Error and uncertainty in habitat models

摘要: 1. Species distribution models (habitat models) relate the occurrence or abundance of a species to environmental and/or geographical predictors that then allow predictions be mapped across an entire region. These are used in range policy settings such as managing greenhouse gases, biosecurity threats and conservation planning. Prediction errors almost ubiquitous habitat models. An understanding source, magnitude pattern these is essential if transparently decision making. 2. This study considered sources It divided them into two main classes, error resulting from data deficiencies introduced by specification model. Common important included missing covariates, samples species' occurrences were small, biased lack absences. affected types could developed probable would occur. Almost all had this significant spatial correlation analysis. 3. A challenging aspect modelling distributions processes operating both space. We differentiated between global (aspatial) local (spatial) errors, discussed how they arise what can done alleviate their effects. 4. Synthesis applications. brings together statistical ecological thinking consider appropriate techniques for modelling. Ecological theory suggests capable defining optima, while allowing interactions variables. Statistical considerations, including impacts suggest deal with multimodality discontinuity response surfaces. Models typically simple approximations true probability surface. use flexible regression techniques, explain makes methods superior The most robust approaches likely those which care taken match model knowledge ecology, each allowed inform other. © 2006 Bureau Rural Sciences.