Performance Analysis of an Ice-Based Buoy Operating from the Packed Ice Zone to the Marginal Ice Zone with an Imaging System

Abstract

Arctic sea ice can be regarded as a sensitive indicator of climate change, and it has declined dramatically in recent decades. The swift decline in Arctic sea ice coverage leads to an expansion of the marginal ice zone (MIZ). In this study, an ice-based buoy with an imaging system is designed for the long-term observation of the changes in sea ice from the packed ice zone to the marginal ice zone in polar regions. The system composition, main buoy, image system, and buoy load were analyzed. An underwater camera supports a 640 × 480 resolution image acquisition, RS485 communication, stable operation at –40 °C, and long-term underwater sealing protection through a titanium alloy housing. During a continuous three-month field deployment in the Arctic, the system successfully captured images of ice-bottom morphology and biological attachment, demonstrating imaging reliability and operational stability under extreme conditions. In addition, the buoy employed a battery state estimation method based on the Extreme Learning Machine (ELM). Compared with LSTM, BP, BiLSTM, SAELSTM, and RF models, the ELM achieved a test set performance of RMSE = 0.05 and MAE = 0.187, significantly outperforming the alternatives and thereby improving energy management and the reliability of long-term autonomous operation. Laboratory flume tests further verified the power generation performance of the wave energy-assisted supply system. However, due to the limited duration of Arctic deployment, full year-round performance has not yet been validated, and the imaging resolution remains insufficient for biological classification. The results indicate that the buoy demonstrates strong innovation and application potential for long-term polar observations, while further improvements are needed through extended deployments and enhanced imaging capability.

1. Introduction

Arctic sea ice can be regarded as a sensitive indicator of climate change and has declined dramatically in recent decades [1,2,3]. Its rapid retreat has not only contributed to Arctic temperature amplification [1] but has also resulted in a long-term reduction in sea ice thickness [2,3]. In the study of sea ice state, thickness is considered the most comprehensive parameter, while snow cover significantly affects the energy exchange at the ice–atmosphere interface [4,5]. The formation and evolution of melt ponds play an important role in sea ice stability and melting processes [6,7,8,9]. With increasing ice permeability, more melt ponds directly connect with seawater, making sea ice more fragile [7,9]. Although various techniques have been developed to observe sea ice thickness and snow depth [10,11], significant uncertainties remain in the marginal ice zone (MIZ) under complex conditions.

The MIZ, as a transitional region between compact ice and open water, has expanded rapidly in recent years with the decline of sea ice [12,13]. In this region, sea ice thickness and dynamics exhibit strong variability [14,15,16], making it difficult for existing numerical models and remote sensing methods to accurately capture its evolution. To address this challenge, a number of ice-based buoys, such as IMB and SIMBA, have been deployed [17,18], and local observations have been obtained using buoy arrays [19] or imaging techniques [20]. Recent developments have also been made by several international teams. For example, the National Snow and Ice Data Center (NSIDC) and NOAA have continuously deployed IMB and SIMB buoys in the Arctic to monitor sea ice thermodynamics. European research groups have carried out marginal ice zone campaigns near Svalbard using drifting buoys to investigate wave–ice interactions and ice dynamics. In addition, Chinese Arctic expeditions have deployed drifting buoy arrays during cruises, providing valuable data on sea ice drift and surface meteorological parameters. These efforts highlight the growing use of buoy-based systems while also underscoring the lack of imaging capability and limited adaptability to diverse polar conditions.

In recent years, several initiatives have further advanced marginal ice zone (MIZ) observations. For instance, Rabault et al. reported the deployment of 79 OpenMetBuoys around Svalbard during 2022–2023, providing high-resolution datasets of waves and sea ice drift in the MIZ [21]. A distributed Arctic MIZ observation network was also established to systematically compare in situ buoy data with model forecasts [22]. Complementarily, Zhu et al. generated a 12-winter record of wave-affected MIZs from CryoSat-2 altimetry, revealing pronounced variability in MIZ width across seasons [23]. Wahlgren et al. deployed drifting SWIFT buoys in the Weddell Sea, providing direct in situ observations of wave–sea ice interactions in the Antarctic marginal ice zone, which further enriches our understanding of MIZ dynamics [24]. However, these approaches still face limitations: remote sensing offers wide spatial coverage but cannot resolve under-ice processes; traditional buoys provide continuous thickness and temperature monitoring but lack imaging capability; and shipborne or radar observations achieve high accuracy but are costly and unsuitable for long-term deployment [25].

