Turbulence Dynamics Unlock New Understanding of Ionospheric Radar Clutter

Comparison of the simulated and measured RD spectra for ionospheric inhomogeneities using a perceptual similarity metric. 
(a) AlexNet-based LPIPS results of dataset A. (b) AlexNet-based LPIPS results of dataset B. 
(c) SqueezeNet-based LPIPS results of dataset A. (d) SqueezeNet-based LPIPS results of dataset B. 
(e) VGG-based LPIPS results of dataset A. (f) VGG-based LPIPS results of dataset

Understanding the Validation Results: How Well Does the Simulation Match Reality?

This figure shows the results of comparing computer-simulated ionospheric radar clutter against real-world radar measurements. Think of it as a "report card" for how accurately the turbulence model recreates what actual radar systems observe.

What Are We Looking At?

Each panel (a through f) contains multiple box plots—those blue boxes with whiskers extending above and below. Each box plot represents one simulated radar spectrum compared against all 1,290 real radar measurements collected from actual high-frequency surface wave radar systems.

The figure is organized into three rows, each using a different validation method:

Top row (a, b): AlexNet-based LPIPS metric
Middle row (c, d): SqueezeNet-based LPIPS metric
Bottom row (e, f): VGG-based LPIPS metric

The left column (a, c, e) shows "Dataset A" results, while the right column (b, d, f) shows "Dataset B" results—these represent simulations run under two different initial ionospheric conditions.

How to Read the Box Plots

Each box plot tells you:

• Blue box: The middle 50% of similarity scores when comparing one simulation to all measurements
• Red horizontal line in the box: The median (middle value) similarity score
• Whiskers (vertical lines): The range of typical scores
• Red plus signs: Outlier measurements that are unusually different
• Red circles above boxes: The average similarity score for that spectrum

Higher values = better match between simulation and reality. Values range from about 0.35 to 0.70, where 1.0 would be a perfect match.

Key Findings

1. Consistent Performance Across Methods

All three deep learning methods (AlexNet, SqueezeNet, and VGG) show similar patterns, which indicates the results are robust and not dependent on one particular validation technique. The median values cluster between 0.50 and 0.65 across most spectra.

2. VGG Shows the Strongest Performance

The bottom row (panels e and f) using VGG consistently shows the highest similarity scores, with many median values exceeding 0.60 and some approaching 0.70. This suggests that deeper neural networks better capture the complex texture patterns in ionospheric radar clutter.

3. Low Variability = Stable Simulations

The relatively compact blue boxes (small interquartile ranges) indicate that each simulation consistently matches most of the real-world data—not just a few cherry-picked examples. This consistency demonstrates the model's reliability across diverse ionospheric conditions.

4. Subtle Improvement Trend

Looking across the x-axis (RD Spectrum No.), there's a gentle trend toward slightly higher scores in later spectra. This suggests the simulations progressively capture certain evolving characteristics of ionospheric behavior.

What This Means in Practice

The high LPIPS scores (0.55-0.70 range) indicate that the turbulence-based simulation successfully recreates the visual and structural patterns that radar operators actually observe. This validation is crucial because:

• For radar engineers: The model can predict what types of interference patterns to expect under different conditions
• For system designers: Accurate simulations enable testing of clutter suppression algorithms before expensive field deployments
• For scientists: The strong match confirms that turbulence dynamics correctly explain the physical mechanisms creating ionospheric irregularities

The fact that simulations match real data so well across 1,290 different measurements—representing various times of day, seasons, and space weather conditions—demonstrates this isn't just curve-fitting to limited data. The turbulence model captures fundamental physics governing how the ionosphere behaves.

In essence, this figure provides visual proof that treating the ionosphere as a turbulent fluid system (like Earth's atmosphere or ocean currents) successfully explains the complex interference patterns that plague high-frequency radar systems.

Based on the research paper, there is no indication that the model, source code, or computational artifacts have been made publicly available for independent verification.

