California Wildlife Habitat Relationships System: Effects of county attributes on prediction accuracy for bird species

We evaluated how prediction accuracy of bird species occurrences from model s in the California Wildlife Habitat Relationships (CWHR) System were affec ted by county location, size, and number of wildlife habitats. Standardized lists of birds known to occur in 40 counties were compared to CWHR System predictions to calculate 5 measures of prediction accuracy. Multivariate ge neral linear models were used to assess the influence of county attributes on accuracy for 3 bird abundance groups. For ail species and common species , total and presence accuracies increased and absence accuracy and omission error decreased as the number of wildlife habitats increased. As county si re increased, commission error for all species decreased. For rare species, total and absence accuracies were greater and omission error was lower for inland counties than for coastal counties. Because accuracy was variously affected by different county attributes, depending on species abundance, im plications of prediction errors depend on the wildlife management task to w hich the models are applied. To minimize error, CWHR System queries should be done for areas as large and habitat as diverse as possible.