Evaluating expert opinion and spatial scale in an amphibian model

When ecological models are used to guide conservation decisions, these models should be based upon substantial data and should be applied at appropriate spatial scales. Yet, ecologists are usually faced with scarce data and must often make subjective choices about scale. To handle limited data, the use of expert panels to parameterize models has become common. However, few studies evaluate the success of expert panels in improving models. In this study, I examine a recent resistant kernel model designed to prioritize amphibian breeding habitat for conservation. I compare the predictive ability of the model as originally parameterized by an expert panel to the predictive ability of simpler models. I optimize parameter values for spatial scale and landscape resistance using 896 ponds from 5 studies of spotted salamanders (Ambystoma maculatum) and wood frogs (Lithobates sylvaticus) in Massachusetts and Rhode Island. In predicting amphibian distributions, models examined in this study that relied upon expert-derived resistance values performed worse than null models with uninformative resistance values. The failure of the resistant kernel model offers support for the use of simple models in the face of complex ecological problems. The best scale for measuring upland habitat in these models was in the range of 1000–3000 m, an order of magnitude larger than the salamander migration scale previously proposed for wetland buffer zones.

► I evaluated models that use land cover to predict amphibian distributions. ► Expert opinion made the models worse. ► Scales between 1 km and 3 km were best for measuring fragmentation.