We have developed a lightweight, ultra‑wideband (UWB) stepped‑frequency ground penetrating radar (GPR) system that is fully integrated with an unmanned aerial vehicle (UAV drone) platform. The system is designed specifically for rapid and high‑precision detection of snow and ice thickness, including river ice, lake ice, land glaciers, and sea ice. The state and evolution of these frozen environments have a profound impact on regional and global climate systems, as well as on the prevention of natural disasters caused by ice melting, the projection of environmental and climate change, and the development and utilization of polar resources.
In this work, we present the complete system design, the methodology for automatic ice‑thickness estimation, and a series of physical experiments that validate the performance of our UAV‑drone GPR system. The core of the system is a stepped‑frequency radar transceiver module that operates from 900 MHz to 4 GHz, coupled with a custom‑designed Vivaldi antenna. The entire GPR subsystem (excluding the UAV drone) weighs less than 2.5 kg, which significantly reduces the payload on the drone and enhances its flight endurance. The radar is mounted on a DJI Matrice 600 Pro hexacopter, which has a maximum payload of 6 kg and can withstand winds up to 8 m·s⁻¹. The system also includes a real‑time kinematic (RTK) positioning module for centimeter‑level spatial referencing, a control and communication subsystem based on an embedded device, and a ground station for real‑time data visualization and processing.
Two main principles govern the ice‑thickness measurement. First, the radar emits a series of stepped‑frequency continuous waves that are swept across the desired bandwidth. After receiving the reflected signals, we convert them to the time domain via an inverse Fourier transform. Second, the thickness is derived from the travel‑time difference between the reflection from the air‑ice interface and that from the ice‑water or ice‑ground interface. The fundamental equation is:
$$ H = \frac{c (t_1 – t_0)}{2 \sqrt{\varepsilon_r}} $$
where \(H\) is the ice thickness, \(t_0\) and \(t_1\) are the two‑way travel times to the top and bottom of the ice layer, respectively, \(c = 3 \times 10^8\) m·s⁻¹ is the speed of light in vacuum, and \(\varepsilon_r\) is the relative permittivity of the ice. For freshwater ice, \(\varepsilon_r = 3.2\) is commonly used, while for saline ice the value can range between 4 and 8, depending on salinity and temperature.
System Architecture and Antenna Design
The GPR subsystem is built around a compact frequency‑domain transceiver with a maximum output power of 0 dBm (1 mW) and a power consumption of only 3.5 W. It supports transmission and reception from 1 MHz to 6 GHz, but we selected the 900 MHz–4 GHz band for optimal penetration and resolution in ice and snow. The Vivaldi antenna is a key component because of its ultra‑wideband capability, simple planar structure, and light weight. Our antenna is fabricated on a PTFE substrate (F4BM‑2, \(\varepsilon_r = 2.65\), thickness 1 mm) and has dimensions of 150 mm × 150 mm. The measured reflection coefficient \(|S_{11}|\) remains below –10 dB throughout the operating band, indicating excellent impedance matching. This antenna provides a directional radiation pattern that concentrates the electromagnetic energy toward the ground, which is essential for UAV drone operations where the antenna is typically 0.5–1.5 m above the surface.
| Parameter | Value |
|---|---|
| Frequency range | 900 MHz – 4 GHz |
| Number of frequency steps | 257 (configurable) |
| Output power | 0 dBm |
| Power consumption | 3.5 W |
| Total GPR weight (without drone) | < 2.5 kg |
| Antenna type | Vivaldi (150 mm × 150 mm) |
| RTK positioning accuracy | ±1 cm (horizontal) |
| Maximum UAV load capacity | 6 kg |
| Maximum flight speed (for GPR acquisition) | 3 m·s⁻¹ (recommended) |
Ground Experiments for Ice Thickness Measurement
To evaluate the performance of the system under controlled conditions, we carried out a series of ground tests. The GPR was suspended on a fixed support above artificial ice blocks, simulating the UAV drone flight scenario. We prepared three sets of ice samples: two freshwater ice blocks and one saline ice block with a salinity of 3.7 % by mass. The first freshwater block had dimensions 85 cm × 35 cm × 16.5 cm (thickness 16.5 cm), the second measured 60 cm × 45 cm × 40 cm, and the saline block measured 60 cm × 45 cm × 39.5 cm. A metal plate was placed beneath each ice block to create a strong reflection that mimics the ice‑water interface in real environments. The radar was operated with a time window of 25 ns and 257 sampling points, using the full 900 MHz–4 GHz bandwidth.
