You are here:
Sobol’ sensitivity analysis for stressor impacts on honeybee colonies
Citation:
Kuan, C., R. Curry, G. DeGrandi-Hoffman, K. Garber, A. Kanarek, M. Snyder, AND Tom Purucker. Sobol’ sensitivity analysis for stressor impacts on honeybee colonies. In Proceedings, 8th International Congress on Environmental Modelling and Software, Toulouse, FRANCE, July 10 - 14, 2016. International Environmental Modelling and Software Society, Manno, Switzerland, 27, (2016).
Impact/Purpose:
To be part of the proceedings of the 8th International Congress on Environmental Modelling and Software
Description:
We employ Monte Carlo simulation and nonlinear sensitivity analysis techniques to describe the dynamics of a bee exposure model, VarroaPop. Daily simulations are performed of hive population trajectories, taking into account queen strength, foraging success, mite impacts, weather, colony resources, population structure, and other important variables. This allows us to test the effects of defined pesticide exposure scenarios versus controlled simulations that lack pesticide exposure. The daily resolution of the model also allows us to conditionally identify sensitivity metrics. We use the variancebased global decomposition sensitivity analysis method, Sobol’, to assess firstand secondorder parameter sensitivities within VarroaPop, allowing us to determine how variance in the output is attributed to each of the input variables across different exposure scenarios. Simulations with VarroaPop indicate queen strength, forager life span and pesticide toxicity parameters are consistent, critical inputs for colony dynamics. Further analysis also reveals that the relative importance of these parameters fluctuates throughout the simulation period according to the status of other inputs. Our preliminary results show that model variability is conditional and can be attributed to different parameters depending on different timescales. By using sensitivity analysis to assess model output and variability, calibrations of simulation models can be better informed to yield more accurate predictions of complex ecological processes.
