| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
National Center for Water Quality Research, Heidelberg College, 310 E. Market Street, Tiffin, OH 44883
* Corresponding author (prichard{at}heidelberg.edu).
Received for publication November 7, 2007.
 |
ABSTRACT |
|---|
 TOPNOTES ABSTRACT INTRODUCTIONMethodsResults and DiscussionConclusionsREFERENCES |
|---|
Abbreviations: BMP, best management practice • lnQ, natural log of flow • lnSS, natural log of TSS concentration • lnSSld, natural log of TSS load • LOWESS, locally weighted scatterplot smoother • NCWQR, National Center for Water Quality Research at Heidelberg College • SSdQ, lnSS adjusted for flow using LOWESS • TSS, total suspended sediment • USGS, U.S. Geological Survey
|
INTRODUCTION |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMethodsResults and DiscussionConclusionsREFERENCES |
|---|
By the late 1960s, Lake Erie was suffering from cultural eutrophication (ES&T, 1967; IJC, 1965, 1969). This manifested itself in many ways, but some of the major ones were increased extent and severity of hypoxia in the Central Basin hypolimnion in late summer (Carr, 1962), the extirpation of mayflies (Krieger et al., 2007) and impacts on other benthic fauna, a simplified trophic structure dominated by alewives and other planktivores, and extensive blooms of Cladophora (Read, 1999) and noxious cyanobacteria such as Microcystis. Rehabilitation of the open lake focused on reducing phosphorus to limit algal productivity (ACOE, 1982; GLWQA, 1972; IJC, 1965, 1969, 1978). As remediation efforts moved beyond point sources to include nonpoint sources of pollution, the control of erosion and sediment transport to Lake Erie became important because much of the nonpoint phosphorus was bound to sediment (IJC, 1978). In this context, improved knowledge of tributary sediment concentrations and loads became a matter of concern.
Sediment was and remains important as a carrier of phosphorus and many other pollutants, but sediment is also important in its own right. Sediment deposited in harbors and shipping channels necessitates expensive dredging and creates problems disposing of the dredge spoils (Myers and Metzker, 2000). The need to remove sediment from source waters increases the costs of drinking water treatment. Sediment deposited in rivers and streams can clog spawning beds and degrade other benthic habitats (Hynes, 1974). Turbidity from suspended sediment may limit light penetration and thereby photosynthesis, favoring planktonic algae over aquatic macrophytes in river mouths and bays (Chow-Fraser, 1999).
The Ohio Lake Erie Commission has identified sediment loading as one of its water quality metrics for Lake Erie, rated it as poor, and called for a 67% reduction from baseline conditions measured in 1991 through 1996 (OLEC, 1998). The Lake Erie watershed of northwestern Ohio has been the site of many demonstration programs and other efforts to stimulate adoption of agricultural best management practices (BMPs) to reduce sediment and phosphorus losses (see Forster and Rausch [2002] for a review). Most recently, the Ohio Lake Erie Conservation Reserve Enhancement Program has been a major effort to promote sediment reduction measures in the western Ohio Lake Erie watershed (see http://www.ohiodnr.com/soilandwater/crephome.htm).
Pre-settlement loading rates were a small fraction of what they are today. Sedimentation rates provided by Sly (1976) suggest that current loading rates are 10 times the long-term historical average. Annualized Agricultural Non-Point Source (AnnAGNPS) pollution model simulations of totally forested conditions for the Upper Auglaize watershed, a sub-watershed of the Maumee River, suggest that pre-settlement loads might have been less than 1% of current levels (Bingner et al., 2004). We will likely never get close to those pre-settlement loads, but it remains important to reduce sediment loads as much as possible and to track our progress or failure in reducing sediment concentrations and loads in Lake Erie tributaries (and elsewhere).
