Toward Efficient Calibration of Higher‐Resolution Earth System Models

Projections of future climate change to support decision‐making require Earth system models (ESMs) running at high spatial resolution, but at present this is computationally prohibitive. A major challenge is the calibration (parameter tuning) during the development of ESMs, which requires running large numbers of simulations to identify optimal values for parameters that are poorly constrained by observations. Here, we train a convolutional neural network (CNN) to emulate perturbed parameter ensembles from two lower‐resolution (and thus much less expensive) versions of the same ESM, and a smaller number of higher‐resolution simulations. Cross‐validated results show that the CNN's skill exceeds that of a climatological baseline for most variables with as few as 5–10 examples of the higher‐resolution ESM, and for all variables (including precipitation) with at least 20 examples. This proof‐of‐concept study demonstrates a machine learning based approach that makes the process of constructing a higher‐resolution emulator 20%–40% more computationally efficient, and thus offers the prospect of significantly more efficient calibration of ESMs. Plain Language Summary To determine how Earth's future climate will respond to greenhouse gas emissions requires building accurate computer models. Building these models requires a time‐consuming calibration process to find optimal values for uncertain constants (parameters) in the model equations that represent small‐scale processes. We took a machine learning method (called CNN) that is commonly used in image recognition applications and inverted it to replicate the calibration process of the climate model. The CNN reproduces all of the main features of the global simulation of the climate model in a fraction of the computational time, including for precipitation which varies a lot from place‐to‐place. The CNN also makes efficient use of information contained in outputs from simpler versions of the climate model, which are available at much lower cost. Our results suggest that inserting an artificial intelligence method, like CNN, in the calibration process for a climate model can significantly reduce the computational time required. Key Points Calibration of poorly constrained parameters in higher‐resolution Earth system models (ESMs) is computationally expensive A machine learning technique from computer vision can replace the ESM during calibration, even for complex variables like precipitation Our machine learning app