Purpose and Scope

The purpose of this report is to present a quantified understanding of current nutrient concentrations and loads in rivers and streams across the State of Mississippi and to examine the variability of concentrations and loads spatially, seasonally, between years, and in response to streamflow. Specifically, this report summarizes concentrations of TN and TP and provides estimates of TN and TP loads and yields using continuous streamflow data measured by the USGS and discrete water-quality samples collected at 22 stream locations across the State of Mississippi. Water-quality data were screened and analyzed for the period of record during water years (WYs) 2008–18. This information can augment efforts to:

Description of Study Area and Study Sites

Sites selected for summary include 20 water-quality monitoring stations operated by MDEQ and 2 water-quality monitoring stations operated by USGS. The USGS operated streamflow gages at all 22 sites (fig. 1; table 1). The MDEQ stations are part of their ambient water-quality monitoring network, which is a component of their Surface Water Monitoring Program (SWMP) that has been maintained and implemented for decades to address mandates of the Federal Clean Water Act under Sections 303(d), 305(b) and 314 of the Clean Water Act (U.S. Code 33 U.S.C. §1251 et seq.) to collect data used to assess the quality of all waters within Mississippi. The two USGS stations are part of a long-term National Water-Quality Assessment (NAWQA) Project (U.S. Department of the Interior, U.S. Geological Survey, 1999) located in the Mississippi Delta.

Mississippi ranks 32nd in the United States in land size, with 125,438 km² of total area (U.S. Census Bureau, 2012). Mississippi is bordered by four States: Alabama, Tennessee, Arkansas, and Louisiana. The western border is formed by the Mississippi River, and the southern border is formed by the Gulf of Mexico.

Natural and human factors, such as population, physiography, geology, climate, hydrology, land uses, and land cover types, collectively affect the physical, chemical, and biological condition of surface water in Mississippi. These factors and their combinations may influence patterns and associations of nutrient conditions across streams and rivers in Mississippi.

Population

Mississippi ranks 32nd in the United States in population with 2.97 million people in 2010, a 4.3-percent increase from 2.84 million in 2000, and a population density of 24 persons per km² (U.S. Census Bureau, 2012). The 2010 census classified 49.3 percent of the population as urban and 50.7 percent as rural. Approximately 40 percent of the population of Mississippi is concentrated in 9 of the 82 counties, located in 4 spatial urban clusters: DeSoto County in northwest Mississippi; Hinds, Rankin, and Madison Counties in central Mississippi; Forrest and Lamar Counties in southeast Mississippi; and Harrison, Jackson, and Hancock Counties in south Mississippi (fig. 2). The highest growth rates during 2000–10 were in DeSoto (50.4 percent), Lamar (42.5 percent), Madison (27.5 percent), and Rankin (22.8 percent) Counties. The lowest growth rates were in the Mississippi Delta in Issaquena (−38.2 percent) and Sharkey (−25.2 percent) Counties.

Physiography and Geology

Mississippi lies within the Coastal Plain Physiographic Province (Fenneman, 1938) and includes a wide variety of landscapes. It is flat and relatively featureless in some areas, but elsewhere it consists of rounded and eroded hills and floodplains of large rivers. The Coastal Plain developed on Mesozoic sedimentary rock. The rocks are mainly composed of chalk, sandstone, limestone, and claystone. The unconsolidated sediment units, composed mainly of sediments, are described as gravels, sands, silts, and clays.

Ecoregions are delineations of areas with similar climate, geology, soils, vegetation, topography, and hydrology (Omernik and Griffith, 2008). They are intended to serve as a framework for research, assessment, management, and monitoring of ecosystems and ecosystem components. Mississippi is composed of parts of four level III ecoregions (fig. 3). The study sites are in 3 of the 4 level III ecoregions: Southeastern Plains (15 sites), Mississippi Alluvial Plain (4 sites), and Mississippi Valley Loess Plains (3 sites) (EPA, 2021).

Land Cover

The 2016 National Land Cover Database (NLCD) was used to identify land cover and use throughout Mississippi (Yang and others, 2018; Homer and others, 2020). The 2016 NLCD is a 16-class, 30-meter-resolution dataset primarily based on a combination of Landsat data and geospatial ancillary data, combined with training datasets as well as decision-tree classifications. The 2016 NLCD provides an updated dataset of land cover data for 2001, 2003, 2006, 2008, 2011, 2013, and 2016. The 2013 land cover dataset with eight aggregated 2016 NLCD Anderson Level I classes (Anderson and others, 1976) was used in this study (table 2). For example, the land cover classes “Deciduous forest,” “Evergreen forest,” and “Mixed forest” were aggregated to “Forest” (table 2). Land cover in Mississippi is primarily forest and agricultural (fig. 4; table 3), with nearly half of the State covered by forests and about a fourth of the State covered by agriculture lands. Only 6.17 percent of Mississippi is classified as developed. The percentages of land cover for the drainage areas of the 22 study sites were calculated by using the 2016 NLCD dataset, with Anderson Level I classifications (table 4) (Homer and others, 2020).

Table 2.    

Classification scheme of Anderson Level I and Level II systems used with the 2016 National Land Cover Data dataset (Anderson and others, 1976).

Table 2.    Classification scheme of Anderson Level I and Level II systems used with the 2016 National Land Cover Data dataset (Anderson and others, 1976).

2013 National Land Cover Data
Level I	Level II
1 Water	11 Open water
2 Developed	21 Developed open space
	22 Developed, low intensity
	23 Developed, medium intensity
	24 Developed, high intensity
3 Barren	31 Bare rock/sand/clay
4 Forest	41 Deciduous forest
	42 Evergreen forest
	43 Mixed forest
5 Shrub	52 Shrub/brush
6 Grassland/herbaceous	71 Grassland/herbaceous
7 Agriculture	81 Pasture/hay
	82 Cultivated crops
8 Wetlands	90 Wooded wetlands
	95 Emergent wetlands

Table 3.    

Percentage of land use/land cover for Mississippi for 2013 from the 2016 National Land Cover Data dataset (Homer and others, 2020) using Anderson Level I classifications (Anderson and others, 1976).

[km², square kilometer]

Table 3.    Percentage of land use/land cover for Mississippi for 2013 from the 2016 National Land Cover Data dataset (Homer and others, 2020) using Anderson Level I classifications (Anderson and others, 1976).

