Post provided by Emma Gardner and Tom Breeze
Pollination underpins >£600 million of British crop production and wild insects provide a substantial contribution to the productivity of many crops. There is mounting evidence that our wild pollinators are struggling and that pollinator populations may be declining. Reliably modelling pollinator populations is important to target conservation efforts and to identify areas at risk of pollination service deficits. In our study, ‘Reliably predicting pollinator abundance: Challenges of calibrating process-based ecological models’, we aimed to develop the first fully validated pollinator model, capable of reliably predicting pollinator abundance across Great Britain.
Poll4pop is a state-of-the-art, process-based pollinator model that predicts pollinator abundance by simulating how foraging bees move around the landscape. It includes population growth over time and dispersal of reproductive females.
Testing and Validating the Model
We compared the model’s predictions to real pollinator abundances observed along transects at 239 survey sites spread across Great Britain.
Landcover maps for each site were built using Centre for Ecology & Hydrology (CEH) Land Cover Map 2015, Ordnance Survey data, crop information from rural payments agencies and hedgerow locations from the CEH Woody Linear Features database.
We tested the model for four groups of bees (ground-nesting and tree-nesting bumblebees, and ground-nesting and cavity-nesting solitary bees).
We compared three different versions of the model using:
- Expert-opinion nesting/floral attractiveness scores, obtained from a survey of 10 UK pollinator experts.
- Data-driven nesting/floral attractiveness scores, obtained by calibrating the scores to get the best match to the observed pollinator abundances.
- Hybrid nesting/floral attractiveness scores, where we allowed the expert opinion estimates to inform the calibration process.
All three model versions showed significant agreement with the survey data for all four groups of bees. However, there were significant differences between the nesting/floral attractiveness scores in the three model versions. The calibration processes found ecologically unrealistic nesting/floral attractiveness scores for some habitats because of biases in the survey data and other factors.
Overall, we concluded that the model version using expert-opinion floral/nesting attractiveness scores is currently the most reliable/realistic version for making predictions.
Our results highlight a key universal challenge of calibrating spatially explicit, process-based ecological models. Notably, the desire to reliably represent complex ecological processes in finely mapped landscapes necessarily generates a large number of parameters, which are challenging to calibrate with ecological and geographical data that are often noisy, biased, asynchronous and sometimes inaccurate.
We therefore suggest a combined approach, where data-driven calibration and expert opinion are integrated into an iterative Delphi-like process that simultaneously combines model calibration and credibility assessment. This may offer the best way to obtain realistic parameter estimates and reliable model predictions for ecological systems with expert knowledge gaps and patchy ecological data.
To read the full study, check out the Methods in Ecology and Evolution article, ‘Reliably predicting pollinator abundance: Challenges of calibrating process-based ecological models’.