In the first experiment, we placed the radar at four different heights above the 16.5 cm freshwater ice: 0.50 m, 0.60 m, 0.90 m, and 1.02 m. The A‑scan signals clearly showed three reflection events: (1) the air‑ice interface, (2) the ice‑metal plate interface, and (3) a secondary multiple reflection. Using the travel‑time difference between events (1) and (2) and the known permittivity \(\varepsilon_r = 3.2\), we calculated the ice thickness. The results are summarized in Table 2.
| Antenna height (m) | Travel‑time difference (ns) | Measured thickness (cm) | Absolute error (cm) | Relative error (%) |
|---|---|---|---|---|
| 0.50 | 2.04 | 17.1 | 0.6 | 3.6 |
| 0.60 | 2.06 | 17.3 | 0.8 | 4.8 |
| 0.90 | 2.01 | 16.9 | 0.7 | 4.3 |
| 1.02 | 2.04 | 17.1 | 0.9 | 5.6 |
The relative errors ranged from 3.6 % to 5.6 %, demonstrating the good accuracy of the UAV drone GPR system for static measurements. We also observed a decrease in reflected signal amplitude with increasing height: the air‑ice reflection amplitude dropped from 0.28 (normalized) at 0.50 m to 0.11 at 1.02 m, and the ice‑metal reflection decreased from 0.83 to 0.21. This amplitude loss is due to geometric spreading in the air layer, which must be considered when planning actual UAV‑drone flights.
Comparison Between Freshwater Ice and Saline Ice
Sea ice is a complex mixture of pure ice, brine, and air, with much higher electrical conductivity and dielectric losses than freshwater ice. To investigate how these differences affect our system’s performance, we conducted comparative experiments using the second freshwater ice block (dimensions 60 cm × 45 cm × 40 cm) and the saline ice block (60 cm × 45 cm × 39.5 cm). Both blocks were placed over a metal plate to ensure a strong bottom reflection.
For the freshwater ice, the A‑scan signal revealed three distinct reflection events: (①) air‑ice, (②) internal cavity or unfrozen water, and (③) ice‑metal plate. The actual thickness of the ice above a large internal cavity was measured to be 14 cm after cutting the block, and the radar estimated 15 cm (using \(\varepsilon_r = 3.2\)), confirming the system’s ability to detect internal structures.
For the saline ice block, we found that the bottom metal plate reflection was severely attenuated and almost unidentifiable. Instead, we observed a series of continuous low‑amplitude echoes between 7 ns and 12 ns. Using two plausible permittivity values (\(\varepsilon_r = 4\) and \(\varepsilon_r = 8\)), we obtained thickness estimates of 12 cm and 8 cm, respectively, while the actual ice thickness above a large internal cavity was 19 cm. The absolute errors ranged from 7 cm to 11 cm. Several factors contribute to this poor performance:
- The higher permittivity of saline ice (4–8) leads to a slower wave speed and greater uncertainty in velocity (±30 %) compared to freshwater ice (±2 %).
- The irregular crystalline structure and internal layering of saline ice cause multiple scattering and continuous low‑amplitude reflections.
- During the experiment, the temperature rose to 27 °C–28 °C, causing some melting and the release of brine, which increased the electrical conductivity and further attenuated the electromagnetic wave.
To further illustrate the difference, we applied continuous wavelet transform (CWT) to the A‑scan signals. The wavelet transform is defined as:
$$ W(a,b) = \frac{1}{a} \int_{-\infty}^{\infty} s(t) \, \psi^{*}\!\left(\frac{t-b}{a}\right) dt $$
where \(a\) is the scale parameter, \(b\) is the translation parameter, and \(\psi(t)\) is the mother wavelet. The time‑frequency analysis showed that the energy of the reflections from saline ice was concentrated in the lower frequency range, and the high‑frequency components were more severely attenuated than those from freshwater ice. This frequency‑dependent attenuation is a direct consequence of the higher conductivity of the brine‑filled medium.
