Data Assimilation into a LES Model: Retrieval of IFN and CCN Concentrations

We performed a series of experiments assimilating real observations into a Large Eddy Simulation (LES) model to analyze the response of cloud-nucleating aerosol concentrations to a potential optimization model state. A Maximum Likelihood Ensemble Filter algorithm (MLEF developed at CSU) was implemented into the LES version of the CSU Regional Atmospheric Modeling System (RAMS@CSU). MLEF calculates optimal estimates of the atmospheric state, model error (bias) and model empirical parameters, employing a maximum likelihood approach. This algorithm presents an important advantage (compared to the classical ensemble Kalman filter) as it does not make any assumption about the shape in the probability density function (PDF) of the model state (e.g., PDF symmetry). The LES model is interfaced with the Los Alamos sea-ice model and its microphysical modules explicitly consider the nucleation of cloud condensation and ice forming nuclei (CCN and IFN, respectively). A well-documented mixed-phase Arctic boundary layer (BL) cloud case was chosen to perform the assimilation experiments. The model state vector was configured to include the concentrations of IFN and CCN, and the number concentration and mixing ratio of six water species (cloud droplets, drizzle drops, rain drops, pristine ice crystals, snow crystals, and aggregates). Data assimilation into a micro-scale model introduces several unique problems. Among them, allowing each ensemble member a (spin-up) time to develop an eddy distribution (stable turbulence statistics) physically consistent with the new optimal model state after each assimilation cycle, and taking into account aspects not resolvable within the framework of a LES model such as large-scale tendencies.

Simulations cover a period of 54 hours, and observed ice and water paths and/or downwelling radiative fluxes at the surface were periodically assimilated every 2 hours. The algorithm has been tested with these simple observational operators with encouraging results. On the one hand, data assimilation enhanced both timing and vertical structure of the simulated BL clouds. It is important to note that only the vertically-integrated values are assimilated and therefore the vertical structure of ice and water contents is "independent" of the assimilated observations. On the other hand, while numerical simulations were initialized with low aerosol concentrations typical of a pristine Arctic environment, the LES model was successful in reproducing the observed presence of a moderately polluted air mass overriding the inversion. Again, the concentrations of IFN and CCN are not assimilated and therefore they are truly retrieved variables.

Now, we are now implementing more complex observational operators (i.e., satellite radiances). Future efforts will be directed towards finding the optimal configuration of the MLEF-LES coupled model to be used as cloud nucleating aerosol retrieval method.