Model Assessment - Simulation models are often used to investigate the processes or mechanisms underlying phenomena, especially in biology and ecology. Developing such models involves a considerable degree of uncertainty in the selection of both the model's components and their representation detail. Merely fitting the model to data does not guarantee that the proposed model structure adequately describes the phenomenon (Reynolds and Ford 1999; S. Wood JRSS 1999). Rather, the model must be assessed to reveal which aspects of the phenomenon it can produce and which it cannot (Reynolds and Ford 1999). Technically, this can be viewed as a multi-objective optimization problem with the goal of approximating a Pareto frontier. This research involves methodological development in modelling (implementation procedures; selecting feature and defining discrepancy measures), technical development in optimization (evolutionary computation optimization algorithms, convergence statistics, algorithm comparison statistics), and philosophical development in statistics (clarifying the relation between model assessment aka structural inference and traditional statistical inference aka parameter inference).
Natural Resource Monitoring - Designing long-term monitoring studies for wildlife refuges in Alaska, where target frames can be measured in millions of hectares, access logistics can overwhelm all other considerations, natural variability can be tremendous, and populations can be highly mobile and/or difficult to detect, forces consideration of a number of key aspects of the design process that often remain in the background in more standard applications.
Evolutionary Computation Optimization Algorithms - Algorithms based, broadly, on natural selection are the most effective means of approximating the Pareto frontier in many multi-criteria optimization problems. The field also offers some interesting statistical challenges in terms of developing methods for comparing algorithm performance.
Discrete Mixture Models - Advances in genetics allow biologists and managers to estimate the contribution proportions of different stocks in a mixture of individuals, such as fish captured in a high seas interception fishery. The statistical models underlying these problems are all mixtures of multinomials, a growing field but one that historically received much less attention than mixtures of continuous distributions.
Software - evolutionary computation software, graphic user interface for MS-SURVIV multi-state mark recapture software, R packages for double observer distance estimation, etc.