To overcome these limitations, this study proposes an innovative ice-based buoy equipped with an imaging system to capture the evolution of under-ice morphology. The buoy adopts a titanium alloy housing with a sapphire glass window, enabling pressure resistance, corrosion resistance, and long-term operation under extreme low temperatures. In addition, an Extreme Learning Machine (ELM)-based method is introduced for battery state estimation. The proposed method achieved a prediction accuracy of RMSE = 0.05 and MAE = 0.187, outperforming LSTM, BP, BiLSTM, SAELSTM, and RF models, thereby enhancing the reliability of long-term autonomous operation. Through both laboratory flume tests and a three-month Arctic field deployment, the buoy’s adaptability under multiple environments was validated.

The main innovations of this work are as follows:

(1)

A novel imaging-based buoy system with a titanium alloy housing and sapphire window was proposed and field-validated, capable of stable operation at −40 °C in complex under-ice environments and continuously capturing under-ice imagery for three months, addressing the lack of imaging in traditional buoys;

(2)

The application of the ELM algorithm significantly improved battery state prediction accuracy (RMSE = 0.05, MAE = 0.187), providing reliable energy assurance for long-term polar buoy operation;

(3)

Low-temperature imaging, intelligent battery prediction, and wave energy-assisted power supply were integrated into a single buoy platform, forming a low-cost, long-term, and autonomous polar observation system with potential for multi-scenario applications.

2. Description of the Ice-Based Buoy with Imaging System

2.1. System Composition

The ice-based buoy with an imaging system is designed for the long-term observation of the changes in sea ice from the packed ice zone to the marginal ice zone in polar regions. It contains an integrated meteorological station, sea ice sensors, and an imaging system for meteorological, environmental, and sea ice observations. The types of observation parameters include wind speed, wind direction, air temperature, humidity, air pressure, sea ice temperature profile, and sea ice images obtained by a freeze-resistant camera system and an underwater camera. Observations of the wind speed, wind direction, air temperature, humidity, and air pressure above the sea ice surface provide basic meteorological data, which can be used in the calculation of surface energy balance. The camera system operating in low-temperature environments should be installed at a height of 1 m above the buoy, which can observe and record the evolution of sea ice. The phenomena of snowfall, freezing rain, polar bears, and ice bottom melting that occur in the polar regions can be captured by cameras in this system. The sea ice temperature can be measured using a thermistor chain, which can be installed through a support structure connected to the buoy. There are three other subsystems in the ice-based buoy, namely the power supply subsystem, control subsystem, and data transmission subsystem. A wave energy generation device and battery form the main components of the power supply subsystem. GNSS, satellite modules, waterproof interfaces, and low temperature-resistant cables are also components of ice-based buoys. The buoy is designed with a low-power strategy in which most subsystems remain in sleep mode and are only activated during data acquisition or transmission. The GNSS module provides a typical positioning accuracy of about 2–3 m in open-sky conditions, with power consumption of 0.1–0.2 W during active positioning and less than 10 mW in standby mode. The Iridium satellite communication unit supports a transmission rate of 2.4 kbps and requires 5–8 W only during short bursts of less than two seconds, corresponding to an average consumption of 0.05–0.1 W. The imaging system consumes 0.5–1 W when capturing ice or underwater images but remains in sleep mode otherwise, resulting in an average of 0.05–0.1 W. The controller and environmental sensors operate with less than 0.1 W on average. Overall, the buoy’s average power consumption is maintained at approximately 0.3–0.5 W, which enables reliable long-term deployment under polar conditions.

