IonoBench: Evaluating Spatiotemporal Models for Ionospheric Forecasting under Solar-Balanced and Storm-Aware Conditions
Published in Remote Sensing (MDPI)

<sub>Storm (Dst −292 nT): 12-hour-ahead forecasts (top) and residuals (bottom), on unseen test samples.

Quiet (Dst +6 nT): same models and layout. Note the residual scale: ±15 TECU here vs. ±45 TECU during the storm.</sub>

---

This project is a benchmark framework for evaluating deep spatiotemporal models on Global Ionospheric Map (GIM) forecasting. The framework provides standardized datasets, evaluation protocols, pretrained models, and configuration-based experimentation.

Overall Performance on IonoBench Test Set

Model	RMSE (↓)	R² (↑)	SSIM (↑)
SimVPv2	2.25 ± 1.35	0.962 ± 0.015	0.969 ± 0.020
DCNN121	2.62 ± 1.66	0.950 ± 0.023	0.963 ± 0.025
SwinLSTM	2.66 ± 1.49	0.946 ± 0.020	0.960 ± 0.023
IRI 2020	6.39 ± 4.53	0.720 ± 0.109	0.852 ± 0.043

Click Open in Colab to test without local setup.

---

Features

• Supports multichannel spatiotemporal models for multistep 24-hour input to 24-hour output setup
• Includes stratified (blocked subsets balancing solar and geomagnetic representation for the models) and chronological datasets
• Model registry and configuration system (for contributors adding new models)
• Reproducible and pretrained models are available via Hugging Face (see tutorial/)
• Experiments for solar enforcing (increasing solar flux over different portions of the solar cycle) and storm behaviour for intense and superintense storms

---

Local Setup

# Clone repository
git clone https://github.com/Mert-chan/IonoBench.git

# Change your directory
cd IonoBench

# Create environment
conda create -n ionobench python=3.11 -y
conda activate ionobench

# Install dependencies
pip install -r requirements.txt

Tested on: Python 3.11.13 · PyTorch 2.5.1 · CUDA 12.4
The environment uses torch==2.5.1, which requires a compatible CUDA build.

---

Command Line Interface (CLI)

Run experiments without notebooks:

# Download dataset
python -m scripts.cli data --type stratified

# Download model
python -m scripts.cli model --name SimVPv2

# Test on test set
python -m scripts.cli test --model SimVPv2 --checkpoint training_sessions/SimVPv2/...pth --session-name test_run

# Solar intensity analysis
python -m scripts.cli solar --checkpoint training_sessions/SimVPv2/...pth --session-name solar_run --save-raw

# Storm event analysis
python -m scripts.cli storm --checkpoint training_sessions/SimVPv2/...pth --session-name storm_run --save-raw

For details: python -m scripts.cli --help

---

Citation

@article{Ionobench2025,
  title   = {IonoBench: Evaluating Spatiotemporal Models for Ionospheric Forecasting under Solar-Balanced and Storm-Aware Conditions},
  author  = {Turkmen, M.C. and Lee, Y.H. and Tan, E.L.},
  journal = {Remote Sensing},
  year    = {2025},
  volume  = {17},
  number  = {15},
  pages   = {2557},
  doi     = {10.3390/rs17152557}
}