Bayesian Calibration of a Large‐Eddy Resolving Model Towards Campaign Measurements With an Ensemble Kalman Smoother

Abstract

Abstract Large‐eddy simulations (LES) are commonly employed to study small‐scale weather phenomena and to validate coarser models. Like larger‐scale weather and climate models, they contain weakly constrained parameters and use physical inputs with associated uncertainties, such as initial conditions or forcing. This study assesses the viability of the ensemble Kalman smoother (EnKS) to tune both parameters and physical inputs of an LES simulation towards campaign measurements. The EnKS relies on a perturbed parameter ensemble, as is common in sensitivity analysis. By returning an updated ensemble over the tuning parameters, it provides optimal values together with uncertainty estimates. In a Bayesian sense, it balances parametric and measurement uncertainties objectively. The methodological simplicity enables an in‐depth interpretation of both model sensitivities and the calibration results. The methodology is demonstrated by calibrating the PyCLES model towards measurements of nocturnal marine stratocumulus clouds (DYCOMS‐II RF01) at a horizontal resolution of 35 m. Using a WENO finite volume scheme for advection, the model with calibrated inputs improves over models compared in Stevens et al. (2005, https://doi.org/10.1175/mwr2930.1 ), indicating that errors in the prescribed initial condition and forcing may have contributed to biases shared by most tested LES models. The repetition of the calibration with two numerical advection schemes highlights that parametric calibration can only reach a satisfying performance of the improved simulation for models that are not dominated by structural errors. The presented methodology provides a transparent and interpretable framework for model calibration using observations with quantified uncertainties. Plain Language Summary Weather and climate models rely on many parameters that are difficult to determine precisely. Incorrect parameters may result in systematic biases, especially in model components that simulate the influence of unresolved processes. This study tests the ensemble Kalman smoother (EnKS) to calibrate both initial conditions and model parameters using measurement data from the DYCOMS‐II field campaign, which observed low‐level clouds off the coast of Southern California. The calibrated model overcomes some but not all biases identified in previous intercomparison studies. Key Points An ensemble of model simulations is combined with measurements to obtain improved model initial conditions, parameters, and forcing The single Bayesian update step of the ensemble Kalman smoother allows for in‐depth interpretation of the calibration The performance of the PyCLES high‐resolution model on the DYCOMS‐II RF01 case of marine stratocumulus is improved through the calibration