Predicting the unexpected: using a qualitative model of a New Zealand dryland ecosystem to anticipate pest management outcomes

Pest management is expensive and there is often uncertainty about the benefits for the resources being protected. There can also be unintended consequences for other parts of the ecosystem, especially in complex food webs. In making decisions managers generally have to rely on qualitative information collected in a piecemeal fashion. A method to assist decision making is a qualitative modelling approach using fuzzy cognitive maps, a directed graphical model related to neural networks that can take account of interactions between pests and conservation assets in complex food webs. Using all available information on relationships between native and exotic resources and consumers, we generated hypotheses about potential consequences of single-species and multi-species pest control on the long-term equilibrium abundances of other biotic components of an ecosystem. We applied the model to a dryland ecosystem in New Zealand because we had good information on its trophic structure, but the information on the strength of species interactions was imprecise. Our model suggested that pest control is unlikely to significantly boost native invertebrates and lizards in this ecosystem, suggesting that other forms of management may be required for these groups. Most of the pest control regimes tested resulted in greater abundances of at least one other pest species, which could potentially lead to other management problems. Some of the predictions were unexpected, such as more birds resulting from possum and mouse control. We also modelled the effects of an increase in invasive rabbits, which led to unexpected declines of stoats, weasels, mice and possums. These unexpected outcomes resulted from complex indirect pathways in the food web. Fuzzy cognitive maps allow rapid construction of prototype models of complex food webs using a wide range of data and expert opinion. Their utility lies in providing direction for future monitoring efforts and generating hypotheses that can be tested with field experiments.