The National Center for Water Quality Research (NCWQR) at Heidelberg College has been monitoring sediment and nutrients in the major U.S. tributaries to Lake Erie for many years, with some records extending back as far as the late 1960s. This study is an analysis of NCWQR sediment data for the 30-yr period beginning with the 1975 water year (which began 1 Oct. 1974) and continuing through the 2004 water year (ending 30 Sept. 2004). The goals of the study were to identify and quantify trends in sediment loads and concentrations and to interpret them in the context of efforts to manage Lake Erie and its watershed.
|
Methods |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMethodsResults and DiscussionConclusionsREFERENCES |
|---|
|
|
|
The results in this paper are based on a total of nearly 75,000 sediment sample analyses. NCWQR sample collection and analysis at these stations continues; data can be accessed at http://wql-data.heidelberg.edu./index2.html.
Sediment Analysis
Total suspended solids (TSS) concentrations are measured using USEPA Method 160, Residue (Gravimetric) (O'Dell, 1993). The TSS concentration is determined by filtering a sample aliquot of known volume through a pre-dried and pre-weighed glass fiber filter, then redrying and reweighing the filter. Results are reported as mg L–1.
Data Preparation and Analysis
Before statistical analysis, daily flow-weighted mean sediment concentrations were calculated by dividing the daily load by the daily discharge. The daily load is the sum of sample loads obtained by multiplying the observed concentration by the instantaneous flow at the time of the sample and a sample time window. The time window is equal to half the time from the previous sample to the current sample (or the time from the beginning of the day for the first sample) plus half the time from the current sample to the next sample (or the time to the end of the day for the last sample). The daily discharge is calculated analogously. For subsequent statistical analysis, the official daily mean flows reported by the U.S. Geological Survey (USGS) were used rather than the daily discharges calculated as just described. For statistical analyses that were based on loads, these loads were re-calculated as the product of the daily flow-weighted mean concentration and the USGS daily mean flow.
For each river, trends in flow, suspended sediment load, and suspended sediment concentration were evaluated. Concentration data were analyzed without and with (natural) log transformation and after adjustment for the influence of flow on concentration. Concentration data were also analyzed by season (winter/spring/summer/fall) and within six classes of flow. LOWESS (Locally Weighted Scatterplot Smoother; Cleveland, 1979) was used for visualization of trends and to evaluate their complexity. A span of 40% was chosen for these plots because it preserved multi-year patterns of change but masked shorter-term fluctuations of lesser interest and dubious interpretability. The significance of trends was determined using various forms of linear regression, with adjustment for autocorrelation.
Data Transformations and Subdivisions
Flow, load, and concentration data were transformed to natural logs before regression analysis to evaluate the statistical significance and magnitude of trends. Henceforth, lnQ will be used for the natural log of flow, lnSS for the natural log of TSS concentration, and lnSSld for the natural log of TSS load.
Concentrations of TSS were adjusted for flow by computing the LOWESS Rough of plots of lnSS versus lnQ; a span of 10% was used for this adjustment. A smaller span was chosen than was used for trend visualization because the lnSS–lnQ relationships lack the intermediate-term excursions that need to be smoothed out of the trend plots. The results are henceforth referred to as SSdQ. This approach was used in preference to alternatives, such as using the residuals from a regression analysis of the same variates or incorporating lnQ as a covariate in the formal trend analysis, because the relationship between lnSS and lnQ is in some cases decidedly nonlinear.
Two seasons of 6 mo duration were determined by iteration of an ANCOVA model of lnSS as a function of lnQ and a dummy variable with the values "summer" and "winter" and interaction between the two. Of the six possible groupings of the months, the one that had the lowest mean square error was chosen to represent summer and winter. These seasons were used in an initial seasonal trend analysis and then divided into four seasons of 3 mo each for a final seasonal analysis. Differences in seasonal average concentrations were evaluated using the non-parametric Mann-Whitney U test on TSS and the two-sample Student's t test on lnSS.
