104
Seine49.260.2678,600.015.800.4004
Severnaya Dvina64.141.9348,000.0105.600.0154
Seyhan36.4334.5319,300.04.800.0104
Skagit48.445−122.33428,010.914.340.0104
Stikine56.7081−132.130351,592.867.690.0214
Susquehanna40.15−76.5271,000.034.000.0084
Swan Canning−32115.9126,021.01.310.0604
Tagus39.75−3.7523,488.20.701.5722
Tana−2.3240.3142,000.04.750.0404
Tejo38.44−9.0876,200.09.600.1484
Tennessee36.75−88.25104,454.227.250.0223
Tennessee35.25−88.2585,003.442.440.0213
Tennessee34.75−85.7558,637.323.770.0133
Tennessee35.75−84.7544,832.714.010.0133
Tocantins−2.12−49.3757,000.0372.000.0034
Tornionjoki65.4824.0839,500.011.860.0044
Tugela−29.2230.530,112.01.100.0514
Uruguay−33.55−58.22240,000.0145.000.0374
Ususmacinta17.25−91.347,700.055.520.0854
Vargem Grande−3.25−68.251,032,887.01449.330.0281
Volga54.7555.75109,867.013.790.0712
Volga  1,350,000.0256.500.0114
Weser53.328.3445,800.010.600.3704
Wisla54.0618.47198,000.032.500.2104
Yana70.8136224,000.032.200.0064
Yellowstone45.75−108.7530,548.95.510.0123
Yellowstone45.75−110.759,197.03.070.0143
Yellowstone47.75−104.25178,975.99.720.0143
Yenisey52.25106.75444,159.033.290.0182
Yenisey50.25103.757,844.22.750.0102
Yenisey67.2586.752,442,750.0392.120.0382
Yenisey69.286.52,440,000.0577.300.0094
Yesil41.2436.3535,960.05.670.0804
Yukon62.39−164.48849,000.0200.000.0104
Zaire−6.0412.243,698,000.01200.000.0244
Zambezi−18.5536.041,330,000.0106.000.0104
Zhujiang22.4113.05437,000.0363.000.0144

Acknowledgments

[39] We are grateful to UNESCO-IOC, NASA, USGS, and CALFED for supporting this work and to Charlie Vörösmarty, Balazs Fekete, Wil Wollheim, and the rest of the Global NEWS working group for useful discussion and feedback. This work was supported by grants to J. A. Harrison from California Sea Grant (award RSF8), from the U.S. Geological Survey 104b program, and from the NASA-IDS program (award 06-IDS06-009). However, any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of NASA, USGS, CALFED, or other funding agencies.

Ancillary