The Use of Spread Models to Inform Eradication Programs: Application to Red Imported Fire Ants

A central theme of the invasion biology literature is to predict the introduction and spread of biological invasions but predictive models rarely are applied to inform invasion management. Here, we demonstrate the utility of a spatio-temporal predictive model that has been used to inform Australia's largest eradication program, the program to eradicate the red imported fire ant (RIFA) from Brisbane. That model has informed eradication efforts in two main ways: estimating the probability of program success and identifying cost-effective control strategies. A third application of the model was to inform research priority setting by identifying uncertain parameters with greatest impact on program outcomes. One of the main findings of our modelling analyses was that due to apparent long distance "jump" dispersal, it is critical to search large areas at sufficient sensitivity to detect all individuals. However, the required spatial coverage and sensitivity of surveillance cannot be achieved with the currently used surveillance method, visual search by trained personnel, due to that method's high cost. Therefore, it is necessary to consider alternative surveillance methods. One such method, remote sensing, has a substantially lower cost than visual search but has insufficient sensitivity to find all RIFA colonies. Therefore, there is an important role for both methods. The combined use of those methods potentially can allow for all infestations to be found and thereby achieve eradication even when large areas are infested at low density. We demonstrate that an effective strategy for combining those methods is to use them sequentially, with remote sensing used in the first stage to identify general areas of infestation followed by higher sensitivity visual search to detect all individuals within each infested area. Simulation analyses demonstrate that small increases in the sensitivity of remote sensing are equivalent to large increases in the areal extent of visual search within the plausible range of sensitivities of the two methods. Threshold surveillance sensitivities for remote sensing were estimated, below which stipulated eradication probabilities cannot be achieved with available resources. Although treatment of large areas with aerial baiting can contain the invasion for an extended period without the need for extensive surveillance, some surveillance is required to improve information on infested locations. There is a risk that treated RIFA nests will release reproductive offspring prior to those nests being treated. Our analyses demonstrate that such "reproductive escape", which rarely is considered in invasion spread models, has a large impact on the probability of eradication. Reproductive escape should, therefore, be further studied to improve estimation of eradication feasibility and identification of effective eradication strategies. Our findings demonstrate that even large biological invasions can potentially be eradicated with available resources using readily available surveillance technologies. That will depend on the detectability of individuals or natal sites with those technologies and, more specifically, whether infestations can be detected before they become a source of long distance spread. Predictive models have much to offer eradication programs despite imperfect predictability of invasions.