Design and Implementation of an Ice-Tethered Observation System for Melt Pond Evolution with Vision and Temperature Profile Measurements

Abstract

Melt pond is one of the most significant and important features of Arctic sea ice in the summer and can dramatically reduce the albedo of ice, promoting more energy into the upper ocean. The observation of the seasonal evolution of melt pond can improve our fundamental understanding of the role and sensitivity of sea ice in the context of global climate change. In this study, an ice-tethered observation system is developed for melt pond evolution with vision and temperature profile measurements. The system composition, structure of the ice-tethered buoy, freeze-resistant camera, and thermistor chain are analyzed. A sealed shell and electric heating wires are used to increase the temperature to around the camera in low-temperature environments. The ice thickness and depth of melt pond can be inverted using a specific interface recognition algorithm. A low-light image enhancement strategy is proposed to improve the quality of images under the low lighting conditions in polar regions. The proposed system was tested in the second reservoir of Fen River, Yellow River, from 15 January to 27 January 2021. An artificial freshwater pond was used as the location for thermistor chain deployment and observation. The differences in mean square error (MSE), peak signal-to-noise ratio (PSNR), and feature similarity index (FSIM) between the original and enhanced images indicate that the proposed algorithm is suitable for low-light image enhancement. The research on the ice-tethered observation system will provide a new framework and technical support for the seasonal observation for melt pond.

1. Introduction

Arctic sea ice has been a sensitive indicator of climate change and has declined dramatically in recent decades [1,2,3]. Melt pond can be considered as one of the most significant and important features of Arctic sea ice in the summer [4,5,6,7,8]. The positive net surface energy balance of the sea ice results in the surface melting on Arctic sea ice, which starts from late spring, and some of the meltwater is retained on the ice surface. The continuous absorption of solar radiation leads to severe localized melting and meltwater area expanding both horizontally and vertically, which eventually forms melt ponds. Melt pond can dramatically reduce the ice albedo and promote more energy into the upper ocean [9,10,11].

The evolution of a melt pond can be divided into four stages, based on the results of observations and pond behavior and control mechanisms [12]. In the first stage, pond coverage rapidly increases and meltwater accumulates on the surface. During the second stage, horizontal transport dominates the process of water loss, and percolation through the ice is also one of the significant causes. High ice permeability and open flaws cause the meltwater to be freely lost to the ocean, with some melt ponds even directly melting through to the ocean during the third stage. The fourth stage is defined as a refreezing period, which is not restricted to the end of the ice season. In this stage, the freezing of the melt ponds can stop meltwater transportation. Numerical modeling is the most widely used and most effective research method to depict the features of sea ice conditions in the Polar Regions. The seasonal evolution of melt ponds is a predominant control of albedo and solar energy partitioning [13], and can be used as an important variable in melt pond modeling [14,15,16,17,18], as the accuracy of sea ice numerical models depends largely on the assimilation of the results of long-term in situ observation.

The observation of the evolution process of the melt pond mainly includes the melt pond coverage, melt pond depth, albedo, melt pond shape and size, and melt pond temperature. Field observation and remote sensing can usually be used to study the morphological changes and physical properties of melt ponds [19,20,21]. In situ observations, aerial surveys and synthetic aperture radar were used together to evaluate the physical, radiative, and electrical properties of the melt ponds of first-year ice and multiyear ice [22]. Under-ice diving observations from 8 June 2008 to 11 June 2008 provided results that melt pond size and shape distributions may affect the regional transmittance of sea ice [9]. Aircraft videography can be used to reveal sea ice surface types, including melt ponds of different colors [23]. Observations of pond coverage, albedo, and ice properties and terrestrial lidar measurements were made on landfast Arctic sea ice to quantitatively identify the mechanisms of pond coverage [5]. A terrestrial laser scanner, snow and sea ice sampling, and surface meteorological characterization were used to study the evolution of melt pond coverage, surface topography, and mass balance on landfast sea ice, estimating the contribution of surface melting to the net radiative flux [24]. Ice mass balance buoy (IMBs) and the Scottish Association of Marine Science (SAMS) sea ice mass balance for the Arctic (SIMBA) have been widely deployed in the Arctic Ocean to automatically observe the sea ice mass balance [25,26,27,28]. The ice mass balance buoy with radiometers (IMB-R) were deployed in a freshwater pond and a saline pond, respectively, revealing the salinity control of thermal evolution of melt ponds [29]. For the first time, continuous observation by ice mass balance instruments provided the formation of a deep melt pond and subsequent false bottom evolution [30]. A two-stream radiative transfer model was proposed and applied to ponded sea ice in the Arctic to examine the upwelling irradiance from the pond surface and provided a potential method to obtain ice thickness through melt pond color [31]. The “LinearPolar” algorithm was proposed based on the polar coordinate transformation and the proposed algorithm had higher accuracy and precision [32].