Flows were divided into six categories: four for storm runoff periods and two for low-flow conditions. Storm runoff periods were defined as periods during which the flow continually exceeded the 60th percentile of all flows. Such periods were divided into large and small individual storms, and each of these was further divided into rising and falling limbs. Non-storm flows were divided into two groups: low flows (between the 0th and 30th percentile) and intermediate flows (between the 30th and 60th percentiles). Large storms were identified as storms with a peak flow greater than the 95th percentile of all flows; small storms had peak flows less than the 95th percentile. Large storms may have some flow values as low as the 60th percentile; consequently, some flows in large and small storms are of comparable magnitude. For efficient notation, rising and falling limbs of large storms were designated LR and LF, respectively; rising and falling limbs of smaller storms were designated SR and SF. Intermediate flows were designated INT, and low flows were designated LOW. Flow class was later used as a categorical variable in an ANCOVA to examine trends within each flow class.
Linear Regression
Multiple linear regression was used to evaluate trends for the period of record as a whole and for shorter segments of the record identified by inspection of LOWESS Smooths of the data. All analyses were done using a natural log-transformed sediment parameter (lnSS, lnSSld, or SSdQ) as the dependent variable and time as the independent variable, with time expressed as decimal years since the beginning of Water Year 1975 (i.e., 1 Oct. 1974). Sine and cosine terms in 2x
xtime were included in the analysis to reduce variance due to seasonality. Results were expressed as percent change in the dependent variable per decade, given by
 |
is the percent change, and bt is the coefficient of the time term from the regression analysis.
Autocorrelation
Regression and related procedures assume that the residuals from the analysis are independent (i.e., that they have no autocorrelation). When autocorrelation is present and positive, the probability values reported by the statistical test are inflated and may suggest that a result is statistically significant when this is not the case (Loftis and Ward, 1980). Analyses using daily data often have highly autocorrelated residuals. The approach to this problem used here is the one described in Richards and Baker (1993, 2002). Autocorrelation is assumed to be adequately described by an autoregressive time series model with a one step lag (AR1), in which case reported t values can be adjusted by multiplying them by a factor calculated from the sample size and the lag-one autocorrelation coefficient. These adjusted t values are then used to determine the probability value for the statistic of interest, usually a trend slope.
For seasonal analyses, autocorrelation is poorly defined because it assumes a constant time step, whereas these data series have 90 or so observations spaced a day apart, followed by a gap of about 270 d, representing the other seasons. For this reason, the correction factor determined for the nonseasonal data series was used to adjust the regression results for these interrupted seasonal series as well.
Piece-Wise Trends
A number of LOWESS displays indicated substantial reversals of trend direction during the period of record. In other cases, points of inflection were present, separating a time of rapid change from a time of slow change in the same direction. In such cases, although the overall linear trend result summarizes the net change during the period of record, it fails to reveal important details. In these cases, the data were split into two or more segments, and each segment's trend was evaluated separately.
Whether a change in slope displayed in a LOWESS plot is substantial enough to justify multiple-slope analysis is a matter of judgment. Values of the LOWESS Smooth were examined to identify times of maxima and minima; points of inflection were identified by interactive examination of the Smooth values and the graph of these values. In general, trend reversals were given more weight than inflections in the slope. Changes with durations shorter than 4 or 5 yr were generally ignored because such fluctuations might be caused by the smoothing of data for as little as one unusual year, and management changes sufficient to cause significant trends are unlikely to occur over such a short time frame. In no case were more than three segments analyzed.
All analyses included sine and cosine terms in 2xxtime to reduce the variance due to seasonality. Slopes were tested for significant difference from those of adjacent segments using a procedure described in Kleinbaum et al. (1998).
ANCOVA for Changes in Sediment–Flow Relationship
Trends may be identified that have no causes other than changes in weather. Such trends are not of particular interest for this paper because they do not reflect success or failure of environmental management efforts. Changes in weather are reflected primarily in changes in flow, among the parameters available for study. Changes in weather are not expected to change the lnQ–lnSS relationship (the "rating curve"). Rather, the data should migrate up and down the rating curve from year to year. Changes in land use and environmental management, on the other hand, are intended to reduce the amount of sediment associated with a given flow, leading to a change in the rating curve itself. Increases in practices like conservation tillage should decrease erosion and cause the regression line to shift downward (lower slopes and/or lower intercepts) over time. Analysis of covariance was used to evaluate this, with lnSS a function of lnQ and YEAR. YEAR was treated as a categorical variable. Examination of preliminary results with and without interaction between lnQ and YEAR demonstrated that the major change over time was typically in the intercept rather than the slope, so the final analysis was done without interaction. With this mode of analysis, an intercept was calculated for each year of the period of record, and these values were not constrained to change in a systematic way over time. The values of YEAR were then regressed against time to determine if there was a systematic shift in the rating curve over time.