Land cover and land use class	Area (km2)	Percentage
1 Water	2,671,603	2.21
2 Developed	7,454,323	6.17
3 Barren	267,342	0.22
4 Forest	50,963,648	42.2
5 Shrub	5,322,886	4.41
6 Grassland/herbaceous	3,826,318	3.17
7 Agriculture	32,113,306	26.6
8 Wetlands	20,835,275	17.3

Table 4.    

Percentage of land use/land cover from the 2016 National Land Cover Data dataset (Homer and others, 2020) using Anderson Level I classification (Anderson and others, 1976) for areas draining selected streamflow and water-quality monitoring stations from the Mississippi Department of Environmental Quality (MDEQ) ambient water-quality monitoring network and the U.S. Geological Survey (USGS) National Water-Quality Assessment (NAWQA) Project.

[<, less than]

Table 4.    Percentage of land use/land cover from the 2016 National Land Cover Data dataset (Homer and others, 2020) using Anderson Level I classification (Anderson and others, 1976) for areas draining selected streamflow and water-quality monitoring stations from the Mississippi Department of Environmental Quality (MDEQ) ambient water-quality monitoring network and the U.S. Geological Survey (USGS) National Water-Quality Assessment (NAWQA) Project.

Site number (fig. 1)	USGS or MDEQ1 station number	Short name	1 Water	2 Developed	3 Barren	4 Forest	5 Shrub	6 Grassland/ Herbaceous	7 Agriculture	8 Wetlands
			Percent
Tombigbee River Basin
1	02436500	Town	1.44	12.0	0.07	35.1	2.32	1.42	43.0	4.71
2	02439400	Buttahatchee	0.22	5.39	0.07	62.3	8.81	3.01	10.8	9.36
3	02443000	Luxapallila	0.43	6.16	0.12	56.0	7.39	4.39	11.7	13.8
Pascagoula River Basin
4	02476600	Okatibbee	3.22	12.3	0.14	43.9	10.8	4.88	9.88	15.0
5	02477000	Chickasawhay upper	0.63	4.76	0.01	52.8	9.27	4.71	10.7	17.1
6	02478500	Chickasawhay lower	0.91	4.53	0.09	49.5	10.1	5.12	7.84	21.9
7	02479300	Red	0.91	4.99	0.26	41.9	9.81	4.15	10.3	27.7
8	02479560	Escatawpa	0.64	3.27	0.07	47.8	8.58	6.36	7.92	25.4
Pearl River Basin
9	02482000	Pearl upper	0.80	5.37	0.03	47.4	3.86	2.61	18.9	21.0
10	02486500	Pearl lower	2.12	10.2	0.03	46.7	3.29	2.79	15.4	19.5
11	02487500	Strong	0.57	4.58	0.01	56.3	4.22	3.28	16.0	15.0
12	02490900	Bogue Chitto	0.38	6.92	0.07	49.2	6.00	4.03	23.9	9.56
Tennessee River Basin
13	03592100	Bear	1.37	7.07	0.21	54.1	5.64	2.36	26.9	2.39
Yazoo River Basin
14	07268000	Little Tallahatchie	0.76	6.28	0.08	45.6	3.81	1.70	35.6	6.19
15	07287120	Yazoo upper	3.14	5.04	0.12	38.4	2.71	1.17	39.6	9.87
16	07288500	Big Sunflower	1.73	5.57	0.06	0.07	<0.01	0.01	82.3	10.3
17	07288650	Bogue Phalia2	0.59	3.95	0.03	0.01	0.01	<0.01	87.7	7.75
18	07288955	Yazoo lower2	3.51	4.29	0.28	13.8	0.36	0.14	58.4	19.2
Big Black River Basin
19	07290000	Big Black	1.10	4.95	0.05	50.3	3.41	2.71	24.9	12.6
South Independent Streams Basin
20	07290650	Bayou Pierre	0.58	3.58	0.17	60.3	6.37	4.26	17.5	7.20
21	07375280	Tangipahoa	0.88	11.4	0.10	46.5	5.81	3.87	21.3	10.1
Coastal Streams Basin
22	111B59	Tuxachanie	0.17	3.81	0.11	51.8	8.86	3.12	2.26	29.8

¹

All site identifiers are USGS site identifiers, except 111B59 which is an MDEQ site identifier.

²

USGS NAWQA Project monitoring station.

Station Selection, Period of Record, and Data Screening

Station selection for loads and yields was based on the availability of long-term water-quality data at a station. Station selection was carried out with assistance from MDEQ staff. Twenty of the 40 current MDEQ ambient water-quality monitoring network stations were chosen for inclusion in this study. Remaining stations were excluded based on an insufficient amount of water-quality data and (or) the ability to associate the water-quality site with a suitable nearby streamgage. The two stations from the USGS NAWQA long-term trend network (USGS, 2019) that were selected for this study are the Bogue Phalia (site number [site no.] 17) and Yazoo lower (site no. 18) (fig. 1).

Streamflow data and monthly water-quality data from the selected sites were investigated, retrieved, compiled, and evaluated for the period 2008–18. All data were retrieved and analyzed by WY. A WY is defined as the 12-month period from October 1 through September 30 and is designated by the calendar year in which the WY ends. Streamflow and water-quality data were retrieved from the USGS National Water Information System database (USGS, 2020); monthly water-quality data were provided by the MDEQ and are available in a companion data release (Crain and others, 2023). During some years, samples were not collected each month for a particular station. A summary of missing data is provided in table 5.

The values used in the load calculations were concentrations of unfiltered total nitrogen as N (TN) and unfiltered total phosphorus as P (TP), both expressed in units of milligrams per liter. If a TN concentration was not available, it was estimated by using the analytic results for dissolved nitrate plus nitrite as N (NO₃+NO₂) and results for total Kjeldahl nitrogen as N (TKN), which includes organic N and ammonia. If neither NO₃+NO₂ nor TKN concentrations were less than the detection limits, the TN concentration was simply their sum. If the concentration of TKN was less than the method detection limit of 0.1 milligram per liter (mg/L) or the concentration of NO₃+NO₂ was less than the method detection limit of 0.02 mg/L, the following rules were applied to estimate concentrations of TN:

Note that a positive bias in a TN concentration is likely when the concentrations of TKN and NO₃+NO₂ are summed because of the reduction of nitrate to ammonium during the Kjeldahl digestion process. This process causes some double counting of NO₃+NO₂ concentrations in the analytical results for TKN concentrations and in the calculated sum of TN (Mueller and Spahr, 2006).

For summaries and analyses of concentrations in this report, TN and TP values less than the detection limit are given as the detection limit and are referred to as “censored values.”