We also conducted dynamic (B‑scan) experiments by moving the ice block laterally beneath the stationary radar. After removing the direct wave and applying singular value decomposition filtering, the resulting images clearly revealed the internal cavity in the freshwater ice, while the saline ice image showed much weaker reflections and greater clutter. These results highlight the challenges that UAV‑drone GPR systems face when deployed over sea ice or highly saline environments.
| Parameter | Freshwater ice | Saline ice (3.7 % salinity) |
|---|---|---|
| Permittivity range | 3.2 | 4–8 |
| Wave speed uncertainty | ±2 % | ±30 % |
| Bottom reflection amplitude | Strong, clearly identifiable | Very weak, almost buried in clutter |
| Internal structure | Uniform, transparent | Layered, irregular crystals |
| Absolute error in thickness | ≤ 1 cm | 7–11 cm |
Field Flight Test: Snow Depth Measurement
In February 2024, our city experienced heavy snowfall and freezing rain, providing an excellent opportunity to test the UAV drone GPR system under real field conditions. We conducted flight tests over snow‑covered ground, with the radar mounted on the drone flying at an altitude of approximately 1 m above the ground and a speed of about 1 m·s⁻¹. Two areas with snow depths of 10 cm and 29 cm were surveyed.
The B‑scan images showed clear reflections from the air‑snow interface and the snow‑ground interface. The travel‑time differences were extracted from multiple A‑scan traces, and the snow thickness was calculated using the wave speed in snow, \(v_{\text{snow}} = 1.7 \times 10^8\) m·s⁻¹ (assuming dry snow). The results are summarized in Table 4.
| Actual snow depth (cm) | Average measured depth (cm) | Absolute error (cm) | Relative error (%) |
|---|---|---|---|
| 10 | 12 | 2 | 20 |
| 29 | 27 | 2 | 7 |
For the 29 cm snowpack, the relative error was only 7 %, demonstrating the reliability of the system for thick snow. For the thinner 10 cm snow, the relative error was higher (20 %), but the absolute error remained at 2 cm, which is still acceptable for many applications. The larger relative error for thin snow is partly due to the inhomogeneous nature of the snow layer and the difficulty in precisely picking the time‑delay from the radar signal when the two interfaces are very close in time.
Discussion
The experimental results confirm that our self‑developed UAV‑drone GPR system is capable of measuring ice and snow thickness with accuracy that meets practical requirements. The system’s lightweight design (under 2.5 kg) is a critical advantage because it allows the UAV drone to carry additional payloads (e.g., RTK, batteries) while maintaining a flight endurance sufficient for surveying tens of kilometers on a single charge. The use of a Vivaldi antenna with 900 MHz–4 GHz bandwidth provides a good balance between penetration depth and resolution. For ice, the penetration depth at these frequencies is typically several meters, which is adequate for most river, lake, and sea‑ice applications.
However, we identified several factors that limit the performance of the system, especially for saline ice. The high electrical conductivity and complex internal structure of sea ice cause strong attenuation and multiple scattering, making it difficult to detect the bottom interface. To improve the accuracy of ice‑thickness measurements in saline environments, future work should include the development of advanced signal‑processing algorithms that can compensate for the frequency‑dependent attenuation and separate the target reflection from clutter. Additionally, we plan to investigate the use of multi‑frequency or polarimetric measurements to better characterize the ice properties.
Another important consideration is the flight speed. During our tests, the UAV drone flew at 1 m·s⁻¹, but for operational efficiency, speeds up to 3 m·s⁻¹ are still acceptable because the spatial sampling interval at a pulse repetition rate of 26 ms per trace is about 8 cm. Higher speeds would lead to coarser sampling and potential loss of detail. The RTK module provided stable positioning, but we observed some minor fluctuations in the reflections caused by drone altitude variations. These can be mitigated by using an automated flight path with constant altitude.
Conclusion
We have designed, built, and tested a lightweight stepped‑frequency GPR system that is integrated with a UAV drone for snow and ice detection. The total weight of the GPR subsystem is less than 2.5 kg, and it operates over the 900 MHz–4 GHz band using a custom Vivaldi antenna. Ground experiments on freshwater ice showed relative errors of 3.6 % to 5.6 % at heights up to 1 m. Field flight tests on snow achieved a relative error of 7 % for 29 cm thick snow. The system also revealed significant differences in radar response between freshwater ice and saline ice: saline ice exhibited continuous low‑amplitude echoes and large thickness errors (7–11 cm) due to its higher permittivity, electrical conductivity, and irregular internal structure.
The proposed UAV‑drone GPR system provides a cost‑effective and efficient solution for rapid snow and ice thickness surveys. It can be deployed in remote or dangerous areas where manual measurements are impractical. Future work will focus on improving the robustness of the system for sea‑ice applications and on developing automated data‑processing tools that can handle the complex signals from saline ice. We believe that this technology will contribute significantly to climate research, water resource management, and polar exploration.