New mathematical method allows scientists to better predict the effects of global warming

Scientists may soon have a better idea of the potential effects of global warming thanks to the work of Tony O’Hagan, Professor of Statistics at the University of Sheffield.

Scientists already know that global warming will have consequences for the earth but are unsure how mild or severe these effects may be, due to the inaccuracies in their prediction models. Professor O’Hagan’s method will allow scientists to better determine these inaccuracies and has major significance in the world of environmental science.

He will be presenting his work today at the British Association’s Festival of Science, University of Leicester. He is currently the Chair of BA’s mathematics section, a post he will be handing over to his successor at today’s meeting.

The basis of his work is to speed up the process of predicting uncertainties in complex mathematical models and to better determine the margin for error in predictions. Traditionally scientists have had to run a complex model many times, randomly altering the input to determine how this will affect the output. For example, they may randomly alter levels of atmospheric C02 to determine how this affects climate change. This series of experiments gives scientists a picture as to how accurate their predictions are. However, this approach is highly impractical when one considers that a single run of the model may take hours, or even days and that the model may need thousands of runs to be useful.

Professor O’Hagan’s method allows scientists to alter the input strategically, doing systematic runs of the model, which give a general picture of the effects of change. This allows them to build a statistical approximation of the model (called an emulator) that can quickly predict the outcome on any given input. The emulator can be created using only a few runs of the model and can calculate outcomes almost instantaneously, allowing scientists to better understand the margin for error in their predictions.

Professor O’Hagan says, “This work is extremely important in a number of fields and makes analysing uncertainties in large and complex models easier and much more accurate. It can be used in a number of diverse situations such as climate change, predicting nuclear contamination after an accident or predicting the yield of wells in an oil field.”