During the past few decades, the Arctic sea ice has undergone tremendous changes, which are manifest in such trends as shrinkage of ice extent, decrease of ice thickness, replacement of multi-year ice with younger first-year ice, decrease of ice concentration, and so on (Comiso et al., 2008; Maslanik et al., 2007, 2011; Stroeve et al., 2012). These changes have a significant influence on mass and energy partitioning in the Arctic atmosphere-ice-ocean system, marine environment and ecosystem, and on global climate (Palmer et al., 2014; Perovich et al., 2011; Serreze et al., 2007). Much effort has been devoted to understanding the observed changes in which the conflation of thermodynamic and dynamic processes was thought to be the primary driver source (Serreze et al., 2007). As one of the most important thermodynamic processes, solar radiation plays a significant role in governing the formation and retreat of sea ice, the distribution of which in the sea ice environment attracts numerous attentions (Ehn et al., 2011, 2008; Grenfell et al., 2006).

Until now, measurement of solar radiation in the Arctic sea ice environment predominantly focused on that above the sea ice, that is, the incident and the reflected solar radiation were simultaneously collected, to calculate the surface albedo of the sea ice (Feister and Grewe, 1995; Perovich et al., 2002; Pirazzini et al., 2006). There are also some studies which obtain the incident solar radiation on the ice surface, the transmitted solar radiation in the underlying ocean, and derive the transmittance of the sea ice based on these measurements (Frey et al., 2011; Nicolaus and Katlein, 2013; Nicolaus et al., 2010; Lei et al., 2012). However, detailed information about how the solar radiation is distributed as a function of depth of sea ice is very limited (Ehn et al., 2008; Light et al., 2008). Because most field studies are conducted manually during the months when the ambient temperature is not too cold (e.g., warmer than –10°C), commercially available oceanographic radiometers are adopted to measure the solar radiation. Among these, the SE590 spectroradiometer (Spectron Engineering, Inc., USA), ASD Ice-1 dual channel spectroradiometer (Analytical Spectral Devices, Inc., USA), and RAMSES-ACC-VIS hyperspectral radiometer (TriOS Optical Sensors, Germany), are widely used, have a spectral range covering the visible to the near-infrared (NIR) (typically 400–1 000 nm), and can meet the requirements of scientific studies (Light et al., 2008; Campbell et al., 2015; Lei et al., 2011). However, these devices are generally large, have high power consumption, and are not suitable for applications where supplemental power availability is limited.

In view of the limited data about the sub-ice-surface solar radiation distribution, a fiber optic spectrometry system was proposed previously to acquire long-term in-situ measurements of solar irradiance at different sea ice depths, by embedding multiple optical fiber probes at the desired depths to collect and transmit solar radiation signals, and using one spectrometer detect the signals sequentially (Wang et al., 2014). The overall performance of the system is evaluated in the laboratory and the results demonstrate that it has adequate performance to warrant further refinement for application in the field. However, the spectral range of this system is determined by the spectrometer adopted (C11009MA, Hamamatsu Photonics K.K., Japan), which only covers the range 340–780 nm. This spectral range might meet the requirement of investigating biological processes that occur within and below the ice, but it is too narrow to evaluate the heat and energy balance in the atmosphere-ice-ocean system where a wavelength of at least 1 000 nm is required (Frey et al., 2011). As a consequence, the application scope of the system is limited.

In this study, a miniature NIR spectrometer was selected to function as an extension to the existing visible spectrometer to increase the spectral range of the spectrometry system to 1000 nm, and assessed. The temperature dependence and integration time dependence of the spectrometer were characterized over the entire target operating temperature range of –50°C to 30°C. The performance of the spectrometer, including dark output, linear operating region, and signal output, were evaluated. These evaluations were further used to develop a dark output prediction model and two signal output correction models to correct temperature-induced biases when using the spectrometer for the Arctic sea ice. The overall performance of the spectrometer was evaluated for sea ice in the Arctic during the 9th Chinese National Arctic Research Expedition. The approach adopted in this study also provides a general framework for assessing the feasibility of using a specified spectrometer for applications apart from measuring solar radiation in cold environments.

Special concerns arise when using spectrometers in Arctic sea ice environments. First, the temperature can easily reach –40°C in the Arctic winter, which is generally far below the minimum operating temperature of most spectrometers. Second, the temperature can vary over a broad range in different seasons in the Arctic (i.e., annual temperatures can vary between approximately –45°C in February and approximately 20°C in August in recent years). In addition, the relative humidity of the air in the Arctic is very high (i.e., the average annual rate can reach to 85%, and in summer it can reach to 100%) (Treffeisen et al., 2007). The airborne water vapor may condense if the ambient temperature drops below the dew point, which might disturb normal operation of the optical and/or electrical parts of the spectrometers. Third, the incident solar radiation presents large temporal and spatial variations in different seasons and at different depths of the sea ice. Thus, the dynamic range of the spectrometer should be sufficiently wide to accommodate this broad variation. Finally, power consumption is a vital concern for autonomous systems running in the Arctic sea ice environment, where the availability of supplemental power is extremely limited.