|
Results and Discussion |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMethodsResults and DiscussionConclusionsREFERENCES |
|---|
. Each figure contains six plots: one each for untransformed flow, sediment concentration, and sediment load, one of flow-adjusted concentration (SSdQ), one of SSdQ divided into "summer" and "winter" seasons as described in the previous section, and one of the change in sediment concentration per decade in each of six flow classes previously described.
|
|
|
|
|
|
|
The River Raisin (Fig. 2) shows a generally downward trend in flows but an upward trend in suspended sediment concentrations and loads, especially since 1995. Summer concentrations are generally higher than winter concentrations, but the winter trend is more strongly increasing, and the LOWESS trends meet about 2005. The greatest increases in concentrations are in the falling limb of the large and small storms and in the intermediate non-storm flow class. When expressed as percent change in concentration per decade, the largest changes are in the low flow categories. There is reason to suspect that these trends may be influenced by sampling artifacts: the Raisin is the only station that is sampled manually, and the data suggest that the sampling bucket may have occasionally bumped the bottom, especially under low flow conditions. Efforts are underway to establish an autosampler station on the Raisin.
The Maumee and Sandusky Rivers (Fig. 3 and 4) show similar patterns, with little change in flows over the period of record, except for an increase in flow in the Sandusky in the last 5 yr of the study period and a reversal of the downward sediment trend in the Maumee between about 1982 and 1988. In both cases, summer concentrations are greater than winter concentrations and are declining more rapidly. The largest decreases in concentration are associated with storm runoff rather than low flow and particularly with the rising limb of larger storms.
Honey Creek (Fig. 5) has fluctuating flows that have no clear overall trend and sediment concentrations and loads that trend downward, especially since about 1994. The flow-adjusted concentrations suggest that before 1995, there was a slight upward trend. Summer concentrations are greater than winter concentrations and show a stronger downward trend since 1994. The largest decreases in concentration are associated with the rising limb of smaller storms. The average decrease (about 18 mg L–1) is modest compared with those for the Sandusky and Maumee Rivers, which exceed 50 mg L–1.
Rock Creek (Fig. 6) has fluctuating flows that indicate an overall decrease until 2000, followed by a marked increase; the resulting overall change is minor. The trend in loads mostly mirrors the trend in flow. Concentrations show a small downward trend, especially when adjusted for effects of flow. Summer concentrations are higher than winter concentrations and decreased more rapidly until 1998; since then winter trends have decreased more rapidly. Changes as a function of flow class are mixed and in no case very large. The sediment concentrations in general are lower than at most of the other stations, the one exception being Honey Creek in the later part of the period of record.
Trends for the Cuyahoga River (Fig. 7) are dominated by a strong increase in flow starting about 2000. This increase is also reflected in suspended solids concentrations and loads. Flow-adjusted concentrations show no trend until about 1998 and a show weak upward trend since that time. Seasonal differences are weak, although increasing trends are somewhat more pronounced in summer than in winter. The largest increases are associated with the rising stage of storm runoff events. The Cuyahoga has several major impoundments in the Akron area, and as a consequence most of the suspended sediment measured at our station at Independence is derived from the lower part of the watershed. The trend observations are consistent with increased erosion of the steep valley sides, including the banks and bed of the river and its tributaries, as a consequence of increased flows.
Trends for the Grand River (Fig. 8) are characterized by a strong decrease in sediment early in the period of record, through 1995, followed by a more stationary period and an increase after 2000. The load trends are visually similar to the flow trends, although the early decrease in loads seems to reflect changes in concentration more than changes in flow. The flow-adjusted concentration trend is similar to the untransformed concentration trend, and seasonal differences are minor. The largest decreases over the period of record are substantial and are associated with the rising limb of large storms.
