Finding the effective parameter perturbations in atmospheric models: the LORENZ63 model as case study

Climate models contain numerous parameters for which the numeric values are uncertain. In the context of climate simulation and prediction, a relevant question is what range of climate outcomes is possible given the range of parameter uncertainties. Which parameter perturbation changes the climate in some predefined sense the most? In the context of the Lorenz 63 model, a method is developed that identifies effective parameter perturbations based on short integrations. Use is made of the adjoint equations to assess the sensitivity of a short integration to a parameter perturbation. A key feature is the selection of initial conditions.