Estimates of Daily Streamflow

Daily mean streamflows were estimated at 7 of the 22 water-quality sampling sites where no streamflow data were available or where the available streamflow record was incomplete (missing) for a part of the water-quality sampling period in WYs 2008–18 (table 6). Streamflow data were not available for two of the water-quality sampling sites; the available streamflow record was incomplete for the remaining five sites. Streamflow was estimated by using either a drainage-area-ratio adjustment or regression methods. For sites lacking streamflow records that were located on the same stream reach (upstream or downstream) or near a streamgage, the drainage-area-ratio adjustment was used to estimate the streamflows at the nearby ungaged location. This method is commonly used when the drainage-area ratio between the two sites is between 0.5 and 1.5 (Risley and others, 2008).

Extension in time (augmentation) of an existing short-term streamgage record (a short-term station) was accomplished by developing a regression relation among concurrent daily mean streamflows at the short-term stations and at one or more nearby stations that are hydrologically similar and have streamflow records through the sampling period (long-term stations). The final regression streamflow estimates were made by using the Maintenance of Variance Extension Type 1 (MOVE.1), as described by Hirsch (1982), in which a correlation regression model (Helsel and others, 2020) was fit to the concurrent daily mean streamflows (SAS Institute, Inc., 2004). The MOVE.1 method provides the estimated streamflows with some of the variance characteristics of the observed streamflow record. The MOVE.1 method is unlike ordinary-least-squares regression, which tends to “smooth” the estimated streamflow and provides lower variance than in the original observed streamflows. The MOVE.1 regression estimate was computed by use of log-transformed values of the concurrent nonzero daily mean streamflows as

Missing time-series streamflow values were estimated by linear interpolation between known values. The linear interpolation formula is defined as:

Load Estimation Methods

A load, also referred to as “mass flux,” is the transport (mass discharge) of a constituent passing a given point in a stream over a specific time period. A load is determined by multiplying the concentration of the constituent by the streamflow. A yield is defined as the load per unit of drainage area and is expressed as kilograms per year per square kilometer.

A variety of quantification methods are available to estimate long-term, mean annual loads. Lee and others (2016) evaluated 11 methods for computing mean annual water-quality loads for various constituents, including TN and TP. Their findings concluded that ratio estimators and Weighted Regressions on Time, Discharge, and Season (WRTDS) allowed for flexibility in determining concentration and streamflow relations and produced the most accurate loads (Lee and others, 2019). Six of the load estimation methods considered and evaluated in this study were based on simple interpolation, ratio estimators, multiple regression, and weighted regression. However, only three of the six methods were chosen to estimate mean annual nutrient loads for this study. The three methods are the Beale’s ratio estimator (BRE), the USGS LOAD ESTimator (LOADEST) seven-parameter regression estimator, and WRTDS.

The BRE method is a simple, data-driven method that is applied by delineating different strata based on either time or streamflow conditions within the sampled record (Beale, 1962; Tin, 1965; Lee and others, 2016). This method has been widely used for estimating loads, particularly for P, at sites in the Great Lakes (Maccoux and others, 2016) and has been shown to produce statistically unbiased results (Richards and Holloway, 1987; Quilbé and others, 2006; Lee and others, 2019).

The USGS LOADEST method computes regression coefficients using the adjusted maximum likelihood estimation method to develop regression relations (Wolynetz, 1979) that relate infrequently available concentration data to various continuous explanatory variables (for example, daily mean streamflow, season, and time) to estimate their effects on water-quality constituent concentrations and loads (Runkel and others, 2004). LOADEST, a FORTRAN program originally developed by Runkel and others (2004), was updated and repackaged as rloadest by Lorenz and others (2015) in the R programming language (R Core Team, 2018).

For this study, the R package loadflex was used to fit and assess nutrient load models and to compare uncertainty in the mean annual load estimate predictions. The loadflex package (available at https://github.com/mcdowelllab/loadflex) facilitates the implementation and comparison of five load estimation models using the same interface (Appling and others, 2015). This package allows the user to efficiently compare multiple load estimation models using model selection criteria to determine the best fitting model. The loadflex package also estimates the uncertainty in load estimations which is important for interpretation (Yanai and others, 2015). For this effort, the loadflex package estimated mean annual loads using both the BRE and LOADEST regression methods from a single dataset. The BRE method was implemented in the stratified form as described in Cochran (1977).

The WRTDS method creates flexible statistical relations using a weighted regression for each day of the estimation period to produce four daily time series for the period of record (Hirsch and De Cicco, 2015). These daily time series include daily concentration, daily load, flow-normalized daily concentration, and flow-normalized daily load. The nonnormalized estimates for daily concentration and load as reported in this study are estimates of the observed daily occurrence as determined by the observed daily streamflow (Hirsch and De Cicco, 2015). Flow-normalized daily concentration and load estimates are calculated by using the mean streamflow for a given day. Flow-normalized calculation reduces the effects of streamflow variability on concentrations and loads and enhances the WRTDS method’s ability to identify trends and protects the analysis from being highly influenced by the random occurrence of a particularly wet or dry year near the end or the beginning of a study period (Sprague and others, 2011).

For this study, no single load estimation method provided the greatest accuracy for all water-quality constituents, sites, and sampling scenarios. Selection of a load estimation method used for each constituent and at each site was based primarily on the degree to which the model fit the data and the use of the flux (load) bias statistic; a dimensionless statistic defined as

Values of the flux bias statistic between −0.1 and +0.1 indicated an acceptable model; values beyond that range indicated an unacceptable model (Hirsch, 2014). Additionally, a variety of graphics were used to identify and characterize issues with the fitted model for each given dataset. A function for producing eight diagnostic graphs is included in the WRTDS distribution (R package EGRET; Hirsch and De Cicco, 2015).

Of the methods considered in this study, the BRE and WRTDS methods were the most consistent in providing accurate and least biased mean annual load estimates of TP. According to Lee and others (2016), methods that create flexible statistical relations between streamflow and load, such as the BRE and WRTDS methods, generally provide more accurate load estimates, particularly for TP which is a constituent that exhibits a strong positive relation between concentration and streamflow. Results of the comparisons of multiple methods considered for estimating mean annual TN loads in this study indicated that the LOADEST regression and WRTDS methods provided more accurate and least biased estimates of TN. Both methods are regression based and work well for TN because this constituent exhibits little variation between concentration and streamflow (Lee and others, 2016).