Statistical Results
Linear Trends for the Entire Period of Record
Representing data for these rivers as a single linear trend in many cases ignores some important details (Fig. 2–8
). Nonetheless, such analyses indicate the net change over the entire period of record, which is a useful result, especially when supplemented by the more detailed analyses that follow. Table 3
presents the results of analyses of lnQ, lnSS, lnSSld, and SSdQ for each tributary.
|
0.001; all of the trends for SSdQ are statistically significant at p 0.0001. Three of seven trends for lnSSld are statistically significant, generally at 0.05 > p > 0.001. The difference in significance levels reflects the greater variance of loads as compared with concentrations and the success of flow-adjustment in decreasing the variance (and autocorrelation) of concentration data. Changes in flow are generally modest (and not statistically significant) and in all cases smaller than changes in sediment concentrations and loads for the same river. The largest decreases in sediment loads and concentrations, generally about 50% per decade, are associated with the Grand River. The largest increases in loads (22%) and concentrations (34%) are associated with the Cuyahoga River.
Summer and Winter Levels and Trends
For each tributary, the ANCOVA analysis described previously identified the optimal seasons as follows: winter is November through April and summer is May through September. Differences in seasonal averages are presented in Table 4
. Differences between summer and winter were significant for all rivers but the Grand; the seasonal contrast for the Cuyahoga was not as strong as for the rivers to the west. In all cases for which there were significant differences, summer average concentrations were higher than winter average concentrations.
|
|
Piece-Wise Trends with Breakpoints Indicated by LOWESS Curves
Examination of Fig. 2 through 8 reveals periods of relatively consistent change separated by maxima, minima, or points of inflection. Flow-adjusted concentration data for each river were divided into segments at these change-points, the trends were analyzed for each segment separately, and the resulting trend slopes were evaluated for statistical significance (i.e., difference from no trend) and for difference from the trends for adjacent segments. Results of these analyses are presented in Table 6
along with the time periods that define each segment.
|
Two of the more striking changes in these trends are the sudden increase in the Cuyahoga since 1996 and the strong decrease in the Grand between 1988 and about 1995. For the Cuyahoga, the recent trend is highly significant, as is the difference between the trends for the two periods. The Cuyahoga watershed has experienced a great deal of urban sprawl during the last decade (personal communication, James White, Cuyahoga River Community Planning Organization, May 2007). Conversion of forest to residential land use has been particularly extensive in several of the sub-watersheds upstream from the NCWQR sampling station. Accompanying this conversion are increases in impervious areas and areas with reduced perviousness (e.g., lawns) and loss of canopy cover. The remaining canopy has been thinned by understory clearing, gypsy moth attacks, and grazing by urban deer. All of these factors tend to lead to increased and flashier runoff and presumably are important causes of the observed trend toward higher sediment concentrations.
The Grand River has a very strong downward trend in the initial period (1988–1995) that is highly significant (p < 0.0001) and a weaker downward trend since then that is not statistically significant (p = 0.1061). The two trend slopes are significantly different from each other. The causes of the observed trends are not clear. Some of the factors that in interaction may be responsible for the trends include extensive development, culminating about 1999, in the Big Creek watershed, which enters the Grand a short distance upstream of the station; increased storm water runoff and flashiness leading to erosion of entrenched tributary channels; farms reverting to forest in parts of the watershed; and the development of a new channel for the Grand River near the station, beginning about 2000, involving the cutting off of a major meander bend (personal communication, Brett Rodstrom, Grand River Partners, May 2007).
ANCOVAs for Change in the Sediment–Flow Relationship
Results of the ANCOVA described in the Methods section are reported in Table 7
(the analysis is for Maumee River data). The graph in Fig. 9
is a plot of the intercept for each year as a function of time. The downward slope of 0.0207 natural log units per year is highly significant (p = 0.0001; R2 = 45.9). This indicates that the amount of sediment expected for a given flow is decreasing over time.
|
|
On the other hand, these results are consistent with the expected benefits of erosion-control practices such as reduced tillage and the use of filter strips. They are also consistent with increases in tile drainage in these watersheds, at least according to modeling studies (Bingner et al., 2004). Tile drainage reduces soil moisture and allows infiltration of more rainfall, especially early in the storm period when erosion and soil loss tend to be greatest. Model scenarios that incorporate tile drainage generate more total discharge but less surface discharge and sediment yield than scenarios that differ only in lacking tile drainage.
