Page Banner

United States Department of Agriculture

Agricultural Research Service

Research Project: OBJECT MODELING AND SCALING OF LANDSCAPE PROCESSES AND CONSERVATION EFFECTS IN AGRICULTURAL SYSTEMS Title: Assessing Modeled Spatial and Temporal Variability in Soil Moisture Using Automated, Multi-Objective, Step-wise Calibration

Authors
item Strudley, Mark
item Green, Timothy
item Erskine, Robert
item David, Olaf - COLORADO STATE UNIV.
item Umemoto, Makiko - USGS/CU

Submitted to: Annual Hydrology Days Conference Proceedings
Publication Type: Abstract Only
Publication Acceptance Date: February 28, 2007
Publication Date: March 19, 2007
Citation: Strudley, M.W., Green, T.R., Erskine, R.H., David, O., Umemoto, M. 2007. Assessing Modeled Spatial and Temporal Variability in Soil Moisture Using Automated, Multi-Objective, Step-wise Calibration. Annual Hydrology Days Conference Proceedings.

Interpretive Summary: The predictive capability of many environmental models is commonly hampered by a profuse set of parameters that are often physically ambiguous and costly to measure in the field. These parameters are fundamental to model formulation and behavior because they help define and quantify the specific environmental processes encapsulated by the model (e.g., runoff, infiltration, etc.). Usually these parameters are assigned values (called model parameter estimation) by experienced scientists, but this rarely allows the task to be fully objective, and it is difficult or impossible for even experienced scientists to be able to “know” the appropriate values for parameters a priori in complex models. Automated procedures have been developed on the computer for accomplishing this task according to predefined rule sets, but model parameter estimation is sometimes hindered by complex dynamical model feedbacks and interactions, resulting in the inability of many automated procedures to identify optimal parameter sets. In the case of modeling and calibration endeavors for field-scale environmental models in agricultural settings, little work has been done in attempting to assess and correct model formulation at different points in space based on parameter estimation and optimization. We attempt to remedy this situation by examining the performance of a model used to estimate depth-dependent soil moisture variability in a dryland agricultural setting (the Root Zone Water Quality Model, RZWQM) in Colorado at different landscape positions within a cropped field using multiple-objective, step-wise calibration. This method employs a “shuffled complex evolution” algorithm which searches for globally optimal parameter sets by modeling the parameter estimation process as an analog to competitive genetic evolution. New parameter sets are generated iteratively (as “offspring), and “surviving” sets propagate according to their “fitness”, or how well they provide model output that matches measured data. In this exercise we examine the calibration of hydraulic parameters that govern water flow and retention in the soil. The incorporation of such parameter optimization and structural assessment techniques into the modeling “toolbox” of government agencies such as the Agricultural Research Service will aid the design and implementation of more accurate predictive tools to advise sound economic and environmental management of agricultural resources.

Technical Abstract: The predictive capability of many environmental models is commonly hampered by a profuse set of parameters that are often physically ambiguous and costly to measure in the field. Furthermore, model parameter estimation is sometimes hindered by complex dynamical model feedbacks and interactions, resulting in the inability of many automated calibration procedures to identify optimal parameter sets. In the case of modeling and calibration endeavors for field-scale environmental models in agricultural settings, little work has been done in attempting to assess and correct model structure at different points in space based on parameter optimization. We attempt to remedy this situation by examining the performance of a vertically-distributed model used to estimate depth-dependent soil moisture variability in a dryland agricultural setting (the Root Zone Water Quality Model, RZWQM) in Colorado at different landscape positions within a cropped field using multiple-objective, step-wise calibration. This method employs a shuffled complex evolution algorithm which uses competitive evolution and complex shuffling to search for globally optimal parameter sets. In this exercise we examine the calibration of hydraulic parameters that govern the formulation of the Brooks and Corey description of the soil water retention curve. The incorporation of such parameter optimization and structural assessment techniques into the modeling “toolbox” of government agencies such as the Agricultural Research Service will aid the design and implementation of more accurate predictive tools to advise sound economic and environmental management of agricultural resources. Future work will examine model performance discretely in space by using spatial dynamic identifiability analysis to quantify the marginal probability distribution of a parameter in terms of spatial statistics such as contributing area, local slope, and partial contributing area.

Last Modified: 10/22/2014
Footer Content Back to Top of Page