Calibration and Uncertainty Quantification of Convective Parameters in an Idealized GCM

Parameters in climate models are usually calibrated manually, exploiting only small subsets of the available data. This precludes both optimal calibration and quantification uncertainties. Traditional Bayesian methods that allow uncertainty too expensive for models; they also not robust presence internal variability. For example, Markov chain Monte Carlo (MCMC) typically require model runs sensitive to variability noise, rendering them infeasible models. Here we demonstrate an approach requires can accommodate The consists three stages: (a) a stage uses variants ensemble Kalman inversion calibrate by minimizing mismatches between data statistics; (b) emulation emulates parameter-to-data map with Gaussian processes (GP), using training; (c) sampling approximates posterior distributions GP emulator MCMC. We feasibility computational efficiency this calibrate-emulate-sample (CES) perfect-model setting. Using idealized general circulation model, estimate parameters simple convection scheme from synthetic generated model. CES generates probability good approximations posteriors, at fraction cost required obtain them. Sampling approximate allows generation predictions quantified parametric