Aerosol-only parameterizations for liquid cloud microphysics

The gradient-boosted tree framework LightGBM [51] is used to train separate models for predicting cloud droplet number concentration (CDNC) and median volume diameter (MVD).Each cloud identified in the combined ACTIVATE and Airbus Norway dataset, constitutesa single datapoint. The input variables for each datapoint are the mass mixing ratios of the eleven aerosol species The target variables are the in situ CDNC and MVD measured by the CDP. To test howsensitive the results are to the train–test division, the model was trained one hundred times using different randomsplits, each with 75 percent of the data for training and 25 percent for testing. The clean data for training the decision trees is available cloud_data_MVD.csv.The aerosol data is from CAMS EAC4 reanalysis.The python script lightGBM_train_plot.py has functions for training and evaluating the decision trees for CDNC and MVD.The raw data files from the Norwegian part of the training data are in the Airbus_Norway_2023_2024.zip file. Data from the ACTIVATE campaign can be accessed here: https://www-air.larc.nasa.gov/missions/activate/index.html