In the context of global climate warming, the state and evolution of snow and ice—including river and lake ice, terrestrial glaciers, and sea ice—significantly influence regional and global earth systems. Monitoring ice thickness, distribution, and dynamic changes in these environments is critical for preventing natural disasters caused by ice melt, predicting environmental and climate shifts, and supporting polar resource development. Conventional methods for ice thickness measurement, such as manual drilling, are labor-intensive, inefficient, and pose safety risks. Ground penetrating radar (GPR) has proven to be an effective tool for estimating ice thickness by measuring the travel time of electromagnetic pulses to the ice-water interface, primarily because electromagnetic waves exhibit extremely low attenuation in ice. However, deploying ground-based GPR in complex or steep terrains remains challenging due to operational difficulties and potential hazards. To overcome these limitations, we have developed a self-designed, lightweight, and low-power stepped-frequency ultra-wideband (UWB) GPR system mounted on a UAV drone platform, specifically tailored for rapid and high-precision snow and ice detection.
This paper presents a comprehensive study of our UAV drone-based GPR system, which operates in the frequency range from 900 MHz to 4 GHz. The entire radar subsystem—excluding the UAV drone platform itself—weighs less than 2.5 kg, significantly reducing the payload burden and enhancing flight endurance. The system integrates a UWB Vivaldi antenna, a frequency-domain transceiver module, a real-time kinematic (RTK) positioning module for centimeter-level accuracy, a control and communication subsystem based on a small embedded device, and a ground station for data processing and visualization. The modular design allows for flexible frequency band adjustments by swapping antennas, making it adaptable to various detection scenarios. We conducted ground-based physical experiments and field flight tests to evaluate the system’s performance, focusing on ice thickness measurement for both freshwater ice and saline ice, as well as snow depth detection under real-world conditions.
Our research objectives were threefold: first, to validate the accuracy and reliability of the UAV drone-mounted GPR system for ice thickness measurement under controlled conditions; second, to systematically compare the electromagnetic responses of freshwater ice and saline ice, as the latter’s complex composition—including brine pockets and irregular crystalline structures—can significantly affect GPR signal propagation; and third, to demonstrate the system’s effectiveness in field flight tests for snow depth mapping. By exploring these aspects, we aim to provide a cost-effective and efficient technical solution for snow and ice monitoring, with potential applications in polar regions, alpine environments, and winter infrastructure management.
The paper is organized as follows: Section 2 describes the system architecture and design principles, including the stepped-frequency technique and the antenna characteristics. Section 3 details the methodology for ice and snow thickness detection, including the underlying physics and signal processing techniques such as wavelet transform. Section 4 presents the experimental setup and results, organized into three subsections: freshwater ice detection at various heights, comparative analysis of freshwater and saline ice, and field flight tests for snow depth measurement. Section 5 provides a discussion of the findings, limitations, and future directions, followed by conclusions in Section 6.
System Architecture and Design Principles
The core of our system is the stepped-frequency GPR, which transmits a sequence of continuous-wave signals with linearly increasing frequency steps to achieve frequency sweeping across the desired band. While this approach results in a lower data acquisition rate compared to impulse GPR systems, it offers significant advantages in terms of detection flexibility and range resolution due to the controllability of each frequency point. The entire GPR subsystem is powered by a lightweight power supply unit and integrated onto a DJI Matrice 600 Pro UAV drone platform, which has a maximum payload of 6 kg and can withstand wind speeds up to 8 m/s. The main subsystems include:
- Transceiver Module: A compact frequency-domain module with a maximum output power of 0 dBm (1 mW) and a power consumption of 3.5 W, capable of efficient signal transmission and reception from 1 MHz to 6 GHz.
- Positioning Subsystem: An RTK module providing ±1 cm horizontal positioning accuracy in FIX mode, with data output via standard NMEA-0183 protocol for precise spatial mapping.
- Control and Communication Subsystem: Based on a small embedded device, it manages the transceiver module via dedicated commands (e.g., start/stop, frequency range setting, data storage) and supports both local data storage and real-time data transmission via Wi-Fi.
- Ground Station: A tablet or laptop that interacts with the UAV drone via a self-organized wireless network for command transmission and data processing.
