This dataset contains the replication data and code for the article "Plasma Density Estimation from Ionograms and Geophysical Parameters with Deep Learning". The project introduces "Kian-Net", a deep learning model designed to estimate electron density profiles in the ionosphere by fusing ionogram images with geophysical parameters.

The Kian-Net model utilizes a fusion architecture (FuDMLP) combining an IonoCNN (Convolutional Neural Network) for processing ionogram images and a GeoDMLP (Deep Multilayer Perceptron) for processing geophysical parameters. These two branches are fused to predict plasma density profiles. The model is trained using plasma density profiles observed by the EISCAT UHF radar as ground truth.

The dataset is organized into modules: 1. Training_KIAN_Net/: Scripts and source data for training the model. 2. Testing_KIAN_Net/: Scripts for evaluating the model on independent test days, including pre-trained weights. 3. Predicting_KIAN_Net/: Scripts for generating predictions on new data. 4. Plotting/: Scripts and data for generating the figures presented in the publication.

Data ranges from 2012 to 2022 and includes magnetometer data and ionograms from the Tromsø Geophysical Observatory, geophysical data from OMNIWeb, and EISCAT UHF radar data.

numpy, 2.2.1

pandas, 2.2.3

scipy, 1.15.2

matplotlib, 3.10.1

pillow, 11.0.0

tqdm, 4.67.1

scikit-learn, 1.6.1

seaborn, 0.13.2

dcor, 0.6

torch, 2.6.0+cu124

torchvision, 0.21.0+cu124