We develop an analytical real options framework for the optimal timing of climate policy when both economic cost and stocks of environmental pollutants evolve stochastically and are correlated. By modelling environmental damage as a mean-reverting pollution stock jointly driven with uncertain economic losses, we derive a two-dimensional optimal stopping boundary, moving beyond the single-state or simplified threshold rules commonly used in earlier work, including Pindyck (2000, 2002). A key theoretical insight is that the option value of waiting depends non-linearly on ecological conditions, reflected in a strictly positive coefficient on pollution in the continuation value, correcting prior simplifications that implied independence from environmental risk. Our analytical characterisation of the exercise boundary shows it is convex and downward-sloping: higher ecological degradation lowers the economic trigger for action. Most importantly, we find that the correlation between state variables governs how rising volatility affects optimal policy timing. When economic and ecological risks are positively correlated, greater volatility accelerates policy adoption, directly contradicting the classical “uncertainty-delay” effect argued by Pindyck, while negative correlation restores the traditional delay incentive. These results underscore that treating uncertainties jointly rather than separately can fundamentally change policy recommendations, with profound implications for climate risk governance and investment under transition and physical risk.