What the Paper Does NOT Mention:

• No GitHub repository or code sharing platform links
• No data repository citations (e.g., Zenodo, Figshare, IEEE DataPort)
• No supplementary materials section with simulation code
• No statements about data/code availability policies
• No contact information for requesting research materials beyond author emails

What IS Available:

The paper provides:

• Detailed mathematical formulations (Navier-Stokes equations, RANS, k-ε turbulence model)
• Specific parameter values (drift velocities, Reynolds numbers, domain dimensions)
• Validation methodology descriptions
• 1,290 measured RD spectra were used, but no mention of public access

Reproducibility Concerns:

Limited reproducibility without:

1. Simulation code: The specific implementation of RANS/k-ε equations with ionospheric boundary conditions
2. Measured radar data: The 1,290 validation spectra from HFSWR campaigns
3. Pre-trained models: The LPIPS validation networks and their configuration
4. Preprocessing pipelines: How raw radar data was converted to RD spectra for comparison
5. Initial conditions: Exact parameters for "Dataset A" and "Dataset B" simulations

Standard Practice vs. This Study:

Modern computational research increasingly requires:

• IEEE policy: Encourages but does not mandate code/data sharing
• Reproducibility standards: Leading journals now expect public repositories
• Community norms: Computational fluid dynamics studies often share meshes, solver settings, and validation datasets

This study appears to follow traditional publication practices where detailed methods are described but implementation artifacts remain with the research group.

Options for Verification:

If you need to reproduce or verify this work:

1. Contact authors directly:

   • Yuanbiao Li: ybuleo@foxmail.com
   • Lei Yu: yu.lei@hit.edu.cn
   • Yinsheng Wei: hitweiysgroup@163.com

2. Request specific materials:

   • Simulation source code
   • Configuration files for COMSOL/OpenFOAM/custom solver
   • Sample RD spectra from validation dataset
   • LPIPS model architectures and weights

3. Partial replication:

   • The mathematical framework is fully specified
   • Commercial CFD software (COMSOL Multiphysics, ANSYS Fluent) can implement RANS/k-ε models
   • LPIPS is open-source: https://github.com/richzhang/PerceptualSimilarity
   • Could replicate methodology with different radar datasets

Recommendation:

For independent verification, direct communication with the Harbin Institute of Technology research group is necessary. The Chinese research institution may have different data sharing policies than Western universities, and military/dual-use radar applications may limit data release even for civilian research.

The lack of publicly available artifacts is a significant limitation for a 2025 publication, especially one making novel methodological claims. Ideally, the validation dataset (even if anonymized/subset) and core simulation code should be shared to enable the scientific community to build upon this work. Perhaps the IEEE should make this a condition of publication.

 

Novel simulation model based on fluid dynamics principles accurately recreates ionospheric irregularities that interfere with high-frequency radar systems

Researchers at Harbin Institute of Technology have developed a groundbreaking approach to modeling ionospheric clutter that combines classical fluid dynamics with deep learning validation, offering new insights into how turbulent plasma structures in Earth's upper atmosphere interfere with high-frequency surface wave radar (HFSWR) systems.

The study, published in IEEE Transactions on Geoscience and Remote Sensing, represents a significant departure from previous ionospheric clutter models that relied primarily on statistical characterizations of radar interference patterns. Instead, the research team led by Yuanbiao Li, Lei Yu, and Yinsheng Wei applied the principles of turbulence dynamics to explain the microphysical mechanisms underlying small-scale ionospheric irregularities.

From Statistical Models to Physical Mechanisms

The ionosphere—Earth's electrically charged upper atmosphere extending from approximately 60 to 1,000 kilometers altitude—serves dual roles in radar operations. It enables long-range communication for skywave radar systems while simultaneously generating problematic clutter for surface wave radar. Understanding and predicting this clutter has been a persistent challenge for radar engineers.

"Even during quiet conditions, the ionosphere exhibits dynamic, small-scale irregularities that interact with radio waves, causing rapid signal fluctuations, discreteness, and spectral broadening," the researchers explain in their paper. These irregularities manifest as five distinct clutter types in radar data: lamellar distribution, dot distribution, spatial correlation, distance-dependent, and target-like patterns.

Previous modeling approaches treated ionospheric clutter primarily as a statistical phenomenon, focusing on amplitude and phase characteristics without fully explaining the underlying physical processes. The new model takes a fundamentally different approach by treating the ionosphere as a turbulent fluid system governed by the Navier-Stokes equations—the same mathematical framework used to describe atmospheric weather patterns and ocean currents.

Turbulence at the Edge of Space

The research team employed Reynolds-averaged Navier-Stokes (RANS) equations combined with the k-ε turbulence model to simulate ionospheric inhomogeneities under various flow conditions. By varying key parameters—particularly drift velocity (10-50 m/s) and Reynolds number (400-4,000)—the simulations revealed how ionospheric irregularities form and evolve through distinct stages.