In these contexts, this paper focused on designing an ice-tethered observation system for melt pond evolution with vision and temperature profile measurements used in the Polar Regions. Section 2 describes the system composition, design of ice-tethered buoy, freeze-resistant camera, and the thermistor chain. Section 3 gives research results on the low-light image enhancement method in the Arctic. In Section 4, the results of field experiments in the Yellow River enable us to evaluate the performance of the thermistor chain, and the effectiveness of the low-light image enhancement algorithm is compared and evaluated. The conclusions and future works are present in final section.

2. Description of Ice-Tethered Observation System for Melt Pond Evolution

2.1. System Composition

The ice-tethered observation system for melt pond evolution proposed in this study is designed for the observation of the seasonal horizontal and vertical evolution of the melt pond in the Arctic. The composition of the proposed system is shown in Figure 1. It integrates several instruments for meteorological, snow, sea ice, melt pond, and hydrological observations, including air temperature, relative humidity, barometric pressure, ultrasonic range finder, up-looking sonar, freeze-resistant camera, thermistor chain, and CT (conductivity and temperature). Observations of the air temperature, relative humidity, and barometric pressure can provide basic meteorological data above the sea ice surface. The ultrasonic range finder can be installed at 1.2 m above the snow/ice surface and measure the distance between the ultrasonic sensor and surface. The up-looking sonar can work in seawater and measure the distance between the sonar and the ice bottom. The freeze-resistant camera, which has the ability to work in low-temperature environments in the Arctic, should be installed on the bracket and aimed at the melt pond in order to capture the surface morphology evolution process of melt pond. A freeze-resistant camera also can provide visual evidence, such as snowfall, ice breaking, or polar bears. The thermistor chain can be installed through a support structure and deployed in the ice through an ice hole and measure a temperature profile from the air to the ocean. CT can achieve the observation of water temperature and salinity. The overall system also includes three modules in addition to instruments: the ice-tethered buoy, the power supply module, and the controller and data logger. The ice-tethered plays the roles of physical carrier of the system and contains the power supply module, the controller and data logger, waterproof interfaces, and buoyancy block. The power supply module consists of low-temperature batteries and an energy management subsystem. The controller and data logger can realize data collection and remote transmission.

2.2. Design of Ice-Tethered Observation System for Melt Pond Evolution

2.2.1. Ice-Tethered Buoy

As shown in Figure 1, the structure of the ice-tethered buoy consists of a main floating body and a bracket. A sealed electronic cabin is installed inside the main buoy, which contains a power supply module, a controller, and a data logger. The bracket on the main buoy is designed to fix the GPS, iridium antenna, air temperature, relative humidity, and barometric pressure sensors. The waterproof joints are attached to the upper cover of the main buoy and connected to the antenna and meteorological sensors. The floating body is made of glass beads. The bottom diameter of the main buoy is 0.25 m, which the main buoy can be easily deployed in sea ice through ice drilling. The upper cover of the main buoy is 1.5 m and can provide enough space for the connectors. The weight of the buoy is 50 kg.

The ultrasonic range finder and up-looking sonar are installed on a 5 m PPR (polypropylene-random) rod. The ultrasonic range finder is installed on a 1 m bar. The up-looking sonar is fixed by a stainless steel case. A thermistor chain and a CT can be deployed in a melt pond. The cables of the thermistor chain and CT are connected to the main buoy. The controller and data logger are designed by Taiyuan University of Technology and in charge of obtaining meteorological data and sea ice temperature profile data, data transmission, and processing.

2.2.2. Freeze-Resistant Camera

The freeze-resistant camera in this observation system is designed based on the modification of a serial port camera (Figure 2). A sealed shell is used to install the camera. This serial port camera has the characteristics of a small size and integration convenience. It is integrated with a high-performance image processing chip, which can compress the original image and output it in standard JPEG image format. The serial port camera can achieve functions such as image capture, compression, and serial output. The commands for the serial port camera are simple that camera reset, shooting, baud rate setting, and image size selection can be achieved with a few instructions. The basic parameters of the serial port camera can be seen in

Table 1. Basic parameters of serial port camera.

