Approximating the Internal Variability of Bias-Corrected Global Temperature Projections with Spatial Stochastic Generators

Decision-making under climate change, from vulnerability assessments to adaptation and mitigation, requires an accurate quantification of the uncertainty in the future climate. Physically constrained projections, in the presence of both observations and climate simulations, can be obtained by establishing an empirical relationship in the historical time period and using it to correct the bias of future simulations. Traditional bias correction approaches do not account for the uncertainty in the climate simulation, and focus on regionally aggregated variables without spatial dependence, with loss of useful information such as the variability of gradients across regions. We propose a new statistical model for bias correction of monthly surface temperatures with sparse and interpretable spatial structure, and we use it to obtain future reanalysis projections with associated uncertainty, using only a small ensemble of global simulations.