Thus, the observed changes in the relationship between sediment and flow seem to reflect beneficial effects of two groups of agricultural practices, both of which are widely applied in these watersheds and have increased in extent and in effectiveness during the period of this study.
Summer and Winter
The same ANCOVA was performed on summer and winter data treated as two separate datasets. The summer decreasing trend is steeper than the overall trend and has a higher R2 and greater statistical significance (Table 8
). The winter trend decreases relatively gradually, explains a minimal amount of variance, and is not statistically significant. Thus, most of the overall trend is due to changes during the summer months.
|
, grouped together, whereas the low flow category corresponded to non-storm categories INT and LOW taken together. Decreasing trends were stronger for the low flow data than for the high flow data. These results are not completely independent from those for summer and winter data because summer months have more frequent low flows than winter months. To try to separate seasonal effects from flow-related effects, the seasonal ANCOVA was performed again using four seasons: winter (November–January), spring (February–April), summer (May–July), and fall (August–October). On average for the period of record, winter and summer each have about 25% of the annual discharge, spring has 40%, and fall has 10%.
The ANCOVA results (Table 9
) are similar to the two-season results reported previously. The winter trend is relatively weak and falls just short of statistical significance (slope = –0.0129; p = 0.0585). The spring season shows virtually no change (slope = –0.0009; p = 0.8957). The summer and fall trends are strong, highly significant, and of similar magnitude (slope = –0.0272 and –0.0289, respectively; p 0.0001 in each case). This pattern does not match the flow pattern, which has similar winter and spring discharges. Consequently, the observed seasonal patterns are at most only partly a consequence of differences in flow.
|
Other Rivers
Results of similar ANCOVAs for the other rivers are also reported in Tables 8 and 9. For a given river, the trend directions are consistent: summer, winter, high flow, low flow, and overall trends are all in the same direction. The only exception is Rock Creek, which has three decreasing trends and two increasing ones. However, the increasing trends are trivially different from zero and far from statistically significant.
For the most part, the other rivers parallel the Maumee in the seasonal aspects of their trends. Trends are more strongly decreasing or less strongly increasing in the summer season than in the winter season. In either case, the summer trend is the "better" outcome. For example, the Raisin winter trend is upward at 0.0650 (p 0.0001), whereas the summer trend is upward at 0.0188 (p = 0.0020), and the Sandusky winter trend is downward at –0.0071 (p = 0.2848, not significant), whereas the summer trend is downward at –0.0253 (p 0.0001). The one exception to this pattern is the Cuyahoga, which shows a larger upward trend in summer (0.0301; p = 0.0020) than in winter (0.0159; p = 0.0941, not significant). The other rivers parallel the Maumee in the flow-related aspects of their trends. In all cases, the trends under low flow conditions decrease more or increase less than those under high flow conditions.
Four-Season Analysis
Results for ANCOVAs using four seasons are basically consistent with the results of the two-season analysis and the four-season trend analyses reported previously. For each river, the season of maximum average daily flow is spring, followed by winter, then summer, then fall. The median of the seasonal discharges for the seven rivers, expressed as percentages of the average annual discharge for each river, is 41.5% in spring (range, 35.6–43.0%), 29.4% in winter (range, 25.5–34.8%), 22.1% in summer (range, 16.2–24.2%), and 8.1% in fall (range, 6.8–13.8%). The Cuyahoga shows the smallest seasonal variation (35.6% in spring and 13.8% in fall). The Grand shows the greatest difference between summer and winter values (16.2 vs. 34.8%, respectively).