Trend-Estimation Methods

Trends in streamflow were assessed in Exploration and Graphics for RivEr Trends (EGRET) by using the Mann-Kendall test with Theil-Sen slope estimates for the period of record (WYs 2008–18) (Helsel and others, 2020). The tau value determined in the Mann-Kendall analyses indicated whether the annual mean daily streamflow trend was upward or downward. Values of tau may range from +1 to −1. A value of +1 indicates continuous increase in streamflow with time, and a value of −1 indicates continuous decrease in streamflow with time. Values of tau reflect the relative strength of the monotonic association between streamflow and time. Monotonic streamflow trend slopes were computed by using the Theil-Sen slope estimator to identify the median slope associated with the trend (Helsel and others, 2020). Monotonic streamflow trend slopes are calculated on the logarithms of the streamflow data and then transformed to percentage changes per year (Hirsch and De Cicco, 2015).

The Exploration and Graphics for RivEr Trends confidence intervals (EGRETci) package (Hirsch and De Cicco, 2019) was used to calculate trends in nutrients and to estimate the uncertainty associated with these trends. A block bootstrap approach contained in the EGRETci package was used to generate confidence intervals to determine the magnitude, direction, and statistical likelihood of trends during WYs 2008–18. The trend results in this report have been summarized by using a likelihood-based approach that uses descriptive statements to describe the likelihood associated with the trend direction over time (Hirsch and others, 2015). This approach is often used when reporting water-quality trends (Oelsner and others, 2017) and is an alternative to the null-hypothesis significance testing approach. The likelihood-based approach provides more straightforward information on the certainty of an estimated trend and gives water-resource managers more useful information when making decisions on basin management (Oelsner and others, 2017). When the likelihood statistic was between greater than or equal to 0.95 and less than or equal to 1.0, the trend was described as “highly likely” for an upward trend (or “highly unlikely” for a downward trend if one wanted to describe the trend likelihood in this manner). Conversely, a trend with a likelihood statistic between greater than or equal to 0.0 and less than or equal to 0.05 was considered “highly unlikely” for an upward trend (or “highly likely” for a downward trend).

Data Analysis Methods for Comparing Streamflow, Water Quality, and Land Use

The Spearman’s rank correlation test (Helsel and others, 2020) was used to determine monotonic relations between streamflow, water-quality conditions, and land use. Spearman’s rank correlation test is a nonparametric test that determines correlations between the ranks of two variables and does not require the variables to be linearly related. Spearman’s correlation coefficient, rho or Spearman’s ρ, ranges between −1 and 1, with values close to these endpoints representing the strongest correlative relations (either negative or positive, respectively) and values near zero representing poor correlative relations.

Hydrology

Environmental factors such as precipitation and streamflow may influence concentrations and loads of nutrients in a basin. Previous studies have found that the transport of nutrients to streams is associated with high-flow events (Zhang and others, 2009; Carpenter and others, 2017; Gao and others, 2018), particularly extreme precipitation events (for example, precipitation exceeding 51 millimeters in 24 hours [Carpenter and others, 2017]). In addition, the influence of extreme precipitation events may result in skewed temporal distributions of nutrient loads. Thus, understanding hydrologic variations over time is important for assessing water-quality conditions and for providing a context to evaluate loads, yields, and trends.

Average annual and monthly precipitation data from three NOAA meteorological stations were used to characterize the relation between precipitation and streamflow across the State. The three selected stations represent the southern (USW00013833, Hattiesburg Chain Municipal Airport), central (USW00063831, Newton 5 ENE), and northern (USW00093862, Tupelo Regional Airport) regions in Mississippi (NOAA, 2021). Changes in the amount of precipitation appeared to generally influence the variability in annual median streamflow during the period of record, WYs 2008–18 (fig. 5A, B). Station-by-station analyses of precipitation data from nearby weather stations and streamflow were not performed; at some sites, streamflow could be responding to factors other than precipitation, such as water release from dams and irrigation return flow. Comparison of the mean annual precipitation (WYs 2008–18) from the combined data of the three meteorological stations to the combined long-term precipitation mean (1981–2010) at the same stations indicated about the same number of years with mean annual precipitation less than and greater than the long-term precipitation annual mean of 55.9 inches. Mean annual precipitation was greater than the long-term precipitation annual mean in 2009, 2012, 2013, 2016, and 2018.

Mean streamflow ranged from 5.06 cubic meters per second at the Tuxachanie (site no. 22) site to 489 cubic meters per second at the Yazoo lower (site no. 18) site (table 7; fig. 6). Average annual mean streamflow was highly related to drainage area (fig. 7). The highest average annual median streamflows of all study sites combined occurred in WYs 2010, 2013, and 2016, and the lowest occurred in WYs 2008, 2011, and 2017 (fig. 5A). The highest average monthly median streamflows occurred in the winter and spring (December–May), and the lowest occurred in the summer and fall (June–November) (fig. 5B).

Flow-duration curves (FDCs) are a useful tool in characterizing water-quality concerns and describing patterns associated with impairments. FDCs provide the ability to associate water-quality concerns to basin processes that may be important when developing watershed-based plans. Analysis between FDCs from different sites can provide some insight into the factors that may affect the movement of water in a drainage basin. The shape of the FDC reflects the ability of a basin to temporarily store water in the ground and to release it later as contributing flow (Leopold, 1994). Factors that affect the hydrologic characteristics of a basin include climate, geology, topography, land cover, and soil properties. These hydrologic characteristics can be inferred based on the slope of the curve. Drainage basins with a large amount of impervious or agricultural areas tend to have a steep slope at the lower end of the FDC, which is characteristic of a highly variable hydrologic system where streamflow is largely driven by direct runoff and there is little groundwater contribution to the stream (fig. 8A). In contrast, streams that have ample groundwater capacity tend to have flat slopes because of substantial groundwater contribution to base flow (fig. 8B). Annual FDCs were prepared from streamflow data for the 10-year period (2008–18) for each site (appendix 1).

In addition to FDCs, streamflow conditions for the sites in Mississippi were assessed by using measured and estimated daily streamflow data (table 7 and fig. 6). Quartiles were applied to describe streamflow. The first quartile, denoted as the 25th percentile, describes a value that exceeds no more than 25 percent of the streamflow data and hence is exceeded by no more than 75 percent of the data. Similarly, the third quartile, denoted as the 75th percentile, is a value that exceeds no more than 75 percent of the streamflow data and hence is exceeded by no more than 25 percent of the data. The 50th percentile (median) is the streamflow that is exceeded by 50 percent of the streamflow data over the analyzed period of record. The range of streamflows between the 25th and 75th percentiles, or interquartile range, represents 50 percent of the annual streamflow durations and indicates the statistical dispersion of the data. To better understand hydrologic context in which water-quality samples were collected, the daily mean streamflow at the time of water-quality sample collection was compared to five different flow-duration intervals for each site (table 8). The results indicated that water-quality samples were evenly collected over a wide range of flows at all sites, particularly at flow-duration intervals ranging from 11 to 90 percent. This finding suggests that load estimations are representative of a wide range of flows at all sites.