Parameter	Performance
Circuit board size	32 mm × 32 mm
Image output format	JPEG
Scanning mode	Progressive scanning
Power supply voltage	5 V
Current	75 mA
Baud rate	115,200
Infrared supplementary light	Optional

.

To avoid the influence of low temperatures, electric heating wires are wrapped around the camera, which the electric heating wire generates heat after being loaded with direct current, increasing the temperature around the camera (Figure 3).

At −40 °C, the serial port camera is affected by low temperatures and cannot take photos properly. When heated to −20 °C, the camera restarted its shooting function, which proves that the heating measures can ensure that the camera operates properly in polar low-temperature environments, without the need to heat up to higher temperatures to generating high energy consumption (Figure 4). The freeze-resistant camera was tested in Arctic, as shown in Figure 4. The images captured by the camera clearly recorded the heating measures to ensure the normal shooting function under unmanned conditions.

The freeze-resistant camera integrates a DC-powered pan-tilt platform to realize automatic rotation. The pan-tilt platform can be used in conjunction with the camera to achieve large-scale monitoring. The pan-tilt platform needs to be configured with coordinates, lens focal length, focus and other parameters in advance, and can store the horizontal angle, vertical angle, and camera lens in the memory. These parameters can be quickly called from memory and we can easily and quickly control the pan-tilt platform. The “Watchman” function is designed to achieve the observation of the surface morphological changes of melt pond. This function means that when the standby time of the pan-tilt platform reaches the set value, it can automatically realize the position adjustment, cruise control, and scanning. The basic information of the pan-tilt platform can be seen in

Table 2. Basic information of the pan-tilt platform.

Parameter	Performance
Voltage	10 V–18 V (DC)
Power	20 W (Maximum)
Rotation angle	0°–360°
Communication mode	RS485
Overall dimension	191 mm × 169 mm × 280 mm
Baud rate	9600
Material	Stainless steel

.

2.2.3. Thermistor Chain

The design of the thermistor chain is illustrated in Figure 5. The circuit design of thermistor chain used master–slaves mode. The slaves complete the temperature measurement, data storage, and data output. The master can realize data calculation and data transmission to the bus. The function of the temperature observation of the thermistor chain is achieved through a bridge circuit, an ADC (Analog-to-digital converter) circuit, and a microcontroller. The analog-to-digital converter ADS1232 (Texas Instruments, Dallas, TX, USA) was used to achieve a high-precision temperature value conversion. The unbalanced bridge can be used to obtain a higher precision resistance value of the temperature measuring resistor R₀ (platinum resistance). MSP430G2553 (Texas Instruments, Dallas, TX, USA) and serial port TTL level to RS485 level converter MAX1483 (Maxim Integrated Products, Sunnyvale, CA, USA) are used together to implement bus construction.

As shown in Figure 3, the output of bridge ∆U(T), which is temperature-dependent, can be calculated by the difference between V_IN1 and V_IN2.

$$\Delta U\left(T\right)=V_{IN1}-V_{IN2}$$

(1)

The output of bridge ∆U(T) can also be described using the bridge measurement circuit and is dominated by R₀(T) and V_REF.

$$\Delta U\left(T\right)=\left[\frac{R_0\left(T\right)}{R_0\left(T\right)+R_a}-\frac{R_c}{R_b+R_c}\right]V_{REF}=f\left[R_0\left(T\right)\right]V_{REF}$$

(2)

$$V_{REF}=V_{REFP}-V_{REFN}$$

(3)

where R₀(T) is a high-precision temperature measurement resistor; R_a, R_b, and R_c are bridge arm resistance, which can be filtered out so that the resistance values are basically constant.

The ADC output R(T) can be described as follows.

$$\begin{array}{l}R\left(T\right)=\frac{1}{2}\left(2^{23}-1\right)Gain\frac{V_{IN1}-V_{IN2}}{V_{REFP}-V_{REFN}}=\frac{1}{2}\left(2^{23}-1\right)Gain\frac{\Delta U\left(T\right)}{V_{REF}}\\ ={\left.kf\left[R_0\left(T\right)\right]\right|}_{R_a,R_b,R_c}\end{array}$$

(4)

where Gain can be chosen as 64; the impact of V_REF has been offset by Equation (4) and the result of R(T) is related to the resistance value of R₀(T). It is simple to obtain the value of R(T), so the value of R₀(T) can be calculated.

The thermistor chain can directly measure the temperature of the surrounding medium (snow, sea ice, melt pond, or ocean). A heating mode is designed and set to easily identify different vertical interfaces, with the additions of the bridge circuit and other resistances that can be heated when outside voltages are applied around the temperature measuring resistors of the thermistor chain.

