This project focuses on the development of ensemble approaches, linking various models and databases to assist in decision support in bioprotection and biosecurity.

Closely related to Machine Learning and Bioclimatic Mapping and Prediction, this project is based on the well-known fact that ecological systems represented by predictive models are highly complex and often stochastic by nature. Subsequently our understanding of them is always incomplete.

No model can capture all the complexity of the system that it represents and therefore prediction error or uncertainty should be expected.

This project aims to develop predictive models for risk assessment and pest management and bioprotection that are more accurate as well as able to represent uncertainty (forecast spread).

Our philosophy arises from the knowledge that unless there is extremely detailed and accurate knowledge of the relationship between a species and its environment, no one model, or model run that attempts to predict that species dynamics can represent reality, and that multiple models or multiple runs, with some sort of aggregation of predictions, are required.

An ensemble of predictions can be used to generate a probability distribution function or a frequency distribution from which confidence limits of some future state of the system can be derived.

Research, particularly in meteorology and climate change forecasting, has shown ensemble models to improve predictions.

Ensemble predictions can be generated by changing:

1) initial conditions
2) model parameters
3) model types or structures
4) input data by intensive bootstrapping or stochastic processes
6) combinations of the foregoing.

Back to Intelligent Systems for Biosecurity homepage