Generalizability of songbird habitat models in boreal mixedwood forests of Alberta

摘要: Abstract Statistical models relating habitat characteristics to species occurrences are increasingly used evaluate the consequences of forest management activities and conservation plans over large spatial temporal scales. In practice, such do not always generalize other locations, hence, they should be validated using independent data. this paper, we assess predictive ability 16 songbird developed in Calling Lake region boreal mixedwood Alberta both internal external validation approaches. Internal relied on same dataset develop while utilised data collected within four landscapes ecological region. Two aspects accuracy were evaluated: agreement between observations predicted values (calibration) models’ classify locations into those which present or absent (discrimination). Calibration was assessed Hosmer–Lemeshow (H–L) statistic plots showing versus observed probabilities occurrence. Discrimination receiver operating characteristic (ROC) curves associated area under ROC curve. With validation, calibration reasonable for all species, however H–L indicated a good fit only eight species. Model discrimination occupied unoccupied sites, hand, 14 models. identified three with fit. The remaining generally over- under-predicted probability External model 10 When re-estimated available data, again indicating 12 Similarly, Several factors may help explain why performance poorer more variable than when These include differences landscape structure disturbance history as well frequency occurrence individual Overall, our analyses several whose degraded considerably, especially measured by calibration. By re-estimating models, increased range variation covariates likely led an improvement We discuss importance ongoing evaluation refinement that will planning scenarios at