Water Quality

Data presented in this section characterize the TN and TP concentrations measured in samples from the 22 selected water-quality network stations in Mississippi from WYs 2008 to 2018 (tables 9 and 10). These data provide the basis for analysis of concentrations and estimation of mean annual loads and yields at the selected stations.

TN and TP Concentrations

Overall, streamflow at the time of sampling showed a significant (p-value of less than 0.01; linear model) positive correlation with TN and TP concentrations (R² = 0.4949, TN; R² = 0.6216, TP). The relation between TN and TP concentrations and streamflow is site-specific, however, with a positive relation at some sites, a negative relation at other sites, and no relation at some sites (figs. 9 and 10). The relation between concentration and streamflow may be more pronounced in streams that are influenced by developed land uses or wastewater treatment facilities. Streams in drainage basins with point-source discharges often have higher nutrient concentrations at low flows, when point-source effluent constitutes a large percentage of streamflow, and often have lower nutrient concentrations at high streamflow, when point-source effluent is diluted by storm runoff. These relations are more pronounced the closer the sampling site is to the point-source discharge site. Five of the 22 sites are near a major point source (wastewater treatment facility). A negative relation between TN concentrations and streamflow was observed at four of the five sampling sites (Town [site no. 1], Okatibbee [site no. 4], Chickasawhay upper [map ref no. 5], and Pearl lower [site no. 10]), and a negative relation between TP concentrations and streamflow was observed at two of the five sampling sites (Town [site no. 1], Pearl lower [site no. 10]). At high streamflows, there is a negative relation with nutrient concentrations at these sites because effluent contributes a smaller portion to the total streamflow which leads to a diluting effect. The exception was the Big Sunflower (site no. 16) site, where a slight positive relation was observed between concentrations and streamflow (figs. 9 and 10). This result may be caused by higher proportions of nonpoint-source contributions in this agriculturally intensive basin.

TN Concentrations

Concentrations of TN varied greatly among sites, ranging from less than 0.05 mg/L to 5.95 mg/L (table 9). The highest median concentration of TN (1.83 mg/L) was measured at Big Sunflower (site no. 16); the lowest median concentration of TN (0.32 mg/L) was measured at Tuxachanie (site no. 22) (table 9; fig. 11A). Six sites had median concentrations of TN greater than 1.0 mg/L (table 9; fig. 12). The spatial distribution of median concentrations of TN observed at the six sites corresponded closely with land use. Three sites are in the Mississippi Delta—Yazoo lower (site no. 18), Big Sunflower (site no. 16), and Bogue Phalia (site no. 17)—where land use is predominantly agriculture (greater than 50 percent). The other three sites are located near urban (developed) areas, downstream of wastewater treatment facilities—Pearl lower (site no. 10, near Jackson), Okatibbee (site no. 4, near Meridian), and Town (site no. 1, near Tupelo).

The seasonal distributions of TN concentrations at the sites were evaluated to identify patterns by examining boxplots showing concentrations as a function of month. Four sites were selected to represent agricultural, mixed, urban, and forested land use in river basins in Mississippi (fig. 13A–D). TN concentrations were typically highest during winter (December–February) and spring (March–May) and lowest during the fall (September–November) at the site representing predominantly agricultural land use basins (fig. 13A). During late fall, plants become dormant, limiting the uptake of available nutrients, and allowing for nutrients to build up in the soil. An increase in precipitation in the spring allows for nutrient runoff into the streams. In addition, nitrogen-based fertilizers are applied in the spring, thus contributing more available nutrients to the soil that potentially can runoff into streams. Conversely, concentrations of TN at the site representing urban land use basins are highest in the summer (June–August) and fall (September–November) because of the constant input of wastewater effluents during lower flows (fig. 13C). Concentrations of TN at the site representing mixed land use basins were lower than at the sites representing agricultural and urban land use basins and slightly increased in late spring and summer (fig. 13B). Concentrations of TN at the site representing forested land use basins generally were very low (less than 1.0 mg/L) (fig. 13D). The low concentrations of TN in forested land use result from nutrient uptake by vegetation and presumably less fertilizer application.

Nutrient concentrations in basin streams are directly related to land use and runoff from associated fertilizer applications and human and animal wastes (Dodds and Smith, 2016). Comparisons of median concentrations of TN among major river basins in Mississippi show that the Yazoo River Basin had the highest median concentration of TN (1.15 mg/L; fig. 14A) when compared to the other basins. The Yazoo River Basin drains large portions of the Mississippi Delta, where agriculture has been and remains a predominant land use. The relation of median TN concentrations to percentage of agricultural land use across sites showed a strong positive correlation as indicated by the Spearman’s rank correlation coefficient (rho = 0.790; p-value = 0.000) (fig. 15A). The Coastal Streams Basin had the lowest median TN concentration (0.32 mg/L; fig. 14A); the Coastal Streams Basin had the highest percentage (51.8 percent) of forested land use and the lowest percentage (2.26 percent) of agricultural land use compared with the other basins. The relation of median TN concentrations to percent forested land use across sites showed a strong negative correlation as indicated by the Spearman’s rank correlation coefficient (rho = −0.624; p-value = 0.000) (fig. 15B). This suggests that forests have considerable capacity to retain and efficiently cycle N and substantially reduce N from entering streams; also, forested areas have minimal N fertilizer application. Median TN concentrations were lower in more developed basins (Tombigbee, Pascagoula, and Pearl) than in predominantly agricultural land use basins, with a median concentration of 0.51 mg/L. Some of the highest TN concentrations in developed sites were downstream of wastewater treatment facilities. However, a weak positive Spearman’s rank correlation coefficient (rho = 0.379; p-value = 1.000) was calculated for the relation between median TN concentrations and percentage of urban land use for the sites in this study (fig. 15C).

