About Us | Help Videos | Contact Us | Subscriptions
 

Book: Methods of Introducing System Models into Agricultural Research
Published by: American Society of Agronomy, Crop Science Society of America, Soil Science Society of America

 

This chapter in METHODS OF INTRODUCING SYSTEM MODELS INTO AGRICULTURAL RESEARCH

  1.  p. 327-363
    Advances in Agricultural Systems Modeling 2.
    Methods of Introducing System Models into Agricultural Research

    Laj R. Ahuja and Liwang Ma (ed.)

    ISBN: 978-0-89118-196-5

     

 View
 Download
 Alerts
 Permissions
 Share

doi:10.2134/advagricsystmodel2.c12

Inverse Modeling with RZWQM2 to Predict Water Quality

  1. Bernard T. Nolan,
  2. Robert W. Malone,
  3. Liwang Ma,
  4. Christopher T. Green,
  5. Michael N. Fienen and
  6. Dan B. Jaynes
  1. B.T. Nolan, USGS, 413 National Center, Reston, VA 20192 (btnolan@usgs.gov); R.W. Malone (rob.malone@ars.usda.gov) and D.B. Jaynes (dan.jaynes@ars.usda.gov), USDA-ARS, National Laboratory for Agriculture and the Environment, 2110 University Boulevard, Ames, IA 50011-0001; L. Ma, USDA-ARS, Agriculture Systems Research, 2150 Centre Ave. Bldg. D, Fort Collins, CO 80526 (liwang.ma@ars.usda.gov); C.T. Green, USGS, Bldg 15, McKelvey Building, 345 Middlefield Road, Menlo Park, CA 94025-3561 (ctgreen@usgs.gov); Michael N. Fienen, USGS, Wisconsin Water Science Center, 8505 Research Way, Middleton, WI 53562 (mnfi enen@usgs.gov).

Abstract

This chapter presents guidelines for autocalibration of the Root Zone Water Quality Model (RZWQM2) by inverse modeling using PEST parameter estimation software (Doherty, 2010). Two sites with diverse climate and management were considered for simulation of N losses by leaching and in drain flow: an almond [Prunus dulcis (Mill.) D.A. Webb] orchard in the San Joaquin Valley, California and the Walnut Creek watershed in central Iowa, which is predominantly in corn (Zea mays L.)–soybean [Glycine max (L.) Merr.] rotation. Inverse modeling provides an objective statistical basis for calibration that involves simultaneous adjustment of model parameters and yields parameter confidence intervals and sensitivities. We describe operation of PEST in both parameter estimation and predictive analysis modes. The goal of parameter estimation is to identify a unique set of parameters that minimize a weighted least squares objective function, and the goal of predictive analysis is to construct a nonlinear confidence interval for a prediction of interest by finding a set of parameters that maximizes or minimizes the prediction while maintaining the model in a calibrated state. We also describe PEST utilities (PAR2PAR, TSPROC) for maintaining ordered relations among model parameters (e.g., soil root growth factor) and for post-processing of RZWQM2 outputs representing different cropping practices at the Iowa site. Inverse modeling provided reasonable fits to observed water and N fluxes and directly benefitted the modeling through: (i) simultaneous adjustment of multiple parameters versus one-at-a-time adjustment in manual approaches; (ii) clear indication by convergence criteria of when calibration is complete; (iii) straightforward detection of nonunique and insensitive parameters, which can affect the stability of PEST and RZWQM2; and (iv) generation of confidence intervals for uncertainty analysis of parameters and model predictions. Composite scaled sensitivities, which reflect the total information provided by the observations for a parameter, indicated that most of the RZWQM2 parameters at the California study site (CA) and Iowa study site (IA) could be reliably estimated by regression. Correlations obtained in the CA case indicated that all model parameters could be uniquely estimated by inverse modeling. Although water content at field capacity was highly correlated with bulk density (−0.94), the correlation is less than the threshold for nonuniqueness (0.95, absolute value basis). Additionally, we used truncated singular value decomposition (SVD) at CA to mitigate potential problems with highly correlated and insensitive parameters. Singular value decomposition estimates linear combinations (eigenvectors) of the original process-model parameters. Parameter confidence intervals (CIs) at CA indicated that parameters were reliably estimated with the possible exception of an organic pool transfer coefficient (R45), which had a comparatively wide CI. However, the 95% confidence interval for R45 (0.03–0.35) is mostly within the range of values reported for this parameter. Predictive analysis at CA generated confidence intervals that were compared with independently measured annual water flux (groundwater recharge) and median nitrate concentration in a collocated monitoring well as part of model evaluation. Both the observed recharge (42.3 cm yr−1) and nitrate concentration (24.3 mg L−1) were within their respective 90% confidence intervals, indicating that overall model error was within acceptable limits.

  Please view the pdf by using the Full Text (PDF) link under 'View' to the left.

Copyright © 2011. Copyright © 2011 ASA, CSSA, SSSA, 5585 Guilford Rd., Madison, WI 53711-5801, USA.