BEST LINEAR UNBIASED PREDICTION OF HERBIVORE PREFERENCES

Generalized linear mixed models were used to obtain best linear unbias ed predictions (BLUP's) of herbivore preferences for range plant speci es from expert knowledge contained in range site descriptions produced by the USDA Soil Conservation Service (currently Natural Resources Co nservation Service). A total of 4,558 assessments of preference for ca ttle, deer, goats, and sheep on 167 plant species were available from 55 range site descriptions for the Edwards Plateau (Texas). Consistenc y of predicted preferences was evaluated through intraclass correlatio n estimated by restricted maximum likelihood. Predictions in observed (3-level ordinal) and legit scales were very similar; rank correlation s between predictions in different scales ranged from 0.994 (P < 0.000 1) for cattle to 0.998 (P < 0.0001) for sheep. Estimated intraclass co rrelations were also high (0.74 to 0.84 in observed scale and 0.82 to 0.92 in legit scale) suggesting consistent rankings of plant species a cross range sites. Metric multidimensional scaling and principal compo nents analysis showed distinct patterns among the 4 herbivores. Grasse s and browse were the most informative forage classes for discriminati ng preferences among herbivores, Deer and cattle exhibited the least s imilar diet preferences. Sheep and goats were intermediate; with sheep closer to cattle and goats most similar to deer. The pair deer-goat s howed the most similar pattern of preferences. BLUP's of plant species preferences showed good agreement with published research on both ind ividual plant species and forage classes. Optimal properties of mixed model procedures can be exploited to predict animal preferences at the range site scale from standardized expert opinion. These estimated pr eferences may be useful for modeling grazing effects at spatial scales compatible with management decisions.