Full Proceedings is available at: http://www.extension.org/72813 Phosphorus (P) indices are a key tool to minimize P loss from agricultural fields but there is insufficient water quality data to fully test them. Our goal is to use the Agricultural Policy/Environmental eXtender Model (APEX), calibrated with existing edge-of-field runoff data, to refine P indices and demonstrate their utility as a field assessment tool capable of protecting water quality. In this phase of the project our goal is to use existing small-watershed data from the Heartland Region (IA, KS, MO and NE) to determine the level of calibration needed for APEX before using the model to generate estimates of P loads appropriate for evaluating a P Index.

1. Estimation of phosphorus loss from agricultural
land in the Heartland region using the APEX model:
a first step to evaluating phosphorus indices.
J.A. Lory, N.O. Nelson, C. Baffaut, A. Senaviratne, M. Van Liew, A. Bhandari,
A Mallarino, M. Helmers, R. Udawatta, D. Sweeney and C. Wortmann
Waste to Worth Conference, Seattle, WA

2. • Modeling team
 Claire Baffaut, USDA-ARS, Univ. of Missouri
 Nathan Nelson, Kansas State Univ.
 Anomaa Senaviratne, Univ. of Missouri
 Ammar Bhandari, Kansas State Univ.
 Mike Van Liew, Univ. of Nebraska
• Runoff study managers
 Matt Helmers, Iowa State Univ.
 Ranjith Uddawatta, Univ. of Missouri
 Dan Sweeney, Kansas State Univ.
• P Index Developers
 John Lory, Univ. of Missouri
 Antonio Mallarino, Iowa State Univ.
 Nathan Nelson, Kansas State Univ.
 Charles Wortmann, Univ. of Nebraska
Regional P Index Assessment Project Team

3. Overall Project Approach
Two-step project
1. How well must we calibrate APEX to generate data to test P
Indices?
2. Use the appropriate strategy to generate data and evaluate P
Indices in IA, KS, MO and NE.

4. Overall Project Approach
Two-step project
1. How well must we calibrate APEX to generate data to test P
Indices?
2. Use the appropriate strategy to generate data and evaluate P
Indices in IA, KS, MO and NE.

5. Why Models?
Heartland region has an impressive amount of runoff
water quality data from small watersheds.
 200 site-years of data.
 But the limits:
• Only 27 soil-crop-management scenarios.
• Only 3 locations have more than 5 years under constant
management.

6. Why Models?
1. Use small watershed event data to calibrate APEX.
2. Use calibrated model to generate long-term average
estimates of losses needed to evaluate P Index.
But how much calibration
is needed?

7. In an ideal world of water quality modeling…
We would have tools to:
• Give feed-back on agricultural management
• Select optimum BMPs based on site characteristics
• Document water quality benefits from BMP implementation
APEX has been promoted for use with limited data
Can we use APEX to do this?
Is APEX reliable without calibration?

8. What is APEX?
Agricultural Policy / Environmental eXtender
Field- to small watershed-scale model
• 1 ac to 100,000 ac
Daily time step
• Provide temperature and rainfall as
daily inputs
Simulates
• Crop growth
• Nutrient cycling in soils
• Runoff
• Erosion
• Nutrient losses
3.16 ha
4.44 ha
1.65 ha
Grass waterways
Flumes
Novelty Watersheds
Greenley, MO

9. APEX Evaluation Objectives
Can APEX predict runoff, sediment loss, and P loss
without calibration?
• Best Professional Judgment (BPJ)
How well can APEX predict runoff, sediment loss, and P
loss with calibration?
• Full calibration
Can we develop a regional calibration for APEX?

10. Evaluation Datasets
Size: 0.4 – 5 ha.
Crops
• Corn/Soybean/Sorghum
• Pasture
Tillage
• No-till/Reduced till
Fertility
• Fertilizer
• Poultry litter
Structures
• Grass waterways
• Buffers
Initially focused on 19 small watersheds of four Tier 1 locations.

11. Best Professional Judgment Parameterization
Model options selected through best professional
judgment
SSURGO soils data (from web soil survey)
Management data from the site
Measured soil test P, total C, and total N.
Parameter file based on best professional judgment,
recommendations from model developers, and
published reports.
• Included discussion with model developers and revision of
APEX source code

12. Full Model Calibration
Start with best professional judgment parameterization
Add site-specific soils data
• Site-specific soil investigation, measured horizon depths
• Measured soil test P, total C, total N, and total P by horizon
• Measured texture (bulk density and hydraulic properties if
possible)
Sensitivity analysis based on model performance
• r2, Nash-Sutcliffe, percent bias, regression slope, minimum
square error
Manual calibration followed up with automated
parameter optimization.
• Event-based calibration at each Tier I site

13. Event-based Model Calibration - Runoff
Franklin County KS, Field 7

14. Event-based Model Calibration – P Loss
Franklin County KS, Field 7

15. “Annual” Comparisons
Event-based calibration for 19 watersheds at 4 locations
• Close communication with model developers to improve APEX
Data were summed at each location by year
• 97 site years of data
Evaluate accuracy of APEX predictions across multiple
sites and management

16. Compare event simulated values vs measured data
Model performance indicators
Coefficient of determinationr2
• Quantifies the correlation between measured and simulated values
• The proportion of the variability in the observed data that is explained by the model
• 0 shows no correlation, 1 shows perfect match
Nash-Sutcliffe EfficiencyNSE
• Compares the model predictions to a simple arithmetic average of measured values.
• Fit to a 1:1 line
• 𝑁𝑆𝐸 = 1 − 𝑖=1
𝑁
𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑑 𝑖 − 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑒𝑑 𝑖
2
𝑖=1
𝑁 (𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑑 𝑖 − 𝐴𝑣 𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑑)2,
• NSE = 0: no better than average (can be negative)
• NSE = 1: perfect match.

17. Runoff
BPJ = Best Professional Judgment

18. Erosion
BPJ = Best Professional Judgment

19. P Loss
BPJ = Best Professional Judgment

20. Conclusions
Relying on best professional judgment produced unsatisfactory
results for simulated runoff, sediment, and total P loss.
With full calibration, APEX can produce satisfactory simulation of
runoff, sediment, and total P loss
• Improved runoff with calibration
• Greatly improved simulation of P loss
Regional parameterization may be successful alternative to best
professional judgment

21. Looking ahead
 Current research: How much calibration is needed to
provide an effective model to predict P loss?
• No calibration? No
• Regional calibration? ??
• Full calibration? Yes
 How much impact does calibration strategy have on
evaluation results of P index?
• Do I get different recommendation with a better calibrated
model?

22. • Modeling Support
• Jimmy Williams
• Jaehak Jeong
• Funding: USDA-NRCS Conservation Innovation Grant
Acknowledgements