Skip to main content
ARS Home » Research » Publications at this Location » Publication #319797
Title: Uncertainity and equifinality driven by rainfall in the APEX model

Author
item PRADA, ANDRES - University Of Illinois
item CHU, MA - University Of Illinois
item GUZMAN, JORAGE - Waterborne Environmental
item Moriasi, Daniel
item King, Kevin
item Bosch, David
item Bjorneberg, David - Dave
item Teet, Stephen

Submitted to: Transactions of the ASABE
Publication Type: Proceedings
Publication Acceptance Date: 7/15/2015
Publication Date: 7/29/2015
Citation: Prada, A.F., Chu, M.L., Guzman, J.A., Moriasi, D.N., King, K.W., Bosch, D.D., Bjorneberg, D.L., Teet, S.B. 2015. Uncertainity and equifinality driven by rainfall in the APEX model. Transactions of the ASABE. Available: http://www.asabe.org/meetings-events/2015/07/2015-annual-international-meeting.aspx.

Interpretive Summary: Uncertainty is an inherent part of complex environmental models. Uncertainty in model inputs, model parameterization, and model structure can propagate non-linearly to the model outputs. Evaluating, quantifying, and reporting uncertainty is crucial when model results are used as basis for managerial decisions and policies. Results should be presented with the full disclosure of the risks associated with uncertainty of the outputs. In this study, we evaluated the uncertainty and equifinality of the Agricultural Policy/Environmental eXtender (APEX) model for a sub-watershed in the Upper Big Walnut Creek in Ohio. Three APEX models were developed using three different rainfall datasets: (1) estimated from 38 NOAA stations surrounding the watershed; (2) measured in the watershed; and (3) generated from PRISM models. A two-step probabilistic approach to calibrate the model was implemented using global uncertainty and sensitivity analysis. A preliminary analysis was conducted using 22 uncertain global parameters. Each parameter was assigned a uniform distribution with ranges derived from measurements, literature, and model range validity. Sampling of the probability distribution functions was performed using the Sobol method. Acceptable models were evaluated using the Nash-Sutcliffe Efficiency Coefficient. Preliminary results indicated that rainfall datasets rather than parameter ranges were driving model uncertainty and equifinality. Model results using the NOAA dataset have the highest model efficiency but also the highest uncertainty and equifinality. Quantifying uncertainty and equifinality can improve model result understanding, increase model robustness, and help practitioners identify the validity of model outcome ranges.

Technical Abstract: Uncertainty is an inherent part of complex environmental models. Uncertainty in model inputs, model parameterization, and model structure can propagate non-linearly to the model outputs. Evaluating, quantifying, and reporting uncertainty is crucial when model results are used as basis for managerial decisions and policies. Results should be presented with the full disclosure of the risks associated with uncertainty of the outputs. In this study, we evaluated the uncertainty and equifinality of the Agricultural Policy/Environmental eXtender (APEX) model for a sub-watershed in the Upper Big Walnut Creek in Ohio. Three APEX models were developed using three different rainfall datasets: (1) estimated from 38 NOAA stations surrounding the watershed; (2) measured in the watershed; and (3) generated from PRISM models. A two-step probabilistic approach to calibrate the model was implemented using global uncertainty and sensitivity analysis. A preliminary analysis was conducted using 22 uncertain global parameters. Each parameter was assigned a uniform distribution with ranges derived from measurements, literature, and model range validity. Sampling of the probability distribution functions was performed using the Sobol method. Acceptable models were evaluated using the Nash-Sutcliffe Efficiency Coefficient. Preliminary results indicated that rainfall datasets rather than parameter ranges were driving model uncertainty and equifinality. Model results using the NOAA dataset have the highest model efficiency but also the highest uncertainty and equifinality. Quantifying uncertainty and equifinality can improve model result understanding, increase model robustness, and help practitioners identify the validity of model outcome ranges.