2.2. Design of Ice-Based Buoy with Imaging System

2.2.1. Main Buoy

The main buoy is a rigid sphere structure made of polyvinyl chloride, as shown in Figure 1. The diameter of the main buoy is 0.8 m. A rigid and watertight electronic cabin is made of 7075 aluminum alloy, which is installed inside the main buoy. A GNSS, an integrated meteorological station, a satellite module and an antenna, a power supply subsystem, a control subsystem, and a data transmission subsystem are placed and fixed in the electronic cabin. A stainless steel base is installed at the bottom of the main buoy. A supporting structure is connected to the main body through a waterproof plug, and a pan–tilt device and low temperature-resistant camera are installed at the top of the structure. The underwater camera and thermistor chain are installed on a support that runs through the sea ice and into the seawater. The bottom of the main buoy is a base. The diameter of the base is 0.4 m, which allows the main buoy to be easily deployed on sea ice.

2.2.2. Image System

The camera system consists of an ice camera and an underwater camera. The ice-based buoy incorporates an ice camera, which is an improved version of a serial port camera (Figure 2a). Encased in a sealed shell (Figure 2c), this serial port camera boasts a compact design and ease of integration. The camera fulfills functions encompassing image capture, compression, and serial output. It is equipped with a high-performance image processor capable of compressing raw images and outputting images in the standard format. Operating the serial port camera is straightforward, as it responds to simple commands for tasks such as resetting, capturing images, adjusting baud rates, and selecting image sizes. The scheme of the camera shooting can be seen in

Table 1. Scheme of the camera shooting.

1: Start Procedure
2: Initialization
Power on.
Delay for 3 s.
Set the size of the captured image using the command code of 56 00 31 05 04 01 00 19 00/11/22. Set the compression rate for the captured image using the command code of 56 00 31 05 01 01 12 04 00~FF. Restart.
3: Photography process
Receive photography instruction.
Read the length of the captured image.
Read the data of the captured image.
Clear image cache.
Restart.
4: End Procedure

.

As shown in Figure 3, the underwater camera consists of a waterproof case, a small camera, a waterproof interface and an observation window (sapphire lenses).

The waterproof case is an important component of the underwater camera. Good water tightness is necessary to ensure the long-term stable operation of the underwater image observation. When designing the waterproof case, the following requirements should be met:

(1)

Good sealing and high material strength;

(2)

It has corrosion resistance and can ensure the long-term operation of the camera in seawater;

(3)

The front observation window lens is thin and has strong light transmission ability, with low light energy loss.

Thus, for the waterproof case, the upper and lower covers are made of titanium alloy, which has good strength and corrosion resistance. The observation window is made of sapphire glass with high hardness and good pressure resistance. After the camera is installed inside the waterproof case, the upper cover, lower cover, and “O” ring are sealed. The underwater camera needs to undergo water tightness testing in a water pipe (Figure 4).

2.3. Analysis of Buoy Load

The ice-based buoy can survive and operate in sea ice or the ocean. To cope with the severe melting of polar sea ice and expand the observation period of buoys, the survival ability of buoys in the marine environment has been enhanced. As shown in Figure 1b, the ice-based buoy in the ocean is subjected to the external force load of the wind, waves, and current at the sea surface and affected by gravity and buoyancy, where the buoy body provides reserve buoyancy.

The wind force of the ice-based buoy can be divided into vertical wind force and horizontal wind force, respectively. The vertical wind force can be ignored in this study. The wind force of the ice-based buoy can be described as follows.

$$F_{Wind}=F_{Wind}^H={\displaystyle \frac{1}{2}}{\rho }_{a}S{\left(V_{Wind}\right)}^2{C}_{AR}$$

(1)

where F_Wind is the wind force of the ice-based buoy; $F_{Wind}^H$ is the horizontal wind force; ρ_a is the air mass density, 0.1228 kg s² m⁻⁴; S is the frontal area acreage of the buoy, 0.3 m²; V_Wind is the wind speed; and C_AR is the air resistance coefficient, 0.75.