The Reynolds number, a dimensionless quantity that characterizes the ratio of inertial forces to viscous forces in a fluid, proved critical to understanding ionospheric behavior. At low Reynolds numbers (around 400) with drift velocities of 10 m/s, the ionosphere maintains relatively laminar flow patterns. As the Reynolds number increases to 4,000 and drift velocities reach 50 m/s, the flow transitions through vortex formation to fully developed turbulence.

"When the Reynolds number reaches 4,000 and the free drift velocity of ionospheric inhomogeneities reaches 50 m/s, the initially homogeneous sheet flow tends to form multiple smaller scale vortices," the researchers observed. This transition manifests in radar data as dot distribution fine structures embedded within broader lamellar distributions—a fractal-like pattern consistent with the self-similar nature of turbulence.

The simulation covered a horizontal range of 5 kilometers and a vertical range of 10 kilometers, representing the E- and F-layers of the ionosphere where most radar-relevant irregularities occur. These layers, located between 90-140 km and extending to 1,000 km respectively, contain molecular and atomic ions that respond to electromagnetic fields and neutral atmospheric winds.

Validation Through Deep Learning

To validate their turbulence-based simulations, the research team took the innovative step of applying deep learning metrics traditionally used in computer vision to compare simulated and measured radar data. They analyzed 1,290 range-Doppler spectra collected from actual HFSWR measurements, each representing 200 range bins and 512 Doppler frequency bins.

Traditional image quality metrics like peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) showed moderate agreement between simulated and measured data. However, these metrics demonstrated limitations in capturing the perceptual similarities crucial for validating complex signal patterns.

The researchers achieved more compelling validation using the Learned Perceptual Image Patch Similarity (LPIPS) metric, which employs deep neural networks to assess perceptual similarity in ways that align more closely with human judgment. Testing LPIPS with three different neural network architectures—AlexNet, SqueezeNet, and VGG—the team found consistently high similarity scores, with median values exceeding 0.6 and peak scores approaching 0.7.

"VGG-based LPIPS boxplots show the strongest performance among all metrics," the paper notes. "Median values consistently exceed 0.6, with several spectra near or above 0.7." The reduced variability in VGG-LPIPS scores indicated high consistency in simulation quality across different ionospheric conditions.

Five Flavors of Ionospheric Interference

The validated simulations revealed distinct clutter types corresponding to different stages of turbulence development. Target-like clutter appears during laminar flow conditions when Reynolds numbers are low and drift velocities slow. This pattern exhibits properties similar to actual radar targets—including directionality and concentration in spatial and range domains—but with dispersed power that can trigger false alarms in detection algorithms.

Lamellar distribution clutter, characterized by high concentrations of ionospheric ions in certain layers, manifests as strong energy aggregation occupying multiple range-Doppler resolution units. This pattern can persist as a quasi-stable state during turbulent conditions.

As flow transitions to the vortex stage, distance-dependent clutter emerges in regions of slowly varying ion concentrations between ionospheric layers, typically occupying multiple resolution units with Doppler frequencies close to zero. Spatial correlation clutter forms in regions of rapidly changing concentrations, producing dispersed power with larger Doppler frequencies and characteristic spatial distributions.

Finally, dot distribution clutter appears as small-scale structures embedded within lamellar patterns during fully developed turbulence. These represent the internal self-similar structure of larger clutter patterns—supporting the application of fractal theory for clutter suppression strategies.

Sensitivity Analysis and Model Limitations

The research team conducted extensive sensitivity analysis to understand how parameter uncertainties affect model predictions. Testing perturbations of 5% and 10% to drift velocity and Reynolds number, they found distinct error patterns for laminar versus turbulent regimes.

For laminar flow conditions, 5% changes in drift velocity produced localized errors confined to shear boundaries, while turbulent flow exhibited broader error propagation aligned with vertical structures. Horizontal errors dominated under drift velocity perturbations—reflecting advection-driven dynamics—whereas vertical errors intensified with Reynolds number changes, indicating stratification-modulated turbulence.

"This anisotropy underscores the importance of multiperspective radar observations to disentangle parameter-specific effects in real-world clutter data," the researchers emphasize.

The study acknowledges several limitations. The RANS model showed reduced accuracy at very high Reynolds numbers, where direct numerical simulation (DNS) or large eddy simulation (LES) would better resolve subrange dynamics. Deep learning validation metrics, while demonstrating strong perceptual alignment, exhibited sensitivity to image scaling. The team also notes fundamental challenges in solving three-dimensional Navier-Stokes equations that constrain model optimization.

