Developing a predictive modelling capacity for a climate change-vulnerable blanket bog habitat: Assessing 1961-1990 baseline relationships

Aim: Understanding the spatial distribution of high priority habitats and developing predictive models using climate and environmental variables to replicate these distributions are desirable conservation goals. The aim of this study was to model and elucidate the contributions of climate and topography to the distribution of a priority blanket bog habitat in Ireland, and to examine how this might inform the development of a climate change predictive capacity for peat-lands in Ireland.Methods: Ten climatic and two topographic variables were recorded for grid cells with a spatial resolution of 10×10 km, covering ~87% of the mainland land surface of Ireland. Presence-absence data were matched to these variables and generalised linear models (GLMs) fitted to identify the main climatic and terrain predictor variables for occurrence of the habitat. Candidate predictor variables were screened for collinearity, and the accuracy of the final fitted GLM was evaluated using fourfold cross-validation based on the area under the curve (AUC) derived from a receiver operating characteristic (ROC) plot. The GLM predicted habitat occurrence probability maps were mapped against the actual distributions using GIS techniques.Results: Despite the apparent parsimony of the initial GLM using only climatic variables, further testing indicated collinearity among temperature and precipitation variables for example. Subsequent elimination of the collinear variables and inclusion of elevation data produced an excellent performance based on the AUC scores of the final GLM. Mean annual temperature and total mean annual precipitation in combination with elevation range were the most powerful explanatory variable group among those explored for the presence of blanket bog habitat.Main conclusions: The results confirm that this habitat distribution in general can be modelled well using the non-collinear climatic and terrain variables tested at the grid resolution used. Mapping the GLM-predicted distribution to the observed distribution produced useful results in replicating the projected occurrence of the habitat distribution over an extensive area. The methods developed will usefully inform future climate change predictive modelling for Ireland.