Due to the temperature chain being deployed in melt pond, interface recognition can be divided into two parts: surface snow and melt pond. We assumed that there were N points in the vertical temperature profile data TC_N and TC_N can be divided into TC_n and TC_h. TC_n can been seen as points in melt pond, sea ice, and ocean. TC_h can be seen as points in air, snow, and sea ice above the melt pond.

$$T{C}_N=T{C}_n+T{C}_h,N=n+h$$

(5)

The vertical temperature profile TC_n is described as follows:

$$T{C}_n=\left\{\begin{array}{l}t{c}_a\left(m\right)+e,1\le a<n_1\\ t{c}_b\left(si\right)+e,n_1\le b<n_2\\ t{c}_c\left(o\right)+e,n_2\le c\le n\end{array}\right.$$

(6)

where tc_a(m), tc_b(si), and tc_c(o) are the temperature values in the melt pond, sea ice, and ocean, respectively; n₁ is the location of the melt pond–sea ice interface; n₂ is the location of the ice–ocean interface; e represents the noise in the data of vertical temperature profile.

During the heating mode, the response of the temperature measuring resistors of the thermistor chain is closely related to the thermal conductivity of the surrounding medium. The vertical temperature profile after heating can be described as follows:

$${\left(TC+\Delta T\right)}_n=\left\{\begin{array}{l}t{c}_a\left(m\right)+\Delta T_a+e,1\le a<n_1\\ t{c}_b\left(si\right)+\Delta T_b+e,n_1\le b<n_2\\ t{c}_c\left(o\right)+\Delta T_c+e,n_2\le c\le n\end{array}\right.$$

(7)

The increment in temperature in different media can be approximately expressed as follows:

$$\Delta T_{media}=\frac{Q}{4\pi k_{media}}\ln \frac{t_{end}}{t_{start}}$$

(8)

where Q represents amount of heat; k_media is thermal conductivity of the medium; t_start and t_end are the start and end times of heating, respectively.

For the vertical temperature profile after heating, the location of the top interface (melt pond–sea ice interface) and bottom interface can be estimated using the least-squares estimation method. The LOI_top and LOI_bottom can be expressed as follows:

$$\underset{1\le k\le s}{LO{I}_{top}}=arg\underset{1\le k\le s}{\min }\left\{\begin{array}{l}{\displaystyle \sum _{i=1}^{k-1}}{\left[{\left(TC+\Delta T\right)}_i-\frac{1}{k-1}{\displaystyle \sum _{i=1}^{k-1}}{\left(TC+\Delta T\right)}_i\right]}^2\\ +{\displaystyle \sum _{i=k}^s}{\left[{\left(TC+\Delta T\right)}_i-\frac{1}{s-k+1}{\displaystyle \sum _{i=k}^s}{\left(TC+\Delta T\right)}_i\right]}^2\end{array}\right\}$$

(9)

$$\underset{s+1\le p\le n}{LO{I}_{bottom}}=arg\underset{s+1\le p\le m}{\min }\left\{\begin{array}{l}{\displaystyle \sum _{i=s+1}^{p-1}}{\left[{\left(TC+\Delta T\right)}_i-\frac{1}{p-s+1}{\displaystyle \sum _{i=s+1}^p}{\left(TC+\Delta T\right)}_i\right]}^2\\ +{\displaystyle \sum _{i=p}^n}{\left[{\left(TC+\Delta T\right)}_i-\frac{1}{n-p+1}{\displaystyle \sum _{i=p+1}^n}{\left(TC+\Delta T\right)}_i\right]}^2\end{array}\right\}$$

(10)

$$\left\{\begin{array}{l}{E}_1\left(i\right)=\frac{1}{k-1}{\displaystyle \sum _{i=1}^{k-1}}{\left(TC+\Delta T\right)}_i\\ E_2\left(i\right)=\frac{1}{s-k+1}{\displaystyle \sum _{i=k}^s}{\left(TC+\Delta T\right)}_i\\ E_3\left(i\right)=\frac{1}{p-s+1}{\displaystyle \sum _{i=s+1}^p}{\left(TC+\Delta T\right)}_i\\ E_4\left(i\right)=\frac{1}{n-p+1}{\displaystyle \sum _{i=p+1}^n}{\left(TC+\Delta T\right)}_i\end{array}\right.$$

(11)

where E₁(i), E₂(i), E₃(i), and E₄(i) are segmented mean estimations.