Concentrations of TN varied among the three level III ecoregions in Mississippi, with median TN concentrations ranging from 0.75 mg/L to 1.38 mg/L (Crain and others, 2023). The Mississippi Alluvial Plain ecoregion had the highest median TN concentration, followed by the Mississippi Valley Loess Plains and Southeastern Plains ecoregions (fig. 16A). Again, the intense agricultural land use in basins that drain sites in the Mississippi Delta is a notable reason for the elevated concentrations of TN in the Mississippi Alluvial Plain ecoregion.

Water-quality conditions at a monitoring station may also be affected by the size of the drainage area draining to the location. The TN concentration dataset for the 22 sites represents a wide range of drainage areas (Crain and others, 2023). Drainage areas range in size from 235 to 34,589 km², with a median value of 1,650 km². Streams with drainage areas greater than 2,500 km² have a higher median TN concentration (0.80 mg/L) than streams with drainage areas smaller than 1,000 km² (median of 0.53 mg/L). However, the overall median concentrations of TN were highly variable across the different-sized drainage areas (fig. 17A). Concentrations of TN had a weak positive correlation with drainage area as indicated by the Spearman’s rank correlation coefficient (rho = 0.206; p-value = 0.357), indicating that concentrations of TN are not highly influenced by the size of the drainage area.

TP Concentrations

Concentrations of TP varied greatly among sites, ranging from less than 0.01 mg/L to 1.72 mg/L (table 10). The highest median concentration of TP (0.31 mg/L) was measured at Big Sunflower (site no. 16); the lowest median concentration of TP (0.02 mg/L) was measured at Tuxachanie (site no. 22) (table 10; figs. 11B and 18). The spatial distribution of median concentrations of TP corresponds closely with land use. Eight sites had median TP concentrations greater than 0.1 mg/L. Four sites with median TP concentrations greater than 0.1 mg/L are in the Mississippi Delta—Yazoo upper (site no. 15), Yazoo lower (site no. 18), Big Sunflower (site no. 16), and Bogue Phalia (site no. 17). The Mississippi Delta has been shown to contain naturally occurring high concentrations of phosphorus in the soils (Coupe, 2002) and groundwater (Welch and others, 2009), as compared to other areas of the state, which may be a source of phosphorus in streams from sediment during precipitation runoff and from return water from agriculture fields during irrigation that uses groundwater as a source. Four additional sites with median TP concentrations greater than 0.1 mg/L are located near urban (developed) areas, downstream of wastewater treatment facilities (Pearl lower [site no. 10, near Jackson] and Town [site no. 1, near Tupelo]), and within mixed land use areas (Big Black [site no. 19] and Strong [site no. 11]).

The seasonal distributions of TP concentrations at the sites were evaluated to identify patterns by examining boxplots showing concentrations as a function of month. Four sites were selected to represent agricultural, mixed, urban, and forested land use in river basins in Mississippi (fig. 19A–D). The relation of TP concentrations to season was like that of median TN concentrations. Median TP concentrations were highest during winter (December–February) and spring (March–May) and lowest during the fall (September–November) at the site representing predominantly agricultural land use basins (fig. 19A). Median concentrations of TP at the site representing mixed land use basins indicate minimal variability among the months (fig. 19B). Median concentrations of TP at the site representing urban land use basins were highest in late summer (July–August) and fall (September–November) (fig. 19C). Median concentrations of TP at the site representing forested land use basins generally were very low (less than 0.02 mg/L) (fig. 19D). Low concentrations of TP in forested land use can result from nutrient uptake and storage by vegetation and minimal phosphorus-based fertilizer application.

Patterns in variability of median TP concentrations among basins were similar to patterns observed for median TN concentrations, except for the Big Black River Basin. The Big Black River Basin and the Yazoo River Basin had the highest median concentrations of TP (0.21 mg/L) (fig. 14B). The Big Black River and most of its tributaries transport large amounts of suspended sediment (MDEQ, 2003), possibly because of minimal riparian vegetation in the headwaters and because of riverbank erosion. Results from the 2012 USGS SPAtially Referenced Regressions on Watershed attributes (SPARROW) suspended-sediment model indicate that the Big Black River ranks third highest for suspended-sediment yields among 16 of the 21 ambient water-quality network sites used as SPARROW calibration sites in Mississippi (Robertson and Saad, 2019). River systems having large amounts of suspended sediment, such as the Big Black River Basin, generally have higher concentrations of P. The higher concentrations of P are due to the sorption of P to soil particles and streambed sediments, runoff of soils, and resuspension of streambed sediments during high-flow periods.

As stated earlier in the discussion of TN concentrations, the Yazoo River Basin drains large portions of the Mississippi Delta, where agriculture has been and remains a predominant land use. The relation of median TP concentrations to percentage of agricultural land use across sites showed a strong positive correlation as indicated by the Spearman’s rank correlation coefficient (rho = 0.834; p-value = 0.000) (fig. 20A). The Coastal Streams Basin had the lowest median TP concentrations (0.02 mg/L; fig. 14B), presumably because the Coastal Streams Basin has the highest percentage (51.8 percent) of forested land use and the lowest percentage (2.26 percent) of agricultural land use compared with the other basins. A negative Spearman’s rank correlation coefficient (rho = −0.555; p-value = 0.000) was calculated for the relation between median TP concentrations and percentage of forested land use across the sites (fig. 20B). A weak positive Spearman’s rank correlation coefficient (rho = 0.047; p-value = 1.000) was calculated for the relation between median TP concentrations and percentage of urban land use across the sites (fig. 20C).

The concentrations of TP varied among the three level III ecoregions in Mississippi as was observed with concentrations of TN. The Mississippi Alluvial Plain ecoregion had the highest median TP concentration, followed by the Mississippi Valley Loess Plains and Southeastern Plains ecoregions (fig. 16B). Again, the source of TP at sites in the Mississippi Alluvial Plain may be from naturally high concentrations in the soils and from irrigation water using groundwater that contains naturally high concentrations of TP.

Streams with drainage-basin areas greater than 2,500 km² have a higher median TP concentration (0.16 mg/L) than basins smaller than 1,000 km² (0.05 mg/L; Crain and others, 2023). However, the overall median concentrations of TP were variable across the different-sized drainage areas (fig. 17B). Concentrations of TP had a strong positive correlation with drainage-basin area, as indicated by a Spearman’s rank correlation coefficient (rho = 0.458; p-value = 0.032). This result suggests that concentrations of TP are somewhat influenced by the size of the drainage-basin area.

