A comparison of modelling techniques for small mammal diversity

Development pressures frequently dictate that managers’ need to make decisions about which local sites will be developed and which will be protected. When management for diversity is the goal, it would be helpful if models could aid these decisions. We compared three methods for modelling site-specific small mammal diversity at 48 0.58-ha study sites distributed within six habitats in foothills of the Sacramento Mountains, south-central New Mexico, spring and fall, 1993–1994. Methods included; 1) direct richness prediction with discriminant analysis (classification success rate of 15.1%, mean error=1.6 species), 2) prediction of richness based upon expected species-specific habitat suitability with discriminant analysis (classification success rate 20.3%, mean error=1.6 species), and 3) prediction of relative richness (high vs. low) (classification success rate=91.1%). The mean error of methods 1 and 2 (1.6 species) exceeds the difference known to distinguish high richness habitats from low (1.3 species) in this ecosystem. Therefore, we conclude that the appropriate conceptual technique for modelling diversity is to proceed by distinguishing high and low diversity habitats. We found this technique preferable when compared to pursuit of error-prone models for actual richness that have mean errors larger than those known to characterize the system.