Forecasting Climate Change using a Multivariate Cointegrated System

A cointegrated vector equilibrium correction model of key climate variables including surface temperature, ocean heat content, Arctic sea-ice extent and sea-level change is built, driven by radiative forcing in which a stochastic trend arises due to anthropogenic emissions of greenhouse gases. A valid and congruent statistical model requires saturation estimation to model breaks in trends, while also conditioning on natural radiative forcings and El Ni˜no–SouthernOscillation. The model is stable over 150 years, reflecting the slow adjustment of the deep oceans to increased greenhouse gas concentrations, and predicts an equilibrium climate sensitivity of 2.6◦C. Projections out to 2100 highlight the many uncertainties over the coming decades.