Climate science and policy making are currently dominated by model-led forecasting as a means of informing decision-making. However, given the very significant uncertainties surrounding our understanding of both the climate and socio-economic systems and their interactions, it appears more reasonable to view climate decision-making as a recursive problem led by updates based on the unfolding observed state of these systems. Not surprisingly, many aspects of the current climate decision making machinery already possess this attribute, embedded as it is in the review cycles that proliferate in this and other areas of decision-making under uncertainty. In this paper, we will illustrate the recursive nature of climate decision-making using a geoengineering case study. Because of the deep uncertainties surrounding these speculative technologies, should geoengineering ever be deployed, it seems most likely it would be deployed sequentially within a review cycle where the magnitude of deployment is conditioned on a combination of environmental observations, model forecasts, risk assessment and cost. Such frameworks contain the essential elements of a Model Predictive Control (MPC) problem. Here, we apply MPC to explore a stratospheric aerosol campaign. The experiment uses the UK Met Office Hadley Centre Global Environment Model (HadGEM2) as a surrogate for the Earth in a blind trial simulation where the objective is to define the magnitude and temporal distribution of SO2 emissions injected into the stratosphere from a northern hemisphere location equivalent to Svalbard in order to recover and then stabilize the minimum extent of the Arctic ice sheet over a period of 80 years. The control algorithm must contend with HadGEM2's considerable internal variability and non-stationary dynamics, mismatch between the control model and HadGEM2, uncertainty in future greenhouse gas forcing and a stochastic volcanic aerosol time series. We present our results and reflect on the experiment.