Estimated Mean Annual Nutrient Loads and Yields, 2008–18

Load represents the mass (expressed in kilograms) of a given water-borne constituent moving past a given point per unit of time. Mean annual loads (in kilograms per year) for TN and TP were estimated by using MDEQ water-quality data and USGS streamflow and water-quality data from WYs 2008 to 2018 at 22 stream-water-quality stations in Mississippi (tables 11 and 12) (Crain and others, 2023).

Because the drainage areas for the 22 monitoring sites range from 235 km² for the Tuxachanie site to 34,589 km² for the Yazoo lower site, yield (load per unit of drainage area and expressed as kilograms per year per square kilometer) is used instead of load so that comparisons in the change of yield can be made among the 22 monitoring sites.

TN Loads and Yields

Estimated mean annual TN loads ranged from 74,700 kilograms per year (kg/yr) at the Tuxachanie site, which is in the smallest basin in the study, to 20,874,000 kg/yr at the Yazoo lower site, which is in the largest basin in the study (table 11). Load variability is predominantly controlled by basin size and overall streamflow; larger basins generate larger nutrient loads.

Investigating loads transported from upstream to downstream can be helpful for identifying mechanisms of transport and evaluating the relative load contributions from different source areas. A transported load, as defined in this report, represents the load added to a river in an upstream-to-downstream reach that ends with a monitoring site. The transported load is computed by subtracting the upstream site load from the downstream site load. In this study, only the Chickasawhay, Pearl, and Yazoo Rivers had an upstream and a downstream site on the main stem. Almost half of the transported load of TN in the Yazoo (46 percent) and Chickasawhay (42 percent) Rivers occurs at their respective upstream sites, but only 22 percent of the transported load for the Pearl River occurs at the Pearl upper site. Most of the transported load in the Pearl River occurs between the upper and lower monitoring sites.

Yields of TN for river basins and ecoregions are often estimated to evaluate the effectiveness of water-quality management programs. A comparison of nutrient yields among river basins and ecoregions can be useful to water-resource managers when prioritizing basins for water-quality improvements. In general, the highest annual yields of TN were estimated at sites in the Yazoo River Basin, with the highest annual TN yield being 1,014 kilograms per year per square kilometer ([kg/yr]/km²) at the Bogue Phalia site (table 11). The lowest annual yields of TN (less than 300 [kg/yr]/km²) were estimated at the Luxapallila site in the Tombigbee River Basin and the Chickasawhay lower site in the Pascagoula River Basin. TN yields for the basins are shown in figure 21.

The relation between land use and TN yields was evaluated to determine whether differences in TN yields are discernible among different percentages of three land use types (agricultural, developed [urban], and forest) at the monitoring sites. A strong positive Spearman’s rank correlation coefficient (rho = 0.781; p-value = 0.000) was calculated for the relation of mean TN yield to percentage of agricultural land use across the sites (fig. 22A). A negative Spearman’s rank correlation coefficient (rho = −0.718; p-value = 0.001) was calculated for the relation of mean TN yield to percentage of forested land use across the sites (fig. 22B). A weak positive Spearman’s rank correlation coefficient (rho = 0.123; p-value = 1.000) was calculated for the relation of mean TN yield to percentage of urban land use across the sites (fig. 22C).

TP Loads and Yields

Estimated mean annual TP loads ranged from 6,700 kg/yr at the Tuxachanie site in the Coastal Streams Basin to 5,717,000 kg/yr at the Yazoo lower site in the Yazoo River Basin (table 12). In this study, only the Chickasawhay, Pearl, and Yazoo Rivers had an upstream and a downstream site on the main stem. Transported TP loads were investigated for all streams with both an upstream and downstream monitoring site. Mean annual loads of TP were 2.7-fold (Yazoo River), 2.8-fold (Chickasawhay River), and 6.7-fold (Pearl River) larger at the downstream monitoring site than at the upstream monitoring site on each river. Based on drainage-area size differences, the Yazoo River receives relatively equal loads above and below the upper site, the Chickasawhay River receives a greater load above the upper site than above the lower site, and the Pearl River receives a greater TP load above the lower site as compared to the upper site. The large loads above the upper site of the Chickasawhay River and between the upper and lower sites of the Pearl River are likely due to effluent from the Meridian and Jackson publicly owned treatment works, respectively, during the period of record.

In general, the highest annual yields of TP were estimated at sites in the Yazoo River Basin; the highest estimated annual TP yield was 234 (kg/yr)/km² at Bogue Phalia (fig. 23; table 12). Sites with the lowest annual yields of TP (less than 30 [kg/yr]/km²) were in the Tombigbee River (one site), Pascagoula River (two sites), and Coastal Streams (one site) Basins.

The relation between land use and TP yields was evaluated to determine whether differences in TP yields are discernible among different percentages of three land use types (agricultural, urban [developed], and forest) at the monitoring sites. A strong positive Spearman’s rank correlation coefficient (rho = 0.691; p-value = 0.001) was calculated for the relation of mean TP yield to percentage of agricultural land use across the sites (fig. 24A). A weak negative Spearman’s rank correlation coefficient (rho = −0.154; p-value = 1.000) was calculated for the relation of mean TP yield to percentage of forested land use across the sites (fig. 24B). A weak positive Spearman’s rank correlation coefficient (rho = 0.182; p-value = 1.000) was calculated for the relation of mean TP yield to percentage of urban land use across the sites (fig. 24C).

Trends in Streamflow

Changes in streamflow are an important influence on nutrient concentrations and loads in streams. Sources and transport of nutrients in a basin can be influenced by increases or decreases in streamflow. Understanding how streamflow changes through time is critical in assessing how and why nutrient concentrations and loads change through time. Trends in streamflow often result from variability in climatological conditions such as precipitation; however, trends in streamflow also can result from variability in water regulation operations such as diversions and reservoir storage and release in some streams and rivers.

Overall, the streamgages associated with the ambient water-quality network sites had more upward trends in streamflow than downward trends. Of the 20 sites, the greatest increase in streamflow was at the Tuxachanie site, with a 60-percent increase during WYs 2008–18 (table 13). Streamflow at all sites in the Pascagoula River Basin increased during the study period; increases ranged from 32 to 46 percent for 2008–18. Significant (p-value less than or equal to 0.05) trends in streamflow occurred at 16 of the 20 sites during WYs 2008–18 (table 13). The significant streamflow trends at the 16 sites were upward at all but 6 sites. These upward trends occurred at various sites throughout the different major river basins and ecoregions of Mississippi. Trends in streamflow could not be assessed at the two NAWQA sites, because trends were not provided on the National Water Quality Network website (USGS, 2022).