The current force of the ice-based buoy can be described as follows.

$$F_{Current}=F_{Current}^H={\displaystyle \frac{1}{2}}{\rho }_w{S}^{\prime }{\left(V_{Curremt}\right)}^2{C}_{WR}$$

(2)

where F_Current is the current force of the ice-based buoy; $F_{Current}^H$ is the horizontal current force; ρ_w is the water mass density, 104.49 kg s² m⁻⁴; S’ is the frontal area acreage of the buoy below the buoy waterline, 0.32 m²; V_Current is the current speed; and C_WR is the water resistance coefficient, 0.3. The vertical current force can be ignored.

The wave force of the ice-based buoy also can be divided into vertical wind force and horizontal wind force, respectively.

$$F_{Wave}^H={\rho }_w{A}_{W}Vol{\displaystyle \frac{4{\pi }^2}{{\left(T_W\right)}^2}}exp\left(-{\displaystyle \frac{4{\pi }^{2}D}{{\left(T_W\right)}^{2}g}}\right)C_{IF}$$

(3)

where $F_{Wave}^H$ is the horizontal wave force; ρ_w is the water mass density; A_W is the max wave amplitude, 10 m; Vol is the buoy wet volume, 0.29 m³; T_W is the wave period; D is the draught depth, 0.41 m; g is the acceleration of gravity; and C_IF is the inertial force coefficient with respect to the horizontal plane, 2.

$$F_{Wave}^V={\displaystyle \frac{1}{2}}{\rho }_w{A}_W{S}_V{\displaystyle \frac{4{\pi }^2}{{\left(T_W\right)}^2}}exp\left(-{\displaystyle \frac{8{\pi }^{2}D}{{\left(T_W\right)}^{2}g}}\right)C_D$$

(4)

where $F_{Wave}^V$ is the vertical wave force; S_V is the acreage of the head wave with respect to the vertical plane, 0.3 m²; and C_D is the drag coefficient with respect to the vertical plane, 0.3.

The net force can be regarded as generated by the weight and buoyancy of the ice-based buoy. The horizontal force of the buoy is equal to $F_{Buoy}^H$.

$$F_{Wind}^H+F_{Current}^H+F_{Wave}^H=F_{Buoy}^H$$

(5)

$F_{Wind}^H$ is 67.69 kg, $F_{Current}^H$ is 180.56 kg, $F_{Wave}^H$ is 123.36 kg. $F_{Buoy}^H$ is equal to the horizontal force of buoy, 371.61 kg.

3. State Estimation of Battery of Ice-Based Buoy Based on ELM

Effective estimation of buoy battery status can predict the safe and reliable operating cycle of equipment. Neural networks are widely used for time series prediction tasks because of their ability to capture nonlinear dynamics. However, conventional feedforward networks (e.g., BP) are slow to train and prone to local minimum, while recurrent models (e.g., LSTM and BiLSTM) provide strong sequence modeling but at the cost of high computational requirements, which makes them unsuitable for low-power buoy systems. To balance prediction accuracy and efficiency, this study adopts a prediction model based on the time series features of the Extreme Learning Machine (ELM) to estimate the state of batteries. ELM is an efficient learning algorithm specifically designed for single-hidden-layer feedforward neural networks (SLFNs). This algorithm simplifies the complex parameter-tuning process of traditional neural networks, significantly improves learning speed, and ensures good generalization performance, making it suitable for handling classification and regression problems in engineering applications. The training process of the ELM algorithm is as follows:

• Random Initialization: The weights W from the input layer to the hidden layer and the bias b are set randomly.
• Calculate Hidden Layer Output: For each input sample matrix X, the hidden layer output matrix H is computed using the formula as follows:

  $$H=g\left(WX+b\right)$$

  (6)

  where g is the activation function.
• Compute Output Layer Weights: The output layer weights β are solved using the least squares method as follows:

  $$\beta =H^{*}Y$$

  (7)

  where H* is the pseudoinverse of the hidden layer output matrix H, and Y is the target output.
• Prediction: For the input sample matrix X_new, the hidden layer output matrix H_new is calculated as:

  $$H_{new}=g\left(W{X}_{new}+b\right)$$

  (8)

