Title: Multimodeling and Model Abstraction Author
Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: October 8, 2009
Publication Date: October 14, 2009
Citation: Pachepsky, Y.A. 2009. Multimodeling and Model Abstraction. Interagency Environmental Modeling Workshop and Public Meeting. Rockville, MD, October 13-15, 2009 Meeting Abstract., p. 18. Technical Abstract: The multiplicity of models of the same process or phenomenon is the commonplace in environmental modeling. Last 10 years brought marked interest to making use of the variety of conceptual approaches instead of attempting to find the best model or using a single preferred model. Two systematic approaches to the concurrent use of several models are currently pursued. One approach is the model abstraction which consists in a systematic simplification of a complex model and generating a series of simpler models. Another approach is the multimodeling which consists in assigning weights to the simulation results from different models and using the weighted averages of these results. Multimodeling combines results from individual model to obtain a single prediction. Model abstraction uses models of different complexity to (a) learn more about the system, (b) improve robustness of predictions, (c) improve communication of modeling results, and (d) improve the performance of modeling systems. The review ov multimodeling and model abstraction methods will be given. The example will be presented that uses nineteen model to simulate flow in variably saturated soil. None of the individual models was calibrated. The multimodel based on the superensemble with the singular value decomposition had the accuracy and reliability comparable to the accuracy and reliability of fully calibrated model. The example will also be presented of the application of the model abstraction to characterize and understand flow and transport in soils in the presence of shallow groundwater. We developed case studies by carrying out several types of field tracer experiments at the USDA-ARS OPE3 Beltsville field site, and applying a sequence of model simplifications based on the TOUGH, HYDRUS and MODFLOW software families. Soil moisture, soil water potential, tracer concentrations in groundwater, groundwater levels, and weather data, along with ground penetration radar surveys, electric resistivity monitoring, and dilution tests complemented borehole log data and laboratory hydraulic measurements to characterize soil heterogeneity. The invoked series of model abstractions showed the important role of subsurface heterogeneity in the vadose zone and groundwater, and substantial improved the conceptualization of the subsurface. The advent of multimodeling and model abstraction may present both challenges and opportunities for software system design and implementation for environmental modeling. Additional flexibility required may be well compensated by the improved insights into systems behavior and predictions reliability.