Based on the above considerations, a low-power miniature spectrometer head (C11010MA, Hamamatsu Photonics K.K., Japan) is adopted in this study. The spectrometer is a compact module with a spectral range of 640–1 050 nm, and spectral resolution of 8 nm, covering the NIR region that we are interested in (i.e., 750–1 000 nm; Table 1). Light enters the spectrometer through a slit and is collimated onto a reflective grating where light of different wavelengths is separated spatially into a spectrum (Fig. 1). The spectrum is then focused onto a photodiode array image sensor to be converted into voltage signals by an internal charge amplifier which are sequentially output via a common VIDEO pin. With the exception of the SMA fiber coupling (i.e., a silica fiber with a core diameter of 600 μm) and electrical interface, all optical components and electronics are sealed inside a waterproof metal case to protect them from the potential contamination of condensation or frost. This configuration improves the reliability of the spectrometer especially in cold environments. The module has two different gain settings (high and low gain) which provide an additional degree of sensitivity control (a factor of approximately 5). The selectable gain setting facilitates the measurement of solar radiation presenting a broad intensity variation. Compared with the relatively complicated electrical interface (e.g., universal serial bus (USB)) provided by most spectrometers, this spectrometer employs a simple interface, sequentially outputting spectral measurements corresponding to each of its 256 pixels when provided with start and clock signals. As a result, most microcontrollers equipped with an accurate analog to digital convertor (ADC) can be used to control the operations of the module.

An 8-bit low-power microcontroller (ATxmega128A1U, Atmel Corporation, USA) is adopted to govern the operation of the spectrometer module and the relevant auxiliaries (Fig. 2). The primary motivation for choosing this chip is the abundance of onboard peripherals which greatly simply the design and integration of the various subsystems described below. The spectrometer prepares for the transmission of one group of spectral measurements upon receiving the rising edge at pin ST, and sequentially outputs each of the 256 voltage signals corresponding to the 256 photodiodes comprising the image sensor via pin VIDEO upon each rising edge of the pin CLK (Table 2). After the voltage signal for the last photodiode is sent, pin EOS of the spectrometer will be pulled low to indicate the completion of the data transmission. A 16-bit ADC (AD7988-5, Analog Devices Inc., USA) is used to digitize the analog output of the spectrometer. The reference voltage for the ADC is generated by an ultralow noise voltage reference chip (ADR435B, Analog Devices Inc., USA) with a nominal voltage of 5 V, which is wide enough to cover the entire output range of the spectrometer since the maximum output voltage of the spectrometer is approximately 4.4 V. To minimize temperature-induced biases in integration time of the spectrometer (i.e., the time interval between two adjacent ST pulses) over the entire operating temperature of –50°C to 30°C, a temperature compensated crystal oscillator (TCXO) is attached to the microcontroller to function as the system clock.

In addition, readout temperature of the spectrometer as indicated by an onboard 12-bit temperature sensor (DS1775R, Maxim Integrated Products, USA) is accomplished through a two-wire interface (TWI) of the microcontroller. Storage of spectral measurements to an external flash memory is realized via a serial peripheral interface (SPI), and one universal synchronous and asynchronous serial receiver and transmitter (USART) is reserved for system debugging. The entire system is powered by a 12 V battery pack, and peak power consumption is approximately 0.3 W when the spectrometer is in operation. To conserve power, sleep mode is activated during inactive times.

As stated above, temperature is a vital concern when using a spectrometer in the cold polar environment, since temperatures in the Arctic winter (e.g., –40°C) are far below the minimum operating temperature of the spectrometer (i.e., 5°C) specified by the manufacturer. Further, this extremely low ambient temperature, is also very close to the minimum operating temperature of most electronic components. Temperature-induced mechanical distortion or stretching of the optical bench, or performance variation of the module’s electronics might alter the performance of the spectrometer. However, due to the limited availability of electrical power (i.e., battery pack), temperature stabilization of the spectrometer is not feasible during field deployment. Therefore, careful evaluation of the temperature and integration time dependence of the spectrometer are essential for use of any spectrometer in such cold environments, especially over seasonal time scales. It is noteworthy to mention that the output of the C11010MA spectrometer module is composed of two combined components of signal output and dark output. The former is generated by the incident light and determined by the intensity of the incident light and integration time. Whereas, the latter is the spectrometer output in the absence of any light and is composed of the dark voltage of the photodiode array and an offset voltage inherent to the internal charge amplifier. Thus, the characteristics of linear operating range, dark output, and signal output of the spectrometer are assessed in this study.