The ANVOCA results using four 3-mo seasons to analyze the concentration–flow relationship are consistent with those using two 6-mo seasons reported previously. For most rivers, the summer and fall results are similar, with the fall results being more negative or less positive. The main exception is the Grand, for which the summer downward trend in the concentration–flow relationship is substantially larger than the fall downward trend. The winter and spring trends are generally less strongly downward or more strongly upward than the summer and spring trends and more similar to each other than they are to the summer and spring trends. Neither the winter nor the spring trend is consistently higher than the other.
Summary of ANCOVA Results Relating Concentration and Flow
The direction of change in the concentration–flow relationship is the same for all seasons in a given river, with only minor exceptions, although some rivers are improving and others are not. Positive changes in the concentration–flow relationship are primarily found in the summer and fall seasons when flows are moderate to minor. Changes in the winter and spring relationships are generally less favorable when evaluated against the goal of reducing the concentration of sediment associated with a given flow. These results point to opportunities to improve winter and spring losses of sediment through increases in BMPs, such as the use of cover crops and the avoidance of fall tillage.
Given the differences in land use between the western, agricultural rivers and the eastern, more urban and forested watersheds, one might expect to see patterns in trends in the concentration–flow relationship that reflect these differences in land use. Such patterns are not obvious. The Maumee and Sandusky Rivers and Honey and Rock Creeks show similar patterns in all aspects of their trends and generally show improvements that seem to be responses to the agricultural management practices that have been applied in these basins.
The River Raisin, on the other hand, shows strong trends toward increasing sediment for a given flow, especially in winter and spring. Measured against the goal of reducing sediment concentrations and loads, it is performing the most poorly of all the watersheds studied. Although some of the Raisin data are suspected of contamination by poor sampling technique, removal of these data does not change the results, except for weakening the pattern for summer and fall. Land use in the basin involves more hay and pasture and less row crop than in the Maumee and Sandusky basins, but the land use has not changed greatly over the period of record, and the changes in agricultural acreage and crop selection parallel those in the Maumee and Sandusky. None of the geographic information available to us leads to a convincing hypothesis about why the Raisin behaves so differently than the Ohio agricultural rivers.
The Cuyahoga River also shows increases in the concentration–flow relationship. The Grand River, on the other hand, has shown the greatest decreases in the concentration–flow relationship, in spite of being a largely forested watershed not strongly targeted for practices to reduce sediment. These changes seem to reflect patterns and timing of urban/suburban development in the watersheds, but the land use changes that are thought to be involved are not well documented quantitatively. A causal relationship cannot be established.
This discussion reminds us that soil type, land use, regional changes in agricultural and other land management practices, weather, and unknown additional factors interact in complex and poorly understood ways to determine the water quality patterns that we observe. The convergence of a number of lines of evidence is needed to establish a credible cause-and-effect relationship.
In the case of the northwest Ohio agricultural watersheds, sustained downward trends in sediment concentrations and loads combined with increased conservation practices and the dominance of agricultural row-crop land use strongly suggest that the trends in sediment are due to changes in agricultural practices. The documentation of consistent decreases over time in the relationship of sediment concentration to flow in these tributaries argues against a weather-related cause for these trends and lends further support to the theory that agricultural management practices have succeeded in lowering sediment export from these watersheds.
Possible Bias in the Results
Two aspects of the methods used in this study may have introduced biases into the concentrations that are the basis for the trend analyses reported here. The USGS is a strong proponent of depth- and width-integrated sampling rather than sampling from a fixed point, as an autosampler does. In addition, they prefer to filter the entire sample rather than filtering an aliquot withdrawn after agitation of the sample, arguing that settling of coarser, heavier sediment fractions leads to a low bias when an aliquot is withdrawn. The magnitudes of these possible biases are difficult to evaluate. Point sampling could produce samples that are biased high or low compared with spatially integrated sampling, depending on the location of the point of sampling. The bias is likely to be smaller under high flow conditions, when turbulence leads to more complete mixing of suspended sediment throughout the water column. Suspended sediment in these rivers, especially in the Maumee and Sandusky, is strongly dominated by clay-sized particles, which are more likely to be well mixed in the river than coarser sediment. Such samples are also less likely to be strongly influenced by particle settling after agitation.
