Published online by Cambridge University Press: 07 November 2017
The aim of this study was to assess the temporal transferability of species distribution models (SDMs) and their potential implications for bird conservation. We quantified the loss and fragmentation of Montagu’s Harrier Circus pygargus and Common Kestrel Falco tinnunculus habitats over 13 years (2001–2014) in a highly dynamic landscape in north-western Spain. For this purpose, priority habitats for the target species were modelled at four different spatial scales using an ensemble forecasting framework. To explore the temporal transferability of our ensemble predictions, the models were back-projected to the land cover conditions in 2001 and evaluated using historical occurrence data. In addition, models calibrated with historical data were projected to the land cover conditions in 2014 and evaluated using updated occurrence data. Changes in availability and connectivity of suitable habitats between both years were estimated at four spatial scales from a set of widely-used indicators. SDMs showed a good predictive accuracy but with limited temporal transferability due to changes in the species-habitat relationships between 2001 and 2014. The results showed a decrease in the avaliability of suitable habitats of 33.4% and 47.7% for Montagu’s Harrier and Common Kestrel, respectively; with the subsequent increase in their fragmentation. However, our estimates were found to be strongly dependent on the scale of analysis and model transferability. Changes in habitat availability and connectivity ranged from -48% to +54% for Montagu’s Harrier, and from +116% to +5.6% for Common Kestrel. We call for caution when using SDMs beyond the model calibration time period to guide bird conservation. This is especially important for raptors, often characterised by low population sizes and large home ranges, and particularly sensitive to unstable, highly dynamic environmental conditions. In light of these results, specific, long-standing monitoring protocols remain essential to ensure accurate modelling performance and reliable future projections.