Effects of species and habitat positional errors on the performance and interpretation of species distribution models

摘要: Aim A key assumption in species distribution modelling is that both and environmental data layers contain no positional errors, yet this will rarely be true. This study assesses the effect of introduced errors on performance interpretation models. Location Baixo Alentejo region Portugal. Methods Data steppe bird occurrence were collected using a random stratified sampling design 1-km2 pixel grid. Environmental sourced from satellite imagery digital maps. Error was deliberately into as shifts random direction 0–1, 2–3, 4–5 0–5 pixels. Whole habitat shifted by 1 to cause mis-registration, cumulative one three investigated. Distribution models built for three algorithms with replicates. Test compared controls without errors. Results Positional led drop model (larger having larger effects – typically up 10% area under curve average), although not enough for rejected. Model more severely affected inconsistencies contributing variables. Errors had similar lesser effects. Main conclusions Models are hard detect, often statistically good, ecologically plausible useful prediction, but interpreting them dangerous. Mis-registered produce smaller probably because shifting entire does break down correlation structure same extent individual observations. Spatial autocorrelation may protect against some relationship complex requires further work. The recommendation must should minimised through careful field processing.