Optimal allocation of limited resources to biosecurity surveillance using a portfolio theory methodology

Abstract

Newly arrived pests and diseases pose a continuous threat to agriculture and the environment, and there are limited resources available for early detection programs. We formulate an optimal resource-allocation framework for early detection. The approach is based on the principles of portfolio theory although, unlike standard portfolio theory and its previous applications to biosecurity problems, return is modelled using a flexible mixed-distribution and an uncertainty measure is defined using the Chernoff bound. The Chernoff bound does not assume underlying symmetry and allows sensitivity to both stochasticity and parameter uncertainty to be incorporated systematically into the optimisation process. Optimisation is a trade-off between expected return values and uncertainty while, at the same time, hedging against low-probability high-impact events. In an application to the islands of Torres Strait (Australia), we demonstrate how uncertainty, stochasticity and linkages between surveillance locations interact to significantly alter relative allocation, when compared with allocation based on expected values alone. Locations of particular interest for biosecurity are identified and results lead to a more even spread of resources, or greater surveillance ‘diversification’.