| Summary: | This study carries out a comprehensive ensemble experiment investigating previously unexplored combinations of model uncertainty in an attempt to quantify and attribute the response uncertainty of future Greenland ice sheet twenty first century simulation. The inclusion of multiple uncertainty sources facilitates the construction of the first ever probability density function (PDF) of Greenland ice sheet surface mass balance (5MB) behaviour over the 21st century. The use of an {insolation temperature' 5MB model permits the inclusion of important ice sheet feedbacks not accounted for in more parameterized models traditionally used in such uncertainty analysis. The lower sensitivity of 'this model (compared to temperature only parameterised models) results in the zi" century 5MB being in the lower end of previously reported ranges. The experiment includes a number of novel methods for downscaling climate data and incorporating new uncertainties such as climate model internal variability. This study also presents a new Bayesian inference method based on summary statistics of present day ice sheet behaviour. The inclusion of internal variability is shown to be crucially important in the Bayesian inference method so that realisations are not highly weighted due to coincidence of random climate fluctuations. Further to this, the long term climate signal is shown to produce too little warming to explain all of the recent runoff trends over the ice sheet, with internal variability accounting for the remainder. Caution must therefore be exercised when extrapolating present ice sheet trends into the future. The application of sensitivity analysis techniques facilitates the identification of important regions that dominate uncertainty, which help to constrain parameter ranges in future sampling within similar models. Climate model uncertainty is shown to dominate zi'' century uncertainty, with the models showing the largest zi" century warming producing the best reconstruction of present day runoff.
|
|---|
