Full Proceedings is available at: http://www.extension.org/72817
The purpose of our work was to determine, within the southern region (AL, AR, FL, GA, KY, LA, MS, NC, OK, SC, TN, and TX), the feasibility of using different models to determine potential phosphorus loss from agricultural fields in lieu of phosphorus indices.
Estimation of phosphorus loss from agricultural land in the southern region of the usa using the apex, tbet, and aple models
1. Estimation of Phosphorus Loss From
Agricultural Land in the Southern Region of
the USA Using the APEX, TBET, and APLE
Models
Deanna Osmond, NC State University
Adam Forsberg and David Radcliffe, University of Georgia
John Ramirez, Mississippi State University
Dan Storm and Aaron Mittelstet, Oklahoma State University
Carl Bolster, ARS
Waste to Worth Conference
Seattle, WA
March 30 – April 2, 2015
4. Southern CIG: Objectives
1. Determine pre-existing watershed or plot-scale (11) sites
where accuracy of P Indices to estimate site P loss potential
can be evaluated.
2. Compare predictions of P-Indices to water quality data from
benchmark sites.
3. Compare fate and transport models (APEX, TBET, APLE)
against water quality data. Use water quality data
(monitored or predicted by model) to guide refinement of P
Indices.
4. Compare predictions of P Indices against fate and transport
water quality models (APEX, TBET, APLE) for calibrated and
uncalibrated models.
5. Refine P Indices to ensure better consistency in ratings across
state boundaries and within physiographic provinces.
5. Locations of Data Sets
TX
OK
FL
AL GA
AR
LA
NC
MS
TN
KY
SC
²0 250 500125 Miles
Albers Equal-Area Conic
6. Southern Field Sites
State # Plots Date range Site-years Crop STP range (ppm)
1 2 3 4
AR 7 2009 – 2011 21 Pasture 81 - 183 Captina (C)
GA 6 1995 – 1998 24 Pasture 14 - 142 Cecil (B) Altavista (C) Sedgefield (C) Helena (C)
NC 5 2011-2013 15
Corn with wheat
cover
44-121 Delanco (C)
MS 2 1996-1999 8
Cotton or soybens
with wheat cover
37-79 Dubbs (B) Tensas (D) Alligator (D) Dundee (C)
OK 1 1972-1976 4 Cotton 20 McLain (C) Reinach (C)
OK 1 2006-2007 1.17 Pasture 50 Clarksville (B)
OK 1 1977-1992 16 Native grass 15 Bethany (C)
OK 1 1980-1985 6 Wheat 35 Norge (B)
TX 1 1998-2001 4 Hay 435 Duffau (B)
TX 1 2005-2008 4 Sorghum/Oats 34 Topsey (C) Brackett (C) Krum (D)
TX 1 2005-2008 4 Native grass 10 Nuff (C)
TX 1 2001-2008 7
Corn with wheat
cover
51 Houston Black (D)
Soil Series (hydro group)
7. Texas BMP Evaluation Tool (TBET)
Climate
• Daily rainfall &
temperature
Soils
• Up to 3 series
Land use
• Crop system
Topography
• Field area
• Field slope
Soil Test P
• Mehlich III
Fertilization
8. TBET Model Process
Calibrated
Single year simulations run on a daily time-step
• (1/1/YYYY – 12/31/YYYY)
2 years of warm-up
• Initialize soil-moisture profile and nutrient pools
Compared model predictions to measured values on an event-basis
• Events within each year were summed for annual comparisons
Runoff events greater than 0.1 mm were compared
• If event-basis runoff spanned more than one day, total runoff for the
entire storm (up to three days) was lumped for analysis
Model evaluation
• Slope, intercept, R-squared
• Nash-Sutcliffe Efficiency
• Percent Bias
9. TBET Baseline Results: Runoff
Overall annual observed vs predicted runoff
y = 1.1204x + 39.535
R² = 0.7099
0
200
400
600
800
1000
0 200 400 600 800 1000
Simulatedrunoff(mm/yr)
Observed runoff (mm/yr)
GA
NC
MS
TX/OK
AR
Site
Linear Relationship
NSE PBIAS
Intercept Slope R2
Overall 40 1.1 0.7 0.3 34
AR 31 2.2 0.9 -5.1 164
GA 120 0.5 0.5 0.1 30
NC 104 1.1 0.7 0.0 46
MS 39 1.3 0.9 0.1 45
TX/OK -18 0.8 0.8 0.7 -30
10. TBET Baseline Results: Sediment
Overall annual observed vs predicted sediment
y = 2.6329x + 4.6557
R² = 0.1062
0
20
40
60
80
100
0 20 40 60 80 100
SimulatedSS(ton/ha/yr)
Observed SS (ton/ha/yr)
NC
MS
TX/OK
AR
Site
Linear Relationship
NSE PBIAS
Intercept Slope R2
Overall 4.7 2.6 0.1 -67.6 488
AR 0.0 3.8 0.3 -66.1 378
GA -- -- -- -- --
NC 20.8 7.1 0.4 -304.6 1698