After obtaining the hidden layer output, the subsequent values of X_new are predicted, i.e., the predicted output matrix Y_new is as follows:

$$Y_{new}=H_{new}\beta$$

(9)

The sparrow search algorithm (SSA) is an intelligent search algorithm that can simulate sparrow foraging behavior and antipredation behavior [26]. The SSA is selected as the machine learning optimization method in this study. Set the population size as N, the dimension of each sparrow as D, and the search space boundaries as [x_min, x_max]. The initial population matrix is defined as X = {x_i,j}_N×D, where i = 1, 2, …, N and j = 1, 2, …, D. Additionally, the fitness value f of each sparrow needs to be calculated, with Fx representing the fitness values of all sparrows. Sparrows with higher fitness values are prioritized in obtaining food during the search process, thereby guiding the entire population toward better foraging directions and regions. The location update formula of the discoverers in SSA is described as follows:

$$X_{i,j}^{t+1}=\left\{\begin{array}{ll}{X}_{i,j}^{t}exp\left({\displaystyle \frac{-i}{{\alpha iter}_{max}}}\right),& R_2<ST\\ X_{i,j}^t+QL,& R_2\ge ST\end{array}\right.$$

(10)

where $X_{i,j}^t$ represents the jth dimension of the ith sparrow of generation t; T_max is the maximum number of iterations; α is the uniform random number in the range of (0, 1]; Q is a random number that follows the standard normal distribution; and L is the d-dimensional row vector, where every entry is 1.

Where $X_{i,j}^t$ represents the value of the ith sparrow in the jth dimension during the tth iteration; iter_max denotes the current maximum number of iterations; α is a random number between 0 and 1; R₂ represents the alarm value ranging from 0 to 1; and ST is the safety threshold used to assess the danger level during foraging, typically set to 0.5. Q is a random number following a normal distribution, and L is a 1 × d matrix with all elements equal to 1. When R₂ < ST, the sparrows with higher fitness engage in extensive exploration. Conversely, when R₂ ≥ ST, all sparrows quickly relocate to safer areas to evade threats. The follower location update formula can be described as follows:

$$X_{i,j}^{t+1}=\left\{\begin{array}{ll}Qexp\left({\displaystyle \frac{X_{worst}^t-X_{i,j}^t}{{\alpha iter}_{max}}}\right),& i>{\displaystyle \frac{n}{2}}\\ X_p^{t+1}+\left|X_{i,j}^t-X_p^{t+1}\right|A^{+}L,& i\le {\displaystyle \frac{n}{2}}\end{array}\right.$$

(11)

where $X_{i,j}^t$ represents the latest position of the follower; X_p denotes the optimal position occupied by the discoverer; X_worst is the current globally explored worst position; L is a 1 × d matrix with elements randomly assigned as 1 or −1; and A⁺ is the Moore–Penrose generalized inverse of matrix A. When i > n/2, it indicates that the ith follower with poorer fitness may be allocated insufficient energy upon joining and thus needs to relocate to other regions.

When a sparrow encounters danger during foraging, its position is updated as follows:

$$X_{i,j}^{t+1}=\left\{\begin{array}{ll}{X}_{best}^t\beta \left|X_{i,j}^t-X_{best}^t\right|,& f_i>f_j\\ X_{i,j}^t+K\left({\displaystyle \frac{\left|X_{i,j}^t-X_{worst}^t\right|}{\left(f_t-f_j\right)+\epsilon }}\right),& f_i=f_j\end{array}\right.$$

(12)

where $X_{i,j}^{t+1}$ represents the updated position of a sparrow that has detected danger; $X_{best}^t$ denotes the current optimal position; β is a random number following a standard normal distribution (mean value is 0, variance is 1), and K is a random number uniformly distributed between −1 and 1. f_i and f_j are the current global best and worst fitness values, respectively; while ε is an extremely small constant used to prevent division by zero. If f_i > f_j, it indicates safety; if f_i = f_j, it signifies danger awareness. The Sparrow Search Algorithm (SSA) iteratively updates the positions of discoverers and followers through the above steps until the termination criteria are met, ultimately outputting the optimal position.

