Quantifying parameter uncertainty in projections of glacier mass change in Iceland under GlacierMIP3 experiments

 

Recent advancements in the numerical modelling of glaciers has enabled projections of glacier mass change at regional and global scales. However, this progress has been facilitated by the use of simple mass balance models that rely heavily on parameterisations, often with poorly constrained parameters. These models are typically calibrated by varying parameters in order to minimise the difference between modelled and observed mass balance. Though intuitive, this procedure risks an over-tuning of model parameters, ultimately resulting in an underestimation of uncertainty in projections of glacier change. In this study, we present a novel application of the calibration technique known as ‘history matching’. Rather than tuning parameters to obtain a single ‘best’ solution, this method is used to obtain an ensemble of plausible parameter sets; ultimately enabling an assessment of parameter uncertainty in projections. Here, we apply this approach to quantify parameter uncertainty in the GO model contribution to GlacierMIP3. We run the model for each experiment in GlacierMIP3 using a 250-member perturbed-parameter ensemble, generated using a Latin hypercube design. We then constrain this ensemble through history matching by filtering ensemble members that are inconsistent with geodetic mass balance observations outside defined limits of plausibility. The ensemble is used to assess the sensitivity of projections to total parameter uncertainty as well as individual model parameters. We find a high degree of equifinality in the ensemble, in which ensemble members that performed equally well in calibration produce contrasting responses to the same climate forcing.