A Statistical Emulator Design for Monthly Averaged Climate Fields

We demonstrate a data-driven method for the emulation of distributions of monthly averaged temperature and relative humidity from Earth system models (ESMs). This emulator model can be used to rapidly infer the response of an ESM to arbitrary climate change scenarios outside of those simulated by the original model. This flexibility could be of particular use in climate impact modeling and policy assessment. While most existing emulation methods focus only on emulating the ensemble mean of climate variables, our method is inherently probabilistic, capturing the ensemble spread directly. We demonstrate that this model is effective at emulating both the means and variances of monthly averaged temperature and relative humidity in previously unseen warming scenarios, including a non-monotonic scenario. We will also discuss applications of our emulator design to other models and climate variables.