In this study, the data transmitted by the buoy in the polar region can be stored in the parameter variable area of the database, and data cleaning, abnormal data removal, and stability judgment processing can be used. Then we can divide the stored data into training and test sets, determine experimental parameters, generate a population for the training data, and perform training using a neural network. In the end, we construct an ELM network model to perform voltage prediction tasks based on time series and calculate the correlation coefficient of prediction accuracy. The data transmitted back by the buoys deployed in the Amundsen Sea will be input into the ELM model for training. The predicted and actual values of the ELM algorithm on the test set are shown in Figure 5. The blue circles in Figure 5 mark the points corresponding to the best training perfor-mance. As can be seen from Figure 5, the predicted values generally follow the actual values. The root mean square error (RMSE) is 0.05, the mean absolute error (MAE) is 0.187. From the above data, it can be concluded that the ELM model exhibits good predictive performance and accurate statistical modeling.

The ELM model was compared with the Long Short-Term Memory (LSTM) model, BP neural network model, Bidirectional Recurrent Neural Network (BiLSTM) model, Self-Attention Long Short-Term Memory (SAELSTM) model, and Random Forest (RF) model. The comparison among algorithms between the training and test sets can be seen from

Table 2. Comparison of six algorithms in terms of MAE and RMSE.

Algorithm		MAE	RMSE
Training Set	ELM	0.1907	0.0577
	LSTM	0.1825	0.0517
	BP	0.1920	0.0570
	BiLSTM	0.1849	0.0020
	SAELSTM	0.1797	0.0501
	RF	0.1745	0.0504
Test Set	ELM	0.1868	0.0588
	LSTM	0.1865	0.0602
	BP	0.1878	0.0617
	BiLSTM	0.1894	0.0635
	SAELSTM	0.1909	0.0625
	RF	0.2584	0.0803

, which displays the results of the six trained models on both the training and test sets. As can be observed from Figure 6, the SAELSTM model exhibited high fitting accuracy on the training set, but its prediction accuracy was the lowest on the test set. Although the ELM model adopted in this study showed relatively lower fitting performance on the training set, its test set accuracy surpassed that of the other models.

4. Experimental Results

4.1. Experiment on Indoor Wave-Current Flume

The ice-based buoy with an imaging system in this study integrates a small wave energy generation device. To test the power generation performance of the wave energy device, the buoy is placed in a large-scale wave–current flume, and the output current, voltage, and power of the power generation device are collected and recorded under different wave environments. Due to the indoor testing environment, a test device was designed to receive the output data of the wave energy generation device (Figure 7). The experimental results indicate that the device can continuously generate voltage under wave excitation, and the overall trend is consistent with the theoretical expectations. This confirms that the basic function of the device has been validated and that it can provide fundamental power support for the buoy system.

After the wave energy device generates electricity, it is converted into direct current through the rectification module. After being amplified by the power resistor, the transmitter sends the collected voltage and current information to the controller. The data is then transmitted to the data logger through the Lora wireless transmission module, and the data can be stored on an SD card. The rectifier adopts the Optimal QLKBPC5010 single rectifier bridge stack (which is produced by Optimal Semiconductor (Shenzhen, China)). Without filtering, the output DC value is about 0.9 times the AC input value, and with filtering capacitors, the output DC value is about 1.2–1.4 times the AC input value. The buoy is fully enclosed and uses an induction switch to control the operation of the data acquisition. The switch adopts the M30 metal induction LJ30A-3153Z/AX NPN (produced by Chint (Wenzhou, China)), which is a normally closed inductive sensor. The induction switch is a position switch that can be operated without direct mechanical contact with moving parts. When an object approaches the sensing surface of the switch to the action distance, no mechanical contact or pressure is required to make the switch operate. The Modbus RTU communication protocol is developed in the data logger to directly transmit information to the data logger. Data can be recorded in the form of text or tables on the SD card. The electricity collection is completed through isolation, and the tested end is completely isolated from the system power supply without interfering with each other.

