CARMA: Scalability with Approximate-Model-Based Adaptation

Many complex physical systems such as biological systems are characterized both by incomplete models and limited empirical data. Accurate prediction of the behavior of such systems requires exploitation of multiple, individually incomplete, knowledge sources. Our approach, called approximate-model-based adaptation, utilizes case-based reasoning to provide an approximate solution and model-based reasoning to adapt this approximation into a more precise solution. This approach is implemented in CARMA, a decision-support system for grasshopper infestation advising which models experts and has been successfully used since 1996. Initially focused on rangeland grasshoppers within the state of Wyoming, CARMA’s capabilities have been extended to support the development and implementation of more environmentally friendly and sustainable strategies and to support advising in nine additional western U.S. states. This paper details our approach to scaling CARMA to the wider geographic region. Prior research indicated that completeness of the model-based knowledge used for matching and adaptation is more important to CARMA’s accuracy than coverage of the case library. Given the importance of the model as a tool for refinement and accuracy, and that the cases are mostly void of region-specific information, our approach is thus to continue using the cases without changes as a general source of approximate predictions, and to extend the region-specific historical information required by the model as necessary to provide regional accuracy. The relative ease with which CARMA has been scaled thus far lends confirmation to the fact that CARMA’s modeling of the experts is accurate.