Recently, the US adopted a National Vegetation Classification (NVC) for the existing vegetation of all lands within the US.  The classification system is a "dynamic content" system where the classes and hierarchy are free to change over time through a peer review process.  The classification methodology is open, meaning that any of several approaches may be used to develop or revise the classes.  While an open methodology is desirable for many reasons, I argue that an effective classification system requires that:

• the classes are based on a system of formal logic
• the methods must be documented, and repeatable
• the quality of the results (homogeneity and distinctness of types) must be quantifiable

Results/Conclusions

I present a flexible framework for development of the National Vegetation Classification using formal quantitative methods, and for analyzing results of classifications developed using other methods.  Specifically, fuzzy relational, geometric, and taxa-based approaches are combined in an optimization framework to simultaneously address multiple classification quality criteria.  The PARTANA function analyzes within-class variability in composition compared to among-class variability in a global context (full dimensional compositional space) and identifies poorly defined classes or poorly assigned samples.  The SILHOUETTE function operates similarly in a more local context (nearby neighborhood in compositional space), and the DISDIAM function analyzes within-class variability across data sets to allow analysis of classification specificity across the NVC.  The proposed methods are freely available in the R statistical environment.  Examples of analyses for two regional data sets are presented.