The wave–current flume used in this study is from the College of Oceanography, Zhejiang University. The wave–current flume can complete the basic theoretical research of ocean engineering and wave–current experiments and is an experimental carrier for studying the impact of wave–current comprehensive factors on ocean engineering equipment. The technical specifications of the flume are shown in

Table 3. Technical specifications of the flume.

Parameter	Performance
Length	75 m
Width	1.8 m
Height	2 m
Maximum test water depth	1.5 m
Maximum flow rate	0.8 m3/s

.

The experiment on the indoor wave–current flume is divided into two parts. The power generation performance of the wave–energy device can be tested at standard wave heights of 0.1–0.3 m and wave periods of 1–2 s, with each change in wave height or period. A wave height of 0.2 m and a period of 1.5 s were selected for a 30 min power generation experiment. The experimental setup and process are shown in Figure 8.

The results show that although posture changes introduce short-term fluctuations in the output, the power supply remains stable without interruption. This demonstrates that the device has a certain degree of adaptability and can continuously support the buoy under varying operating conditions.

As shown in Figure 9, the output power remains within a relatively stable range, despite fluctuations under different wave conditions. The average power level is sufficient to meet the operational demand of the buoy imaging system, which further verifies the practical energy-supply capability of the device.

4.2. Image Acquisition Experiment in the Arctic Ocean

The ice-based buoy with an imaging system was deployed in the Arctic Ocean on 7 September 2023. The objective of this experiment is to evaluate the performance of the underwater image system, and the ice-based buoy was equipped with dedicated batteries as the power supply. All data were transmitted back via the Iridium satellite system. After passing laboratory waterproof testing, the underwater camera demonstrated stable performance during three months of Arctic deployment, successfully withstanding the challenges of low-temperature seawater. As shown in Figure 10, the buoy captured valuable images of ice-bottom morphology and biological features such as algal attachment, confirming the feasibility of under-ice observations. An unexpected yet important result was the successful capture of a fish within the field of view, in addition to consistent imaging under low-light conditions. These outcomes show that the system not only operated reliably but also provided insights into biological activity beneath the ice. At the same time, the limited resolution of the current camera restricted detailed species identification, which points to a clear direction for improvement. Overall, the deployment validated the robustness of the design and highlighted priorities for future work, including higher-resolution imaging, extended deployment duration, and the integration of additional sensors to support more comprehensive polar observations.

5. Conclusions and Future Work

This study proposed and validated an ice-based buoy integrating imaging and energy management. The main findings are as follows:

5.1. Experimental and Simulation Results

• Laboratory wave-tank tests demonstrated that the buoy can operate stably under mild wave conditions, with both the imaging system and the energy-harvesting module showing reliable performance;
• A three-month Arctic field deployment further confirmed the feasibility of the buoy in extreme low-temperature environments;
• Numerical simulations exhibited consistent trends with the experimental observations, verifying buoy stability and energy-harvesting performance under different wave conditions.

5.2. Limitations

• The laboratory tests were conducted under mild wave conditions and cannot fully represent Arctic storm environments;
• The imaging resolution was insufficient for detailed ecological classification under ice;
• Battery lifetime and energy management strategies have not yet been validated on a year-round scale.

5.3. Future Improvements

• Conduct tests under more energetic wave conditions (e.g., in large-scale facilities or offshore test sites) to further assess performance in extreme environments;
• Enhance the imaging system with higher resolution and better low-light performance;
• Extend the duration of field deployments and adopt more advanced energy management and prediction algorithms to support year-round continuous operation.

In summary, both experiments and simulations confirm the buoy’s feasibility for long-term polar observations, but further validation and optimization are required under broader environmental conditions.