MS 0.3 1.3 0.8 0.4 43
TX/OK 0.4 0.4 0.3 0.1 -40
11. TBET Baseline Results: Total P
Overall annual observed vs predicted total P
Site
Linear Relationship
NSE PBIAS
Intercept Slope R2
Overall 3.5 1.7 0.1 -26.6 166
AR 0.7 0.8 0.5 0.1 39
GA 2.1 0.3 0.1 -0.4 -43
NC 21.7 5.9 0.5 -158.3 961
MS 1.0 0.4 0.4 0.4 -16
TX/OK 0.6 0.4 0.2 0.1 -36
y = 1.6941x + 3.5249
R² = 0.1028
0
50
100
150
0 50 100 150
SimulatedTP(kg/ha/yr)
Observed TP (kg/ha/yr)
GA
NC
MS
TX/OK
AR
12. TBET Baseline Results: Dissolved P
Overall annual observed vs predicted dissolved P
y = 0.469x + 0.2736
R² = 0.4852
0
2
4
6
8
10
12
14
16
18
20
0 5 10 15 20
SimulatedDP(kg/ha/yr)
Observed DP (kg/ha/yr)
GA
NC
MS
TX/OK
AR
Site
Linear Relationship
NSE PBIASInterce
pt
Slop
e
R2
Overall 0.3 0.5 0.5 0.4 -41
AR 0.4 0.5 0.4 0.4 -10
GA 1.6 0.3 0.2 -0.2 -40
NC 0.2 0.3 0.5 0.4 -36
MS 0.0 0.3 0.9 -0.7 -72
TX/OK 0.0 0.2 0.2 -0.5 -77
13. TBET Preliminary Conclusions
TBET was used after being
calibrated
• Runoff predictions are satisfactory
with slight overprediction
• Sediment for AR & NC is
overpredicted
• Total P is affected by
overprediction of sediment and
underprediction of dissolved P,
which is systematically
underpredicted
• Modeling TBET was very time
consuming with uncertain
outcomes thus it may not be an
appropriate field-based tool for
predicting P loss especially if it is
uncalibrated
R2 NSE PBIAS
Runoff 0.74 0.42 22
Sediment 0.07 -77.61 489
Total P 0.08 -30.53 176
Dissolved P 0.49 0.40 -44
17. APEX Preliminary Conclusion (Uncalibrated)
•Acceptable model performance predicting runoff
•Very inaccurate predictions for phosphorus losses
•Inaccurate soil erosion prediction in small plots
•Tillage practices appears to be a factor that
determines model performance (e.g. overprediction or
underprediction)
•Model setup required additional information that was
not available in the southern databases and most
producers would not have this information either
•Modeling APEX was very time consuming with
uncertain outcomes thus it may not be an appropriate
field-based tool for predicting P loss especially if it is
uncalibrated
18. Annual P Loss Estimator (APLE)
•User-friendly spreadsheet
•Annual time step
•Requires runoff and
erosion as inputs
•Does not require
calibration
•Has most up-to-date
fertilizer and manure
application algorithm
19. APLE Model Process
Uncalibrated
Runoff and erosion values obtained from TBET model simulations
Does not require warm-up
Compared model predictions to measured values on an annual
basis
• Events within each year were summed for annual comparisons
Model evaluation
• Slope, intercept, R2
• Nash-Sutcliffe Efficiency (NSE)
• Percent Bias (PBIAS)
20. APLE Results: Total P
Overall annual observed vs predicted total P
Site
Linear Relationship
NSE PBIASInterce
pt
Slop
e
R2
Overall 3.4 1.5 0.2 -18 -140
AR 0.7 1.0 0.6 0.3 -64
GA 1.5 0.4 0.3 0.1 37
NC 15 4.4 0.4 -3 -670
MS -0.4 1.4 0.8 -18 -24
TX/OK 0.3 0.2 0.4 -0.1 66
Measured TP loss (kg/ha)
0 5 10 15 20 25
PredictedTPloss(kg/ha)
0
20
40
60
80
100
120
NC
MS
GA
AR
TX/OK
21. APLE Results: Dissolved P
Overall annual observed vs predicted dissolved P
Site
Linear Relationship
NSE PBIASInterce
pt
Slop
e
R2
Overall 0.99 0.6 0.5 0.5 4.3
AR 0.5 1.0 0.7 0.3 -51
GA 1.2 0.5 0.4 0.1 28
NC 1.6 0.5 0.1 -3 -180
MS 0.3 2.0 0.3 -18 -160
TX/OK 0.3 -0.1 0.03 -1.4 49
Measured DRP loss (kg/ha)
0 5 10 15
PredictedDRPloss(kg/ha)
0
2
4
6
8
10
12
14
16
NC
MS
GA
AR
TX/OK
22. APLE Preliminary Conclusions
• APLE is uncalibrated
• APLE uses modeled runoff and
erosion
• Dissolved P is better than total P
R2 NSE PBIAS
Runoff -- -- --
Sediment -- --- --
Total P 0.2 -18 -140
Dissolved P 0.5 0.5 4.3
23. Conclusions
• Flow generally predicted better
than sediment, TP or DP
• Modeling was very time
consuming with uncertain
outcomes thus it may not be an
appropriate field-based tool for
predicting P loss
North Carolina
Higher predictions from APEX settings
Soil loss and TP overprediction (both settings)
MUSLE predicted better than MUSS
Very inaccurate prediction for DP