The proposed interface recognition algorithm based on the vertical temperature profile after heating is briefly summarized in

Table 3. Scheme of the interface recognition algorithm based on the vertical temperature profile after heating.

1: Start Procedure
2: Initialization
Measure the vertical temperature profile in melt pond and sea ice.
Set time and activate heating mode.
Obtain the vertical temperature profile after heating.
3: Interface recognition
Use the vertical temperature profile after heating (TC + ∆T)n as input.
Calculate point d∈[1, n] using Equation (9).
Output (TC + ∆T)i=1,2,3…,d and (TC + ∆T)i=d+1,d+2,d+3…,n.
Use (TC + ∆T)i=1,2,3…,d as input and calculate point d1∈[1, d] using Equation (9).
Use (TC + ∆T)i=d1+1,d1+2,d1+3…,n as input and calculate point d2∈[d1 + 1, n] using Equation (10).
Output E1(i).
Use (TC + ∆T)i=d+1,d+2,d+3…,n as input and calculate d4∈[d + 1, n] using Equation (10).
Use (TC + ∆T)i=1,2,3…,d4 as input and calculate d3∈[1, d4] using Equation (9).
Output E2(i).
3: While E1(i) < E2(i) do
Output d1 and d2.
4: end while
Output d3 and d4.
5: End Procedure

.

3. Low-Light Image Enhancement in Arctic

The images of sea ice or melt pond cannot achieve the quality of images under ideal lighting conditions, and some substances in the air can also cause light attenuation and scattering, resulting in low brightness, low contrast, and severe noise in the collected images. Therefore, a low-light image enhancement algorithm in Arctic is proposed.

The sparrow search algorithm (SSA) is an intelligent search algorithm, which can simulate sparrow foraging behavior and antipredation behavior [33,34]. In this algorithm, some of the sparrows with better fitness can be used as producers and the rest are followers. A certain proportion of individuals in the population are selected for detection and early warning.

The location update formulas of the producers, followers, and alerter of the sparrow search algorithm are as follows:

The location update formula of the discoverers in SSA is described as follows [35]:

$$X_{i,j}^{t+1}=\left\{\begin{array}{l}{X}_{i,j}^t·\exp \left(\frac{-i}{\alpha ·T_{\max }}\right), R_2<S_T\\ X_{i,j}^t+Q·L,\hspace{1em}\hspace{1em} R_2\ge S_T\end{array}\right.$$

(12)

where X^t_i,j 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; L is the d-dimensional row vector, where every entry is 1.

The follower location update formula can be described as follows:

$$X_{i,j}^{t+1}=\left\{\begin{array}{l}Q·\exp \left(\frac{X_{worst}^t-X_{i,j}^t}{i^2}\right),\hspace{1em}\hspace{1em}\hspace{1em} i>\frac{N}{2}\\ X_P^{t+1}+\left|X_{i,j}^t-X_P^{t+1}\right|·A^{+}·L,\hspace{1em} i\le \frac{N}{2}\end{array}\right.$$

(13)

where X^t_i,j is the best position occupied by the producer; X^t_worst is the worst position; A is a 1 × D matrix; and each element is randomly assigned 1 or −1; and A⁺ = A^T(AA^T)⁻¹.

The position update formula for the investigator is described as follows:

$$X_{i,j}^{t+1}=\left\{\begin{array}{l}{X}_{best}^t+\beta ·\left|X_{i,j}^t-X_{best}^t\right|,\hspace{1em} f_i>f_g\\ X_{i,j}^t+K·\left(\frac{\left|X_{i,j}^t-X_{worst}^t\right|}{\left(f_i-f_w\right)+\sigma }\right),\hspace{1em}{f}_i=f_g\end{array}\right.$$

(14)

where X^t_best is the best global optimal location; β is a step size control parameter and satisfies the normal distribution random number, with the mean value of 0 and variance of 1. K is a random step size control parameter in the interval [−1, 1]; σ is a minimum constant to prevent the numerator being zero; f_i is the fitness value of the ith sparrow; f_g is the optimal fitness value of the current sparrow population; and f_w is the worst fitness value of the current sparrow species.

$$X_{i,j}^{t+1}=\left\{\begin{array}{l}{X}_{best}^t+\beta ·\left|X_{i,j}^t-X_{best}^t\right|, \hspace{1em}{f}_i>f_g\\ X_{i,j}^t+K·\left(\frac{\left|X_{i,j}^t-X_{worst}^t\right|}{\left(f_i-f_w\right)+\sigma }\right), \hspace{1em}{f}_i=f_g\end{array}\right.$$

