Development and Evaluation of Climate Simulations Using Machine Learning Enhanced Aerosol Model in OpenIFS Atmospheric Model

Atmospheric aerosols have a significant impact on cloud formation and life cycle. Aerosols enhance cloud formation and affect microphysical and radiative properties of clouds by acting as Cloud Condensation Nuclei (CCN). Aerosol-cloud interactions are very complex, and thus accurate global-scale simulations are challenging. 

Aerosol-cloud interactions occur at a microscopic level, but cloud systems are often on a scale of tens or hundreds of kilometers. Accurate modeling of all aerosol-cloud processes at such a large scale is computationally demanding. Therefore, models simulating aerosols and their interactions with radiation and clouds, are usually greatly simplified, making them inaccurate. In this study, the accuracy of a simple aerosol model HAM-Lite will be enhanced with a machine-learning component, and the enhanced model will be coupled with a global kilometer-scale Numerical Weather Prediction (NWP) model OpenIFS. 

OpenIFS is used for global climate simulations and weather forecasting. It is an easy-to-use version of Integrated Forecasting System (IFS) by the European Centre for Medium-Range Weather Forecasts (ECMWF). IFS models the atmosphere in EC-Earth 3 climate model and it is developed by the European Consortium of National Meteorological Services and Research Institutes. OpenIFS will be the main atmospheric model in the upcoming EC-Earth version 4. 

HAM-Lite is a simplified version of a more complex aerosol model HAM-M7. While HAM-M7 includes seven log-normal aerosol modes and a total of twenty-five tracers, HAM-Lite describes only four tracers. HAM-M7 calculates microphysical processes, like nucleation, condensation and coagulation, as well as other processes like emissions and dry and wet deposition. HAM-Lite simplifies the processes by assuming constant hygroscopicity and very simplified calculations for extinction. These simplifications make the model computationally lighter. 

Since aerosol hygroscopicity and extinction are highly simplified in HAM-Lite, we will incorporate machine learning methods to provide a more accurate representation, bringing its performance closer to that of HAM-M7. Training data for the machine learning component will be produced with HAM-M7 coupled with OpenIFS. The new enhanced HAM-Lite aerosol model will also be coupled with OpenIFS for improved global scale simulations. 

By coupling the new enhanced aerosol model with the climate model, the aim is to make the system more accurate without significantly increasing the computational cost. Results from OpenIFS, with and without the enhanced aerosol model, will be compared to in situ measurements, satellite data, and simulations with other models. Expectation is that OpenIFS, coupled with the light aerosol module and machine learning methods, will achieve higher accuracy with reduced computational cost compared to OpenIFS coupled with HAM-M7. 

This research is funded by the European Union's Horizon EU -project Digital Twin of Earth System for Cryosphere, Land Surface, and Related Interactions – TerraDT 101187992.