Trends in Nutrient Loads

Changes in water quality are often obscured by the year-to-year variations in streamflow; therefore, flow-normalized trends in load are better suited to illustrate how water quality has changed through time. Descriptions for the likelihood of an upward or downward trend depend on the likelihood of the trend direction; for example, if an upward trend is highly likely, then this also means that a downward trend is highly unlikely (table 14). Temporal trends are presented as changes in flow-normalized loads for WYs 2008–18. Trend probability is provided based on the percentage changes for the period of analysis.

Comparing Study Results to Other Published Nutrient Annual Yields

Sixteen of the 22 ambient water-quality network sites were used as SPARROW calibration sites to estimate loads and yields of TN and TP (Saad and others, 2019). The period of water-quality record used for these sites for SPARROW was 2000–14. This period overlaps with the period of record for load and yield estimates in this study (2008–18) and thus may provide an independent validation of results of this study.

Although this study and the 2012 SPARROW-calibration used different load estimation methods and had slightly different periods of record for water-quality samples, the study-derived TN and TP yields compared favorably with the 2012 SPARROW-calibration TN and TP yields (tables 16 and 17). Overall, the total average percent differences between the results from this study and the 2012 SPARROW-calibration results were 10.5 percent for TN yields and 16.5 percent for TP yields. Six of the 16 sites had differences in TN yields less than 10 percent, and 7 of the 16 sites had differences in TP yields less than 10 percent. All other sites had less than a 20 percent difference in TN yields. For TP yields, 11 sites had differences less than 20 percent, 3 had differences between 20 and 30 percent, and 2 had differences greater than 40 percent. Differences for mean annual TN yields varied, with 7 of the 16 sites having higher yields from this study (average of 11.2 percent difference) and 9 having lower yields from this study (average of 10.0 percent difference). Differences in TP yields varied as well, with 7 of 16 sites having higher yields from this study (average of 14.7 percent difference) and 9 having lower yields from this study (average of 17.9 percent difference). In general, the TP mean annual yields have a higher percent difference than the TN mean annual yields, possibly because of the transport processes involving sediment-adsorbed P.

Four of the 22 ambient water-quality network sites are part of the south-central United States regional-scale LOADEST model (Rebich and Demcheck, 2007); the model was based on nutrient data collected from WYs 1993 to 2004. The study-derived mean annual TN yields at the four sites (Little Tallahatchie, Bogue Phalia, Yazoo lower, and Bayou Pierre) were lower than the south-central United States regional-scale LOADEST model results, with an average percent difference of 32.1 percent. Only three sites (Bogue Phalia, Yazoo lower, and Bayou Pierre) could be used to compare the study-derived TP yields with the LOADEST-derived TP yields. Two sites (Bogue Phalia and Bayou Pierre) had lower TP yields from this study, with an average percent difference of 64.7 percent. The Yazoo lower site had a higher TP yield from this study, with a difference of 20.0 percent. The results indicate that yields of TN and TP may be decreasing as compared to 1993–2004 conditions, except for TP at the Yazoo lower site.

The study-derived TN and TP yields for the Yazoo lower site were compared to the results from the Mississippi Embayment National Water-Quality Assessment ESTIMATOR regression model which was based on data from 1996 to 1997 (Coupe, 1998). The TN yield estimated by Coupe (1998) was higher than the TN yield estimated by Rebich and Demcheck (2007) and by this study, providing further evidence that the TN yield has possibly decreased during 1997–2018. The TP yield estimated by Coupe (1998) was more similar to the TP yield estimated during this study (6.3 percent difference) and was higher than the TP yield estimated by Rebich and Demcheck (2007), indicating that TP yield at the Yazoo lower site is slightly higher now (2018) than in previous decades.

Comparing Study Results to 2012 SPARROW-Derived Yields

The SPARROW model is a hybrid statistical and mechanistic model that was developed from monitoring results at numerous long-term monitoring sites. The annual nutrient yields from this study were compared to the 2012 SPARROW-prediction model results at 22 ambient water-quality network sites in Mississippi. Study-derived mean annual TN and TP yields for the 22 ambient water-quality network sites varied in consistency with the 2012 SPARROW-predicted yields (fig. 25), with a difference in TN yields of 20.0 percent and a difference in TP yields of 38.6 percent.

The 2012 SPARROW-prediction model overestimated mean annual TN yields for 8 sites (18.8 average percent difference) and underestimated mean annual TN yields for 14 sites (20.7 average percent difference) as compared to TN yield estimates from this study (fig. 25A). TN yields at five of the six sites with the highest TN yields were underestimated by the 2012 SPARROW-prediction model. The TN yields at the five sites in the Pascagoula River Basin were similar, with four of the five sites having less than an average 2 percent difference (table 18). The TN yields at the three sites in the Tombigbee River Basin were overestimated, with an average difference of 19.0 percent. Four of the five monitoring sites in the Yazoo River Basin are in the Mississippi Delta, and TN yields at these four sites were underestimated (average 25.4 percent difference) by the 2012 SPARROW-prediction model. The sites within the remaining basins had over- and underestimations of TN yields.

Study-derived mean annual TP yields for the 22 ambient water-quality network sites are less consistent with the 2012 SPARROW-predicted TP yields as compared to TN yields (fig. 25B), with an overall average percent difference of 38.6 percent. The 2012 SPARROW-prediction model overestimated mean annual TP yields for 10 sites (average of 44.4 percent difference) and underestimated mean annual TP yields for 12 sites (average of 33.9 percent difference) as compared to TP yield estimates from this study. Yield comparisons at sites with lower yields tended to be consistent. Yield comparisons at sites with midrange yields tended to be overestimated by the 2012 SPARROW-prediction models, and those at sites with high yields tended to be underestimated by the 2012 SPARROW-prediction models.

Study-derived mean annual TP yields for sites in the Pascagoula River Basin were similar to the 2012 SPARROW-prediction estimated yields. The study-derived TP yields and the 2012 SPARROW-prediction yields were similar at two sites in the Tombigbee River Basin and were overestimated by the SPARROW model at one site. The 2012 SPARROW-prediction model overestimated the mean annual TP yields at three sites in the Pearl River Basin and underestimated them at one site, all by an average difference of 63.1 percent. Four of the five monitoring sites in the Yazoo River Basin are in the Mississippi Delta and are associated with very high yields; TP yields were underestimated by the 2012 SPARROW-prediction model by an average difference of 44.1 percent. The sites within the remaining basins had over- and underestimations of TP yields