(15)

Considering the characteristics of the Arctic, bilateral gamma correction is used to enhance the low brightness areas while suppressing the local bright areas of low light images. Bilateral gamma adjustment can be expressed as follows [36]:

$$\left\{\begin{array}{l}G\left(x\right)=\alpha ·G_1\left(x\right)+\beta ·G_2\left(x\right)\\ \alpha +\beta =1\end{array}\right.$$

(16)

The result of Equation (16) can be calculated by two gamma functions G₁(x) and G₂(x). α and β are two variables, the values of which can be found in [0, 1]. Assuming that the gray value range of the image is normalized to the range of [0, 1]. The two gamma functions can be described as follows:

$$\left\{\begin{array}{l}{G}_1\left(x\right)=\sqrt[\gamma ]{x}\\ G_2\left(x\right)=1-\sqrt[\gamma ]{1-x}\end{array}\right.$$

(17)

where x is the gray value of image; γ is an adjustable variable to adjust the degree of image enhancement; G₁(x) is a convex function for enhancing dark areas; G₂(x) is a concave function for suppressing bright areas of the image.

$$\left\{\begin{array}{l}{I}^{\prime }\left(x,y\right)=\raisebox{1ex}{$I\left(x,y\right)$}\!\left/ \!\raisebox{-1ex}{$256$}\right.\\ I^{″}\left(x,y\right)=G\left[I^{\prime }\left(x,y\right)\right]\end{array}\right.$$

(18)

where I(x, y) is the acquired image matrix; (x, y) represents the pixel cells of the image matrix; I′(x, y) is the results after being carried out progress of normalization; I″(x, y) is the output corrected by the bilateral gamma adjustment function.

The range of the gray value after the bilateral gamma adjustment function can be adjusted to [0, 255] as follows:

$$I^{‴}\left(x,y\right)=I^{″}\left(x,y\right)·256$$

(19)

In this study, the optimization strategy was determined based on a trade-off between three aspects. Therefore, the objective function is the maximum sum of E_image, S_edge, and STV. Figure 6 shows the schematic diagram of the proposed strategy for a low-light image enhancement in the Arctic.

The objective function is formulated as follows:

$$F_{obj}=\max \left(\phi ·E_{image}+\eta ·S_{edge}+\mu ·STV\right)$$

(20)

where φ, η, and μ are the weighting factors and the sum of them should be equal to 1. E_image is the entropy value of the enhanced image; S_edge is the edge content of the enhanced image; STV is the gray standard variance of the enhanced image.

The entropy value of the image E_image indicates the information in the image and can be expressed as follows:

$$E_{image}=-{\displaystyle \sum _{i=0}^{k}P\left(i\right)·{\log }_2\left[P\left(i\right)\right]}$$

(21)

where P(i) is the probability of the ith gray value in the enhanced image.

S_edge can be calculated using the ratio of the sum of non-zero pixels and the total number of pixels in the image.

$$S_{edge}=\frac{{\displaystyle \sum count\left(P_{non-zero}\right)}}{{\displaystyle \sum count\left(P_{total}\right)}}$$

(22)

4. Experimental Results

4.1. Temperature Observation in the Freshwater Pond in the Second Reservoir of Fen River, Yellow River

The thermistor chain of the ice-tethered observation system for melt pond evolution was deployed in the second reservoir area of Fen River, Yellow River, on 15 January 2021. The thermistor chain used in the field experiment is 1.2 m long assembled 60 temperature sensors. All temperature sensors were soldered on FPCBs. Hot wires were selected to measure ice thickness, which can be regarded as an effective method for measuring ice thickness [4]. The accuracy of the thermistor chain is 0.01 °C (Figure 7). Before the field experiment, the whole thermistor chain was put into the GDJS-series low temperature test chamber over a range of −50–50 °C. The temperature sensors of the thermistor chain were immersed in antifreeze and a high-precision resistance temperature detector (accuracy 0.001 °C) was installed in a same container. The results of measured temperature showed that the discrepancy between actual temperatures and temperature observed by the thermistor chain remain within 0.01 °C. The thermistor chain and hot wire were deployed and installed in the frozen river ice, which penetrated vertically through the air, ice, and water.

In order to test and evaluate the ability of the temperature chain to invert the internal structure of the melt pond, an artificial freshwater pond was used as the location for deployment. As shown in Figure 8b,c, we obtained a space of 1 m × 0.7 m × 0.3 m using drilling and cutting methods and freshwater filled this space.