Implications for Radar Design and Space Physics

The turbulence-based approach offers practical benefits for HFSWR system design and operation. By accurately simulating different clutter types under varying ionospheric conditions, the model provides a predictive tool for developing adaptive signal processing strategies. The correlation between clutter characteristics and ionospheric conditions suggests promising avenues for real-time clutter mitigation.

The research also contributes to broader understanding of ionospheric physics. The validation of small-scale turbulent mechanisms complements existing models of large-scale field-aligned irregularities based on geomagnetic field models like the International Geomagnetic Reference Field (IGRF). While IGRF-based approaches successfully characterize macroscale stability that enables coherent skywave propagation, the turbulence model explains the chaotic microscale dynamics that generate diffuse clutter patterns.

"The chaotic dynamics of these substructures, unresolved in IGRF-based analyses, directly explain the diffuse clutter patterns observed in HFSWR systems," the researchers note.

This complementary relationship between large-scale stability and small-scale chaos proves essential for optimizing both skywave and surface-wave radar designs. Understanding when and where turbulent conditions develop allows radar operators to anticipate interference patterns and adjust system parameters accordingly.

Future Directions

The research team plans to extend their work in several directions. Future studies will implement more sophisticated simulation techniques including DNS and LES to better capture high Reynolds number turbulence. The model's scope will expand to incorporate additional ionospheric variables such as magnetic field variations, solar activity indicators, and seasonal patterns.

The team also intends to investigate the model's applicability across diverse frequency ranges and radar configurations. Current validations focused on typical HFSWR operating frequencies, but ionospheric effects vary significantly across the electromagnetic spectrum.

Integration of real-time ionospheric observations could enable predictive capabilities, allowing radar systems to anticipate clutter conditions based on current space weather. Such integration would require automated data assimilation pipelines connecting ionosonde measurements, satellite observations, and ground-based radar to the turbulence simulation framework.

"Continued refinement of simulation models through integration of diverse observational data is expected to significantly improve HFSWR system reliability and operational effectiveness," the researchers conclude.

The work demonstrates how classical physics principles, modern computational methods, and machine learning validation can combine to solve practical engineering challenges while advancing fundamental scientific understanding. As space weather monitoring capabilities improve and computational resources expand, turbulence-based ionospheric modeling may become standard practice for radar system design and optimization.

---

Sources

1. Li, Y., Yu, L., & Wei, Y. (2025). 
   Radar Echo Modeling and Mechanism Analysis of Ionospheric Clutter Based on Turbulence Dynamics | IEEE Journals & Magazine | IEEE Xplore. IEEE Transactions on Geoscience and Remote Sensing, 63, 4110613. https://doi.org/10.1109/TGRS.2025.3618480

2. Ho, A. Y., Kuo, S. P., & Lee, M. C. (1994). Analysis of electromagnetic wave scattering by ionospheric irregularities. Radio Science, 29(5), 1179-1186. https://doi.org/10.1029/94RS01103

3. Retterer, J. M., et al. (2010). Modeling the low-latitude ionospheric electron density and plasma turbulence in the November 2004 storm period. Journal of Atmospheric and Solar-Terrestrial Physics, 72(4), 350-357. https://doi.org/10.1016/j.jastp.2009.12.005

4. Kintner, P., & Seyler, C. (1985). The status of observations and theory of high latitude ionospheric and magnetospheric plasma turbulence. Space Science Reviews, 41(1-2), 91-129. https://doi.org/10.1007/BF00241347

5. Su, C.-L., et al. (2024). Range aliasing effect of frequency modulation interrupted continuous wave (FMICW) high-frequency surface wave radar—Ionospheric clutter examination. IEEE Transactions on Geoscience and Remote Sensing, 62, 5111411. https://doi.org/10.1109/TGRS.2024.3482231

6. Zhang, R., et al. (2018). The unreasonable effectiveness of deep features as a perceptual metric. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 586-595. https://doi.org/10.1109/CVPR.2018.00068

7. Dai, Z., et al. (2023). A statistical analysis of conjugate equatorial plasma bubbles based on 630 nm airglow observations. Science China Earth Sciences, 66(3), 675-684. https://doi.org/10.1007/s11430-021-1013-4

8. Zhang, R., Liu, L., Ma, H., Chen, Y., & Le, H. (2022). ICON observations of equatorial ionospheric vertical ExB and field-aligned plasma drifts during the 2020–2021 SSW. Geophysical Research Letters, 49(16), e2022GL099238. https://doi.org/10.1029/2022GL099238