A comparative analysis of parallel processing and super-individual methods for improving the computational performance of a large individual-based model

Individual-based modelling approaches are being used to simulate larger complex spatial systems in ecology and in other fields of research. Several novel model development issues now face researchers: in particular how to simulate large numbers of individuals with high levels of complexity, given finite computing resources. A case study of a spatially-explicit simulation of aphid population dynamics was used to assess two strategies for coping with a large number of individuals: the use of ‘super-individuals’ and parallel computing. Parallelisation of the model maintained the model structure and thus the simulation results were comparable to the original model. However, the super-individual implementation of the model caused significant changes to the model dynamics, both spatially and temporally. When super-individuals represented more than around 10 individuals it became evident that aggregate statistics generated from a super-individual model can hide more detailed deviations from an individual-level model. Improvements in memory use and model speed were perceived with both approaches. For the parallel approach, significant speed-up was only achieved when more than five processors were used and memory availability was only increased once five or more processors were used. The super-individual approach has potential to improve model speed and memory use dramatically, however this paper cautions the use of this approach for a density-dependent spatially-explicit model, unless individual variability is better taken into account.