readme.md

MieAI

A neural network for calculating optical properties of internally mixed aerosols in atmospheric models (https://arxiv.org/abs/2312.06497).

Here is the companion fortran repository: MieAI-Fortran

Paper

Look at the paper at this link for the details about MieAI.

Usage

Here is an example usage of MieAI for calculating aerosol optical depth (AOD) of mixed mode aerosols using ICON-ART simulation:

import xarray as xr
from aop import MieAI

core = ['dust', 'soot', 'na', 'cl']
shell = ['h2o', 'so4', 'nh4', 'no3']
wavelength = 0.55 # [in micrometer]
mode = 'acc' # for accumulation mode

dx = xr.open_dataset('icon-art-aging-aero_DOM01_ML_0012.nc')
mie = MieAI(dx, wavelength, core=core, shell=shell, mode=mode)
aod = mie.get_aod()

To calculate the bulk optical properties like extinction coefficient ($k_e$), scattering coefficient ($k_e$), single scattering albedo ($\omega$) and asymmetry parameter (g) at all vertical levels:

aop = mie.get_aop(dx)

To calculate optical properties at a particular height:

dy = dx.isel(height_2=50)
aop = mie.emulate(dy)

Paper codes

Analysis

1. Mie calculation for core-shell configuration
2. Compilation of Mie results in a single file
3. MLP training without quantile transform: x1
4. MLP training without quantile transform: x2
5. Quantile Transformation: Figure3 in paper
6. Hyperparamter optimisation of MLP architecture: Part1
7. Hyperparamter optimisation of MLP architecture: Part2
8. Hyperparamter optimisation Table
9. MieAI Training
10. MieAI performance: Figure4 in paper
11. AOD calculation using MieAI

utility codes

1. Codes translated from MATLAB2python
2. Functions for preprocessing, mie run and MLP
3. AOD calculation using ICON-ART data

Case studies

1. Code for Biomass Burning Event
2. Code for Volcanic Eruption Event
3. Code for Dust Event
4. Code for Volcanic Eruption Event using the MLP without quantile transform

References

1. Papers
2. Packages