An integrated approach to high-resolution modelling of a species range expansion using presence-only data

Electronic versions

Documents

  • Owain Barton

    Research areas

  • MScRes, School of Biological Sciences

Abstract

1. Predicting the potential distribution of species in novel areas is invaluable for
conservation planning and the utility of distribution modelling for conservation decisionmaking is well recognised. However, concern over the uncertainties and ecological relevance of techniques have limited their use by practitioners. Model functionality is constrained by the quality of available species data which imposes a trade-off between the resolution and scale that species-environment relationships can be estimated at. Distribution modelling is particularly challenging in non-equilibrium contexts, such as range expansion, due to the paucity of data appropriate for evaluation. Roe deer (Capreolus capreolus) in the UK are used as a case study to explore the efficacy of an integrated approach to modelling species range expansion using low-quality, presence-only data.
2. A total of 3,843 sightings of roe deer and eight ecogeographic predictor variables were used to produce a habitat suitability map using a MaxEnt model. Predictions of habitat suitability in a novel region (Wales) at a 100 m2 resolution were made using a model developed for a populated, neighbouring region (England and Scotland). The contribution of each variable to the model was assessed using a jackknife test, while the performance of the model in each region was evaluated using 10-fold and 푛 − 1 cross validation as well as a
novel, qualitative method based on the relative occurrence ratio (ROR). The map was integrated with a recently developed, mechanistic model (RangeShifter) to estimate the pattern of range expansion over time.
3. The key, fine-scale drivers of roe deer distribution in the UK were identified by the MaxEnt model with a strong association with woodland habitat. The area under the receiver-operating-curve (AUC) values from the 10-fold and 푛 − 1 cross validation were 0.794 ± 0.015 (mean ± standard deviation) and 0.803 ± 0.208, respectively for England and Scotland and 0.664 ± 0.073 and 0.672 ± 0.243 for Wales. The AUC and ROR results indicated that suitability in Wales may have been under-predicted by the MaxEnt model, although both methods are likely to be sensitive to sample size. The RangeShifter model described the expected pattern of range expansion across Wales. Based on established
estimates of expansion rates for free-ranging populations, it is predicted that 92% of suitable habitat will be occupied within 21 to 47 years.
4. The model predictions of this study support the development of a proactive management strategy for roe deer in Wales. Integration of correlative and mechanistic models enabled the prediction of species distribution at a higher resolution across a larger geographic scale than is typically achievable using presence-only data. Details of ecologically-driven decisions in the modelling process are provided to promote greater confidence in techniques and
encourage their application to a wide range of conservation objectives across taxa.

Details

Original languageEnglish
Awarding Institution
Supervisors/Advisors
Thesis sponsors
  • Knowledge Economy Skills Scholarship (KESS)
  • European Social Fund
Award date12 Sep 2018