Integrating Population Dynamics

RUWPA are at the forefront of developing and applying statistical methods that combine models of wildlife population dynamics and the data available on these populations. This enables us to estimate parameters of interest (such as birth and death rates, population trends, etc) and also perform scenario planning and population viability analysis, all within a rigerious statistical framework.

• State-space Modelling

• Current Research

Above: Graphical representation of a model of the metapopulation dynamics of the British grey seal, click on image to enlarge).

Recent developments in computing and statistics mean that it is now possible to fit quite complex stochastic models of wildlife population dynamics to diverse sources of data, especially data collected over time. This enables us to:

• obtain estimates of population size and trend that are constrained to be biologically realistic
• estimate population parameters such as survival and birth rates, even though they are not directly observed
• use model selection methods to compare the evidence for different biological hypotheses about the processes underlying the population dynamics (such as movement, density dependence, etc.)

• project forward into the future under different scenarios (e.g. different management scenarios such as different harvest levels or remediation measures), giving a coherent method of risk assessment and planning

   

The underlying statistical framework is a state space model (or more generally a hidden Markov model). This models the system using two linked time series: one for the true (but generally unknown) state of the population (such as numbers of animals of different ages or stages in different areas) and the other for observations made on these states (e.g. survey or census data). The true states evolve using a stochastic model of the population dynamics (the "process model"), which can be divided into sub-processes such as over winter survival, movement among populations, breeding, etc. Using simple building blocks, complex models can be built relatively easily. The true, unknown states are linked to the data via a stochastic model of the data collection process (the "observation model"). For more information, see Buckland et al. (2005) and Newman et al. (in press).

Right: Pictorial representation of a state-space model

Specifying complex, biologically realistic models is relatively straightforward, but fitting them is not. Typically, computer-intensive Bayesian methods such as Markov chain Monte Carlo or particle filtering are required.

Left: Posterior parameter estimates (histograms) and priors (solid lines) from British grey seal model. For more information, see Thomas et al. (2005).

Members of RUWPA are at the forefront of developing and applying this framework to addressing real-world problems. Current research areas include:

Right: Pup production data (circles) and filtered estimates (lines) from the British grey seal model. Dashed lines represent 95% credibility interval. For more details, see Thomas et al. (2005).

We are also part of a larger consortium of researchers working on these methods, both within the university (the Centre for Research into Ecological and Environmental Modelling) and in a joint centre with the universities of Kent and Cambridge (the National Centre for Statistical Ecology).

Left: Sperm whale, photo: Howard Hall, Dominica