1. T. Kevin O’Donnell1, Claire Baffaut2, Stephen H. Anderson1
1 Department of Soil, Environmental and Atmospheric Sciences,
University of Missouri, Columbia, Missouri USA
2 USDA-ARS, Cropping Systems and Water Quality Research Unit,
Columbia, Missouri, USA
1
Evaluation of Water Quality Trends in
Goodwater Creek Experimental Watershed:
Implications for Monitoring Strategies and
Objective Setting
2. Upper Mississippi River Basin:
River Restoration
Limited accountability
for projects
Poor project objective
setting
Poorly designed
monitoring plans
Decreased chances to
demonstrate success
2
3. Study Objectives
Assess BMP Effects in the Goodwater Creek
Experimental Watershed (GCEW)
Atrazine Event Loads
Nitrate Event Loads
Evaluate GCEW Monitoring Network
Goal: “Documenting Impact of Prevailing
Farming Systems on Herbicide Contamination
and Transport”
Future Objective Setting for BMP
Implementation
3
4. Methods
GCEW Monitoring Data: Hydrologic Data
Stream Stage at Three
Weirs
12, 31, 73 km2
Precipitation
Max. 10 rainfall gages
4
5. Methods
GCEW Monitoring Data: Atrazine & Nitrate
Autosamplers
Weir 1
Atrazine (1993 to 2006)
Nitrate (1993 to 2004)
Weir 9 (1993 to April 1997)
Weir 11 (1993 to April 1997)
Weekly Grab Samples
5
13. Methods
Gradual Change, Multiple Regression Model
Y = atrazine/nitrate event load (kg)
β0 = intercept
F = 4 streamflow/hydrograph predictors
P = 9 rainfall predictors
BMP = chronology of GCEW area affected by 5
BMP types (ha)
є = residual
εβ ++++= BMPPFY 0
13
14. Methods
Gradual Change, Multiple Regression Model
Step 1: Identify best subset of streamflow/rainfall
predictors by:
R-squared (r2)
Akaike’s Information Criterion (AIC)
Remove temporal autocorrelation of residuals
εβ ++++= BMPPFY 0
14
15. Methods
Gradual Change, Multiple Regression Model
Step 2: Identify/quantify BMP effects on loads:
Akaike’s weights (wi)
All possible combinations of BMP predictors
included in model
εβ ++++= BMPPFY 0
15
16. Methods
Gradual Change, Multiple Regression Model
BMP Predictors
Grass waterways
Terraces drained by grass waterways
Terraces drained by underground outlets
Vegetative filter strips
Permanent vegetation
32 possible model combinations ranked by wi (0-1)
εβ ++++= BMPPFY 0
16
17. Methods
Future Monitoring Needed for Reduction Scenarios
n = number of sampled events needed to show a %
change in mean atrazine/nitrate load
MSE = mean squared error of regression models
t = two tailed, Student’s t value at an α=0.05
D = % change in mean atrazine/nitrate event load
5%, 10%, 20%, and 25% change
2
)*2(
2
**2
D
tMSE
n
knpre −
=
17
26. Discussion
No Evidence of BMP Effects: Atrazine
No watershed-scale
objectives
No-till adoption
~30% ↑ in county crop area
20% ↑ in runoff
120% ↑ in atrazine loss
↑ Split-application of
atrazine?
26
28. Discussion
No Evidence of BMP Effects: Nitrate
Unintended effects
of terraces
Groundwater
leaching?
BMP choice for
reducing nitrate
Winter cover crop
(Blevins et al. 1996)28
29. Discussion
Potential Errors in Event Load Determinations
Delay in 1st
autosample
77% of Weir 1
sampled events
Mean: 8.4 hr delay
Mean: 20% volume
missed
29
30. Discussion
Potential Errors in Event Load Determinations
Delay in 1st
autosample
77% of Weir 1
sampled events
Mean: 8.4 hr delay
Mean: 20% volume
missed
30
31. Discussion
Potential Errors in Event Load Determinations
Delay in 1st
autosample
77% of Weir 1
sampled events
Mean: 8.4 hr delay
Mean: 20% volume
missed
31
32. Discussion
Potential Errors in Event Load Determinations
Delay in 1st
autosample
77% of Weir 1
sampled events
Mean: 8.4 hr delay
Mean: 20% volume
missed
Sample compositing
between events
22% of Weir 1 events
32
33. Discussion
Potential Errors in Event Load Determinations
Delay in 1st
autosample
77% of Weir 1
sampled events
Mean: 8.4 hr delay
Mean: 20% volume
missed
Sample compositing
between events
22% of Weir 1 events
33
34. Discussion
Monitoring Goals: Atrazine vs. Nitrate
Goal: “Documenting Impact of Prevailing Farming
Systems on Herbicide Contamination and Transport”
Nitrate r2: 0.76
Atrazine r2: 0.42
Prevailing Farming Systems are not monitored
Targeted monitoring needed
Decrease potential errors in load by adapting flow-
proportional autosampling?
34