Although integrated sampling may be preferable, it is costly in time and human effort, and datasets of the size used in this study would not be feasible with this methodology. Indeed, USGS has stopped its daily sediment sampling program on these rivers.
Comparison of a limited set of data for the Maumee River, assembled by matching USGS sediment samples with NCWQR samples taken on the same day, indicates that the NCWQR sediment concentrations are, on average, about 90% of the USGS concentrations. It is unknown whether this level of bias applies to the other rivers.
Any possible bias in the sediment concentrations has been consistent throughout the study because the methods have remained the same. Thus, the bias is unlikely to affect the trend results, especially when they are expressed as percent change over time.
|
Conclusions |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMethodsResults and DiscussionConclusionsREFERENCES |
|---|
Over the 30-yr period of record, linear regression of the lnQ revealed no significant trends. Trends in lnSS, lnSSld, and SSdQ were in the same direction for a given river, increasing for the Raisin and Cuyahoga and decreasing for the other rivers. Most of these trends were statistically significant even after adjustment for autocorrelation. Rates of change ranged from decreases of more than 50% per decade to increases of more than 35% per decade.
Analysis of trends using two seasons revealed that the summer season generally had more substantial decreases or smaller increases (depending on the river) than the winter season. Analysis using four seasons revealed that summer and fall changes were similar, with fall decreases usually being larger. The spring season generally had the smallest decreases or the largest increases.
Analysis of trends for data divided into six flow classes showed that the largest percent decreases or smallest percent increases were often associated with the two low-flow categories. However, when these percent changes were applied to the average concentrations associated with the different flow categories, the largest decreases (smallest increases) in concentrations were associated with the higher-flow categories. Details differed from river to river.
Seasonal trends and trends by flow class are not totally independent of each other because some seasons have higher or lower flows than others. The winter season accounts for about 29% of the annual discharge, the spring season 42%, the summer season 22%, and the fall season 8%. Thus, the season with the highest flows shows the least improvement in sediment concentrations, and the season with the lowest flows shows the greatest improvement.
For the Cuyahoga and the Raisin, trends during the early part of the period of record are weakly upward and have been more strongly upward during the last decade or so. The Maumee and Sandusky Rivers and Honey and Rock Creeks show sustained or accelerating downward trends, especially when adjusted for SSdQ. For the Grand River, early trends were strongly downward, and more recent trends are flat to increasing.
Analysis of the relationship between lnQ and lnSS (the sediment rating curve) shows that this relationship has changed systematically over time (by a shift in the intercept of this linear relationship) in parallel with increasing or decreasing trends in sediment. This systematic change in the rating curve would not be expected as a result of changes in weather but is consistent with changes in land use.
Sustained decreases in sediment concentrations and loads and in the sediment rating curves for the agricultural Maumee and Sandusky Rivers and Honey and Rock Creeks seem to reflect the success of agricultural management programs in these watersheds in reducing erosion and delivery of sediment to the tributary system. The fact that the smallest changes are associated with the winter and spring—periods of above-average discharge—indicates that significant opportunities to further reduce sediment loads and concentrations exist in these seasons. An increased focus on BMPs, such as winter cover crops and avoidance of fall tillage, might pay large dividends here.
Causes for the recent increases in sediment concentrations and loads in the more urban/forested Cuyahoga River are less clear, as are the reasons for the early decline in sediment concentrations and loads in the Grand River. Local experts point to effects of development in these watersheds, but data needed to document a cause-and-effect relationship are lacking.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMethodsResults and DiscussionConclusionsREFERENCES |
|---|
|
REFERENCES |
|---|
|
TOPNOTESABSTRACTINTRODUCTIONMethodsResults and DiscussionConclusionsREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  R.P. Richards, D.B. Baker, and J.P. Crumrine Improved water quality in Ohio tributaries to Lake Erie: A consequence of conservation practices Journal of Soil and Water Conservation, May 1, 2009; 64(3): 200 - 211. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Agronomy Journal | Crop Science | |||
| Journal of Natural Resources and Life Sciences Education |
Vadose Zone Journal | ||||
| Soil Science Society of America Journal | Journal of Plant Registrations | The Plant Genome | |||
