AOD Imputation with GAIN: Reconstructing Satellite AOD Gaps Using Deep Generative Models

Table of Contents

• Overview
• Package Installation
• Data Preparation
• Running the Imputation Model
• Validation with AERONET
• License
• Citation

Overview

This repository provides code and workflows for imputing missing Aerosol Optical Depth (AOD) data using a GAIN-based deep learning model. The imputation framework is designed for satellite-derived AOD products with missing observations and validated using AERONET ground truth data.

Package Installation

This project requires Python 3.10 or later. Install all required dependencies using the requirements.yml file:

# Create and activate the Conda environment
conda env create -f requirements.yml
conda activate aq-env

Data Preparation

1. Download the .h5 training dataset from the following link: Download training data

2. After downloading, place the .h5 file into the data/HDF5 directory of the project:

3. Run the following script to prepare the training and test data:

python data_preparation.py

This script will extract relevant features and split the data for training and evaluation.

Running the Imputation Model

To run the GAIN-based AOD imputation model, execute:

python gain_model.py

The model will train on the incomplete AOD dataset and generate reconstructed AOD values.

Validation with AERONET

To evaluate the model performance against AERONET ground-truth measurements, use the validation script:

python aeronet_validation.py

This will compute RMSE, MAE, and correlation statistics comparing imputed AOD with AERONET measurements.

License

This work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/

Citation