Due to seasonal reasons, the surface of the freshwater pond quickly froze and gradually grew downwards. The thermistor chain can be deployed through a small ice hole which quickly froze in a few hours. Hot wires need smaller ice holes that were considered to have a minimal impact on ice growth and decay. The new ice thickness, freshwater pond depth, and ice thickness under the pond can be seen from Figure 9.

The field experiment lasted from 15 January to 27 January. We selected data from 21 January to 27 January for analysis, as these included the observations of the growth of new ice. Hot wires can be used to obtain the ice thickness four times a day. The thermistor chain can realize temperature data collection every one hour. The initial ice thickness was 0.4 m, for which the freshwater pond was not prepared. The new ice thickness was 5.8 cm on 21 January. At the end of the field experiment, the new ice thickness was 14.2 cm and the average growth rate of the new ice was 1.2 cm/d. Although the ice hole froze, the bottom of the freshwater pond did not show significant changes during the experimental period. Therefore, it was easy to observe the depth changes of the freshwater pond through the thermistor chain and hot wire. The depth of the pond decreased from 24.2 cm to 15.8 cm. The ice thickness under the pond showed an upward trend due to the decrease in winter temperature, and ice thickness under the pond increased from 13.5 cm to 24.2 cm. The average growth rate of ice thickness was about 1.1 cm/d.

4.2. Analysis of Image Enhancement

For a comparative analysis, the processing results of a single-scale retinex (SSR), multiscale retinex (MSR), multiscale retinex with a color recovery factor C (MSRCR), contrast limited adaptive histogram equalization (CLAHE), and the proposed algorithm are shown in Figure 10, which includes the original images and enhanced images. Four original images were selected for comparative analysis. Two low-light images in a conventional outdoor environment and two low-light images taken in polar region were selected to evaluate the performance of the algorithms. The scene in low-light image 1 depicts dusk and is backlit with dark paths and trees by the lake, which is not conducive to observing the details of the image. Low-light image 2 was captured in extremely low light conditions at night, and most areas of the image cannot be recognized. Low-light images 3 and 4 were both captured in the polar region, for which the overall images are relatively dark and not conducive to image segmentation and recognition processing in the later stage. Therefore, it is necessary to improve clarity and brightness for the images.

Figure 11 shows gray-scale images after image enhancement.

The mean square error (MSE) between the original image and the enhanced image can be regarded as a relatively common quality evaluation method. The MSE can be calculated as follows:

$$\mathbf{MSE}=\frac{1}{MN}{{\displaystyle \sum _{i=1}^M{\displaystyle \sum _{j=1}^N\left|R\left(i,j\right)-F\left(i,j\right)\right|}}}^2$$

(23)

where M is the image height; N is the image width; R(i, j) is the gray value of the pixels in the original picture; and F(i, j) is the gray value of the pixels in the enhanced image.

As shown in

Table 4. Comparison of four algorithms in terms of MSE.

Algorithm	SSR	MSR	MSRCR	CLAHE	Proposed Algorithm
Image 1	1.5476	14.1193	7.1250	0.0014	0.0006
Image 2	0.0091	0.6312	0.0061	0.0016	0.0017
Image 3	4.5851	0.3071	5.62330	15.4081	0.0042
Image 4	11.5233	0.8400	4.40380	23.1143	0.0155

, the MSE of low-light images processed by the proposed algorithm in this study remain at a relatively small level, which can indicate that the difference between the processed image and the original image is the smallest.

The MSE can be transformed to obtain the peak signal-to-noise ratio (PSNR), the latter being the most commonly used objective metric for evaluating image quality [37]. The larger the PSNR value between the two images, the more similar the images are. The formula of PSNR is as follows:

$$\mathbf{PSNR}=10{\log }_{10}\left[\frac{{\left(2^n-1\right)}^2}{\mathbf{MSE}}\right]$$

(24)

It can be seen from

Table 5. Comparison of the four algorithms in terms of PSNR.

Algorithm	SSR	MSR	MSRCR	CLAHE	Proposed Algorithm
Image 1	15.4172	5.8157	8.7860	45.9364	49.8883
Image 2	35.3989	16.9699	37.0965	42.8546	42.6372
Image 3	12.6805	24.4210	11.7941	7.4165	43.0946
Image 4	8.6784	20.0514	12.8559	5.6554	37.3879

that the PSNR of the proposed algorithm has a stronger ability to suppress noise, and indicates that, while suppressing noise, it preserves the details of the image while effectively enhancing the low-light images.