The Vivaldi antenna is a key component of our system due to its ultra-wideband characteristics, simple structure, light weight, and ease of fabrication. We designed and manufactured an antipodal Vivaldi antenna covering 600 MHz to 6 GHz, with dimensions of 150 mm × 150 mm. The antenna is printed on a PTFE substrate (F4BM-2) with a relative permittivity of 2.65 ± 0.1 and a thickness of 1 mm. The radiation part and feed structure consist of thin copper foil on the substrate, with an exponentially tapered inner edge on the front side to optimize radiation efficiency, and parallel slot lines to enhance gain and directivity. The reflection coefficient (S11) remains below -10 dB across the 900 MHz to 4 GHz operating band, indicating good impedance matching.
| Parameter | Value/Specification |
|---|---|
| Frequency Range | 900 MHz – 4 GHz |
| Antenna Type | Antipodal Vivaldi |
| Antenna Dimensions | 150 mm × 150 mm × 1 mm |
| Substrate Material | PTFE (F4BM-2) |
| Relative Permittivity (Substrate) | 2.65 ± 0.1 |
| Total System Weight (GPR) | < 2.5 kg |
| Positioning Accuracy | ±1 cm (RTK, FIX mode) |
| Maximum Transmit Power | 0 dBm (1 mW) |
| Power Consumption | 3.5 W |
| UAV Drone Platform | DJI Matrice 600 Pro, max payload 6 kg |
Methodology for Ice and Snow Thickness Detection
GPR operates by transmitting electromagnetic waves and receiving the reflections from subsurface interfaces. The detection mechanism relies on the contrast in electrical properties, particularly the dielectric permittivity, between adjacent media. When electromagnetic waves propagate from air into ice, part of the energy is reflected at the air-ice interface, and the transmitted component continues to travel through the ice until it encounters the ice-water or ice-bedrock interface, where another reflection occurs. By measuring the two-way travel time difference between these reflections, we can estimate the ice thickness using the following formula:
$$
H = \frac{c (t_1 – t_0)}{2 \sqrt{\varepsilon_r}}
$$
where \(H\) is the ice thickness, \(t_0\) and \(t_1\) are the two-way travel times of the reflections from the upper and lower ice surfaces, respectively, \(c = 3 \times 10^8\) m/s is the speed of light in free space, and \(\varepsilon_r\) is the relative permittivity of ice. For freshwater ice, \(\varepsilon_r \approx 3.2\), resulting in a wave velocity of approximately \(1.68 \times 10^8\) m/s. For saline ice, the permittivity is higher (typically ranging from 4 to 8), leading to slower wave speeds and higher velocity uncertainty. The large dielectric contrast between ice (εr ≈ 3.2) and water (εr ≈ 81) ensures a strong reflection at the ice-water interface, facilitating thickness estimation.
To enhance signal analysis, we applied the continuous wavelet transform to the time-domain GPR signals. Wavelet transform provides multi-resolution analysis by decomposing the signal into time-frequency components, making it possible to identify localized variations in both time and frequency domains. For a given GPR signal \(s(t)\), its continuous wavelet transform \(W(a,b)\) is computed as:
$$
W(a,b) = \frac{1}{\sqrt{a}} \int_{-\infty}^{\infty} s(t) \psi^{*} \left( \frac{t – b}{a} \right) dt
$$
where \(a\) is the scale parameter (controlling the width of the wavelet), \(b\) is the translation parameter (determining the position along the time axis), and \(\psi^{*}(t)\) is the complex conjugate of the mother wavelet function \(\psi(t)\). This approach allowed us to analyze the time-frequency characteristics of ice reflections and differentiate between signal features from air-ice, ice-water, and ice-metal interfaces.
Experimental Setup and Results
To evaluate the performance of our UAV drone GPR system, we designed a series of controlled laboratory experiments and a field flight test. The laboratory experiments simulated airborne detection conditions by using a fixed support structure to suspend the GPR system above ice blocks, which were moved manually to simulate dynamic measurements. The time window was set to 25 ns with 257 sampling points, and the operating frequency band was 900 MHz to 4 GHz. Ice blocks of varying thicknesses and types were prepared by freezing pure water (freshwater ice) and saline water with a salinity of 3.7% by mass (saline ice) in plastic containers at -40°C for one week.
Freshwater Ice Detection at Different Heights
We conducted measurements on a freshwater ice block (85 cm × 35 cm × 16.5 cm) at four different antenna heights above the ice surface: 0.50 m, 0.60 m, 0.90 m, and 1.02 m. The A-scan waveforms revealed three characteristic reflection events: the first corresponding to the air-ice interface, the second to the ice-metal plate interface (a metal plate was placed beneath the ice to simulate the ice-water/bedrock interface), and the third to a secondary multiple reflection. The results are summarized in the table below.
