From general research questions to specific answers: Underspecificity as a source of uncertainty in biological conservation

Species distribution modelling may support ecologists in conservation decision-making. However, the applicability of management recommendations depends on the uncertainty associated to the modelling process. A key source of uncertainty is the underspecificity of the research question. Modelling specific questions is straightforward since they drive clearly the methodological choices about input data and model building. Nevertheless, when the research questions remain underspecific, modellers must choose among a wide spectrum of choices, with each decision sequence driving to a different outcome that explain partially the target question. We show how the underspecificity associated to a general research question about Great Bustard breeding success at geographic scale drives to multiple decision choices, leads to a variety of model outcomes and hampers the identification of specific conservation actions. We ran generalised linear models using multi-model inference on a set of databases built according to specific sequences of methodological choices. Then, we evaluated variations in model performance, complexity (parsimony) and nature of predictors, as well as averaged model predictions and spatial congruence among model outputs. Deviance and parsimony varied widely (11.46% to 83.33% and 7 to 18, respectively), as did model averaged mean predictions in occupied areas, contributing predictors and spatial congruence among outputs (rPearson = 0.44 ± 0.23 for models calibrated in occupied areas; 0.48 ± 0.06 for models calibrated in potential/accessible areas). We recommend to carefully fix research questions and associated methodological options through collaborative working frameworks to conceptualize modelling approaches and, thus, to mitigate problems arising from underspecificity and other forms of uncertainty in conservation applications.