The feature similarity index (FSIM) is used to calculate the feature similarity between the original image and the enhanced image [38]. The higher the FSIM value, the more similar the reference image and the image to be measured are. The formula of FSIM can be described as follows and a detailed description of FSIM is provided in Appendix A:

$$\mathbf{FSIM}=\frac{{\displaystyle {\sum }_{x\in \Omega }{S}_L\left(x\right)·P{C}_m\left(x\right)}}{{\displaystyle {\sum }_{x\in \Omega }P{C}_m\left(x\right)}}$$

(25)

It can be seen that the proposed algorithm resulted in a higher FSIM and greater similarity in the enhancement of images. The distortion of the image can be regarded as not significant after being processed by the algorithm proposed in this article, and the details of the original image are well preserved at the same time. See

Table 6. Comparison of four algorithms in terms of FSIM.

Algorithm	SSR	MSR	MSRCR	CLAHE	Proposed Algorithm
Image 1	0.7138	0.6586	0.8492	0.7798	0.9051
Image 2	0.5376	0.4623	0.7096	0.5262	0.8264
Image 3	0.3525	0.8278	0.9133	0.4027	0.9323
Image 4	0.4024	0.8744	0.9120	0.3697	0.9341

.

5. Conclusions and Future Work

In this study, an ice-tethered observation system for melt pond evolution was designed and presented. This system includes meteorological instruments, as well as instruments for observation for the ice mass balance, the visual observation of the melt pond, and the thermodynamic observation of melt pond. A power supply module, a controller, and a data logger are installed inside the main buoy of the ice-tethered buoy. In order to improve the survival ability of the system, the main buoy and floating body are small and can be easily deployed in sea ice through ice drilling. The morphological changes of the melt pond and sea ice can be studied using the freeze-resistant camera in this observation system. For the seasonal observation of the melt pond in the Arctic, a serial port camera is installed in a designed sealed shell and electric heating wires are wrapped around the camera, increasing the temperature to around the camera to reduce the impact of low temperatures. A DC-powered pan-tilt platform is used to realize the automatic rotation for observation of changes in surface morphology of sea ice and melt pond within a certain range. The design of the thermistor chain is based on the master–slaves mode and the construction of the data transmission bus. The function of the temperature observation of thermistor chain is achieved through a bridge circuit and high-precision temperature measurement sensors realize temperature observation, which can reduce the error of temperature inversion. An interface recognition algorithm designed specifically for the thermistor chain relies on the vertical temperature profile after heating and can improve the efficiency of the inversion of the internal structure of melt ponds. A low-light image enhancement strategy is proposed and applied based on the sparrow search algorithm and bilateral gamma correction. Field experiments were completed and the thermistor chain can be proved high precision and high stability in the temperature profile in the second reservoir of Fen River, Yellow River. During the field experiment on ice, an artificial freshwater pond of 1 m × 0.7 m × 0.3 m was built using drilling and cutting methods. The observation results of new ice thickness, pond depth, and ice thickness under the freshwater pond were evaluated and analyzed. The results showed that the average growth rates of new ice thickness and ice thickness under the freshwater pond were 1.2 cm/d and 1.1 cm/d, respectively. The pond depth decreased from 24.2 cm to 15.8 cm from 21 January to 27 January. For the low-light images, the processing results of the proposed visual enhancement algorithm, single-scale retinex (SSR), multiscale retinex (MSR), multiscale retinex with a color recovery factor C (MSRCR), and contrast-limited adaptive histogram equalization (CLAHE) were assessed and verified. The mean square error (MSE), peak signal-to-noise ratio (PSNR) and feature similarity index (FSIM) between the original and enhanced images indicate that the proposed algorithm is suitable for low-light image enhancement in the Arctic based on comparative analysis of image quality, ability to suppress noise and feature similarity between the processed images and the original images.

More design optimization can be realized to overcome potential risks in Arctic observations. In future work, the application of renewable energy in polar regions is a key point, which can provide guarantees for the seasonal observation of melt pond. More observation plans for the different types of melt ponds should be considered and more sensors with different parameters, such as chlorophyll and dissolved oxygen can be assembled into the system to observe the impact of melt pond evolution on the marine environment.

Obtaining meaningful information by extracting specific features from images is still a challenge [38]. Images obtained from ice-tethered observation system for melt pond evolution can provide a wealth of information, requiring feature extraction detection and recognition. Entropy plays a pivotal role in image processing. A method for establishing an automatic and efficient image retrieval system is designed and can improve the retrieval ability and accuracy of feature extraction [39]. Image encryption is necessary to protect the digital image transmission and image data security should be considered [40].