| Antenna Height (m) | Actual Ice Thickness (cm) | Measured Ice Thickness (cm) | Absolute Error (cm) | Relative Error (%) |
|---|---|---|---|---|
| 0.50 | 16.5 | 17.1 | 0.6 | 3.6 |
| 0.60 | 16.5 | 17.3 | 0.8 | 4.8 |
| 0.90 | 16.2 | 16.9 | 0.7 | 4.3 |
| 1.02 | 16.2 | 17.1 | 0.9 | 5.6 |
These results confirm that our UAV drone-mountable GPR system can measure freshwater ice thickness with an absolute error of approximately 1 cm and a relative error ranging from 3.6% to 5.6% under heights of up to 1 m. The slight variation in actual ice thickness was due to melting caused by the ambient temperature of 27–28°C during the experiment. Furthermore, the signal amplitude of both the air-ice and ice-metal interfaces decreased as the antenna height increased, from normalized amplitudes of 0.28 and 0.83 at 0.50 m to 0.11 and 0.21 at 1.02 m, respectively. This attenuation is primarily due to geometrical spreading loss in free space, highlighting the importance of selecting an appropriate flight height for UAV drone-based measurements.
Comparative Detection of Freshwater Ice vs. Saline Ice
To investigate the effect of ice composition on GPR signals, we compared a freshwater ice block (60 cm × 45 cm × 40 cm) and a saline ice block (60 cm × 45 cm × 39.5 cm) with a salinity of 3.7%. Both blocks were placed on a metal plate during the experiments. Visual inspection revealed significant structural differences: freshwater ice was homogeneous and transparent, while saline ice exhibited irregular crystalline structures and distinct layering due to salt rejection during freezing. Additionally, the saline ice block contained a large internal cavity (48 cm × 33 cm × 20 cm) beneath a 19 cm thick upper layer.
The A-scan signals for the two ice types showed markedly different characteristics. For freshwater ice, the reflection from the ice-metal plate was clear and strong, allowing accurate thickness estimation. The measured thickness of the ice above the internal cavity was approximately 15 cm, which matched the actual value of 14–16 cm. In contrast, the saline ice produced continuous low-amplitude echoes from 7 to 12 ns, and the reflection from the metal plate was so attenuated that it was difficult to identify. Using a permittivity range of 4 to 8 for saline ice, the estimated ice thickness above the cavity was only 8 cm (for εr = 8) to 12 cm (for εr = 4), resulting in absolute errors of 7 cm to 11 cm compared to the actual 19 cm. This substantial error is attributed to the higher conductivity of the saline ice due to brine drainage during melting, as well as the complex internal structure causing multiple reflections and scattering.
Wavelet transform analysis further illustrated the time-frequency differences. The signal energy was concentrated around the system’s center frequency (2–3 GHz) at the air-ice interface for both ice types. However, for saline ice, the high-frequency components attenuated more rapidly, and the energy of the lower interface reflection showed a downshift in frequency, making it nearly undetectable after 6 ns. In contrast, freshwater ice maintained a clear reflection at around 10 ns. These findings confirm that high salinity and internal crystalline structure significantly alter GPR signal propagation, posing challenges for accurate thickness estimation of sea ice using UAV drones.
The B-scan images of dynamic measurements (where the ice blocks were moved beneath the stationary radar) confirmed these observations. Freshwater ice data clearly showed the internal cavity and the metal plate reflection at the bottom, although strong background clutter was present due to multiple reflections. Saline ice data, while still able to delineate the cavity, exhibited much weaker signal strength and lower contrast, making feature identification more difficult. The presence of double hyperbolic reflections in the saline ice B-scan was attributed to the manual back-and-forth movement of the ice block during the experiment.
Field Flight Test for Snow Depth Detection
In February 2024, during a period of heavy snowfall and freezing rain in Wuhan, we conducted a field flight test to evaluate our UAV drone GPR system for snow depth measurement. The UAV drone flew at an altitude of approximately 1 m above the ground, with a flight speed controlled at 1 m/s. The snowpack had varying thicknesses of 10 cm and 29 cm at different locations.
The B-scan profiles clearly showed reflections from the air-snow interface (upper boundary) and the snow-ground interface (lower boundary). To validate the accuracy of travel time extraction, we performed a calibration experiment by varying the distance between the antenna and a metal plate (0.51 m to 0.72 m in 0.07 m increments). The measured reflection travel times showed an average absolute error of 0.075 ns and an average relative error of 1.83% compared to theoretical values, confirming the reliability of our timing measurements.
Using multiple A-scan traces and the snow wave velocity \(v_{\text{snow}} = 1.7 \times 10^8\) m/s, we calculated the average snow thicknesses to be 12 cm and 27 cm for the two locations, respectively. The results are summarized below.
| Actual Snow Thickness (cm) | Measured Snow Thickness (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 effectiveness of our UAV drone GPR system for reliable snow depth mapping. However, for the thinner snowpack (10 cm), the relative error increased to 20%, even though the absolute error remained 2 cm. This discrepancy was likely due to uneven snow distribution on the ground, as variations of 1–2 cm in different A-scan traces were observed. Overall, these field results validate the system’s practical utility in real-world snow and ice monitoring applications using UAV drones.
