NeuralMie (v1.0) supplementary data

Supplementary data for "NeuralMie (v1.0): an aerosol optics emulator." The preprint of the corresponding manuscript can be found here: https://gmd.copernicus.org/preprints/gmd-2024-30/, and the Github repository is located here: https://github.com/pnnl/NEURALMIE. This Zenodo repository contains training data and trained neural networks. Training data can also be generated using the code in the Github repository.

The training data is contained in four files:

inputs.npy - The inputs. Both the 'sphere' and 'coreshell' neural networks use the same input file. The sphere network ignores the last three columns of inputs. Please refer to the Github repository for details on how to load and train on the data.

sphere_targets.npy - The training targets for the neural network that simulates scattering by homogeneous spheres.

coreshell_targets.npy - The training targets for the neural network that simulates scattering by coated spheres.

upper_x.npy - The upper bound of the particle size parameter distributions in the training set (such that 99.5% of the distribution is below this threshold).

Trained Keras neural networks are provided in these files. These are also available from the Github repository along with Fortran-Keras bridge .txt formatted versions of the neural networks:

sphere.h5

coreshell.h5

hyperparameter_search_ models.zip - Contains the randomly generated neural networks that were part of the hyperparameter search process. These should not be used operationally.