Jerzy Filar

Risk and Uncertainty Quantification in Environmental Modelling

Mathematical models of environmental problems often demand understanding of complex dynamics and interactions between many physical and biological variables on the one hand, and human inputs on the other. Uncertainties accompanying such models stem from multiple sources. Sometimes they manifest themselves as cascading errors and at other times they involve the risk of key variables crossing undesirable thresholds. In both cases they undermine confidence in either the model or, worse still, the underlying science.

The accompanying mathematical problems can be studied using a wide range of approaches including (but not limited to) perturbation theory, stochastic processes, partially observable Markov decision processes, statistical methods, dynamical systems and simulation. They can also be applied in several important contexts including (but not limited to) conservation of natural resources, optimizing harvests of fish subject to sustainability constraints or generating warning signals for species whose abundance drops to low levels. One particularly challenging problem is that of designing controls that minimize the probability of a catastrophe, consistently over time, while achieving satisfactory and sustainable resource consumption. A related problem, also stemming from fishery science applications, is that of devising a “balanced harvest” strategy that ultimately restores the proportions of age cohorts of the harvested species to those that are natural for that species.

There are several PhD, Masters’ or Honours’ research projects that can be designed on this general theme and tailored to the particular student’s background and interests. For some projects co-supervision with scientists from the Queensland Department of Agriculture and Fisheries, or CSIRO may be required.

Fishery-dependent monitoring of Queensland's fisheries

Review and evaluate efficient sampling programs: Is the right amount of sampling occurring for each species? Are there any significant biases in the sampling programs for each species? Assess whether routine analyses are being carried out correctly and to develop new analyses for fisheries management.

Project components include developing: Quantitative analyses to optimise fishery-dependent sampling across multiple species and regions. Routine methods for assessing precision of current sampling of fish length and age. New methods for turning fish length and age data into advice (indicators) about fishing pressure and the status of fish stocks. A corresponding harvest strategy and reference points for judging the performance of the indicators.

Queensland state-wide estimation of recreational fish catches

Improved estimation of state-wide recreational harvests, including resampling, bootstrap and MCMC techniques. Quantify changes in survey angler avidity and recall bias between survey years and methodologies; adjust previous survey data to obtain improved estimates. Evaluating sampling frames - develop methods to generate state-wide harvest estimates (and associated measures of uncertainty) from several synchronous samples taken from different sampling frames (e.g. a licence frame and a residential telephone number list). Develop hierarchical and conditional mixed models for estimation of recreational fish catch and catch rates. Investigate the statistical modelling of recreational survey data collected from multiple survey methods.

From survey to analysis: dealing with differences in the scale at which survey data are collected and the scale at which data are analysed. Examine appropriate estimation methods for different fish species. Develop statistical methods for low fish abundance or recreational species caught by ‘hard-to-reach’ fishers. Develop methods to engage and retain recreational fishers in volunteer data contribution programs.