Discussion
The experimental results collectively demonstrate that our self-developed lightweight stepped-frequency UAV drone GPR system exhibits promising detection capabilities for both ice and snow thickness measurements. The ground tests on freshwater ice produced relative errors of 3.6% to 5.6%, while the field flight test on a 29 cm snowpack achieved a relative error of 7%. These values are within acceptable ranges for many environmental and engineering applications, such as monitoring river ice, assessing winter road safety, and studying glacier mass balance.
The comparative study of freshwater ice and saline ice highlights a critical challenge for UAV drone-based sea ice detection. The continuous low-amplitude echoes observed in saline ice signals are primarily caused by two factors: (1) the wide range of dielectric permittivity (4–8) for saline ice, leading to up to ±30% uncertainty in wave velocity estimation, and (2) the irregular crystalline structure and internal layering that cause multiple scattering and increased signal attenuation. Furthermore, the release of brine from melting saline ice increases the bulk conductivity, which further attenuates electromagnetic wave energy. In our experiments, the absolute error for saline ice thickness estimation was 7–11 cm, more than ten times higher than that for freshwater ice. Natural sea ice, with its even more complex composition and dynamic processes (e.g., temperature and salinity variations), is expected to pose even greater challenges. Future work should focus on developing multi-parameter inversion techniques and advanced signal processing algorithms to mitigate these effects and improve accuracy.
During actual UAV drone flights, the DJI M600 Pro platform can reach a maximum horizontal speed of 18 m/s. However, at this speed, our GPR system’s spatial sampling interval becomes 0.50 m (since each A-scan acquisition takes 26 ms), which may be insufficient for high-resolution detection. To balance efficiency and accuracy, we recommend controlling the flight speed at or below 3 m/s, corresponding to a sampling interval of ≤8 cm. Additionally, human-operated flight paths often introduce altitude fluctuations that cause continuous shifts in B-scan reflection waveforms. Using pre-programmed autonomous flight routes can minimize such errors and improve data quality. The results also show that a flight altitude of approximately 1 m is a reasonable compromise between reducing geometrical spreading losses and maintaining safe clearance above uneven terrain or obstacles.
Despite the promising results, this study has several limitations. First, we only validated the system up to a depth of about 40 cm (for ice) and 29 cm (for snow), and further tests in deeper snow or ice settings are needed to fully characterize the penetration depth at 900 MHz to 4 GHz. Second, our experiments were conducted in a relatively controlled environment (laboratory and urban frozen snow) rather than true polar or alpine conditions. Factors such as low temperatures, high wind speeds, large-scale terrain undulations, and varying salinity levels in natural sea ice were not fully accounted for. Future studies should test the system in real glacier and sea ice environments to assess its robustness and adaptability. We also plan to optimize the antenna design (e.g., multi-band integration), increase the signal transmission power, and improve signal processing algorithms to enhance deep detection capabilities and environmental resilience.
Conclusion
We have successfully developed a lightweight, low-power stepped-frequency ultra-wideband ground penetrating radar system designed for deployment on UAV drones. The total weight of the GPR subsystem is under 2.5 kg, which minimizes the payload on the UAV drone and significantly extends its flight endurance. The system incorporates a Vivaldi antenna operating from 900 MHz to 4 GHz, an RTK positioning module for precise spatial referencing, and a modular control architecture for flexible operation. Through a series of controlled ground experiments and a field flight test, we have validated the system’s effectiveness in measuring ice and snow thickness.
Key findings from our study include:
- The system can measure freshwater ice thickness with a relative error of 3.6% to 5.6% at antenna heights up to 1 m, confirming its accuracy for non-contact ice profiling.
- In field flight tests, the system measured a 29 cm snowpack with a relative error of 7%, demonstrating its practical utility in real-world scenarios using UAV drones.
- Saline ice presents a significant challenge due to its higher permittivity uncertainty, increased conductivity, and complex internal structure, resulting in thickness errors of 7–11 cm—an order of magnitude larger than for freshwater ice. This finding underscores the need for improved data processing methods for sea ice applications.
The proposed system offers a cost-effective, efficient, and safe alternative to traditional manual drilling or ground-based GPR for snow and ice monitoring. By mounting on widely available UAV drones, it enables rapid deployment over large or inaccessible areas, making it particularly valuable for polar research, alpine glacier monitoring, and winter infrastructure management. Future work will focus on field validation in diverse ice and snow environments, as well as advancements in hardware and software to further improve the system’s performance and versatility.