MMA: A Computer Code for Multimodel Analysis

Groundwater environments are complex and subject to multiple interpretations and mathematical descriptions. This recognition has led to a growing tendency among groundwater modelers to conduct multimodel analysis (MMA), as opposed to model analysis using a single model. MMA includes the following major steps: (1) postulate several alternative models for a site, (2) calibrate the models based on the same dataset, (3) rank these models using various criteria, (4) eliminate some of the less plausible models, and (5) weight and average predictions and statistics generated by the models (Neuman 2003; Ye et al. 2004, Poeter and Anderson 2005; Refsgaard et al. 2006). In addition to the conventional procedure of model development and calibration, MMA entails processing model calibration results for ranking the models, analyzing residuals/predictions of the models, calculating model probabilities or weights, and averaging model predictions. Postcalibration processing for many models (e.g., more than five) can be time-consuming, prone to human errors, and tedious, especially when the model calibrations are continually updated as new data and information are collected. Therefore, an automated postcalibration processing is crucial for MMA. The MMA computer code (Poeter and Hill 2007) was developed to perform automated postcalibration processing, which is made possible by utilizing model calibration results obtained from executing UCODE_2005 (Poeter et al. 2005) or any code that produces the needed files. Because UCODE_2005 is a universal code for sensitivity analysis, calibration, and uncertainty evaluation, MMA can be used for any set of models, not only groundwater models. This column provides a review of the basic features and capabilities of the MMA code.