To quantify the temperature and integration time dependence of the spectrometer module, we placed it inside an environmental test chamber (GDS-500, Ayashilin, China) together with the relevant driver electronics, and made measurements under constant illumination, gradually lowering the temperature from 30°C to –50°C at an interval of 5°C (Fig. 3). This 80°C range of test temperatures should cover the annual temperature variation in the Arctic sea ice environment. A tungsten halogen bulb (6319, Newport Corporation, USA) was used as the light source. It was placed at room temperature (i.e., approximately 25°C) and was powered by a stabilized DC power supply, to ensure stability. The emitted light was split by a beam splitter, one beam was guided to the spectrometer via an optical fiber, the other beam was guided to a photodiode (DET100A/M, Thorlabs, Inc., USA) to monitor for intensity fluctuations of the light source during measurements. At each measurement temperature, we waited approximately 30 min for the spectrometer temperature to stabilize, as indicated by the onboard temperature sensor. We then measured spectrometer output over 15 different integration times ranging from 0.001 s to 16 s. The primary consideration for choosing such an integration time range is to make the spectrometer output fall in the linear and non-linear operating region for the shorter and longer integration time, respectively. The measurement was carried out 20 times for each integration time, to minimize the statistical uncertainty. Meanwhile, we also measured the dark output of the spectrometer at each of the above test temperatures for 17 integration time lengths ranging from 0.001 s to 30 s (including the 15 integration time lengths adopted above), by blocking the emitted light from entering the spectrometer via inserting an opaque board in the filter holder (Fig. 3).

To testify the overall performance of the C11010MA miniature spectrometer module, it was integrated into a refined fiber optic spectrometry system together with the relevant driver electronics and measured the solar irradiance distribution within the ice cover with an ice thickness of 1.75 m and snow thickness of 0.1 m during the 9th Chinese National Arctic Research Expedition. The solar radiation signals at 8 different levels of the sea ice were collected and transmitted by 8 fiber probes included in the spectrometry system, respectively. The distribution of the fiber probes was shown in Fig. 14. Two probes were placed ~1.5 m above the snow surface to measure the downwelling and upwelling solar irradiance signals separately. Four probes were embedded within the ice cover at a vertical depth interval of 0.434 m, with the upmost probe located at a distance of 0.1 m below the snow surface. The remaining two probes were deployed 0.4 and 1 m below the ice bottom, respectively. The spectrometer was aligned to each of the fiber probe in sequence under the control of a rotary spectrometer switching device. Spectral sensitivity of each combination of the spectrometer and each of the fiber probe was determined by a simultaneous comparison study carried out using a radiometrically calibrated hyperspectral irradiance sensor (RAMSES-ACC-VIS, TriOS Mess- und Datentechnik GmbH, Germany) as a standard, at room temperature (i.e., 25°C).

To determine solar irradiance at different levels of the sea ice environment, we first corrected the influence of dark output in the spectral measurement by subtraction to get the signal output. The dark output was estimated using Eq. (5) with the same integration time as the corresponding spectral measurement and temperature of the spectrometer module as the inputs. We then converted the signal output to that at T_ref and t_ref by dividing with Eq. (13). We finally corrected spectral sensitivity of the spectrometer and fiber probe to get spectral intensity of the solar irradiance. The results denoted that general shape of the solar irradiance at different level of the sea ice was similar, and general shape of the solar irradiance above the snow surface measured by the spectrometer agreed well with that measured by other oceanographic spectroradiometers (Fig. 15a) (Nicolaus et al., 2010). However, spectral intensity of the solar irradiance decreased much faster at the near-infrared region than at the visible region, as the position is deeper in the ice (lines 3–6 in Fig. 15a) and in the underlying water (lines 7 and 8 in Fig. 15a). This characteristic ascribed to the fact that spectral absorption coefficient of the ice and water was much bigger at the near-infrared region than at the visible region (Grenfell and Perovich, 1981). When we calculated the spectral albedo of the sea ice using the upwelling irradiance to divide the downwelling irradiance above the snow surface (i.e., lines 1 and 2 in Fig. 15a), we found that spectral albedo of the sea ice generally decreased with longer wavelengths, from 0.80 at 640 nm to 0.46 at 950 nm (Fig. 15b). This characteristic basically agreed with spectral albedo of the sea ice covered with melting old snow (Riihelä et al., 2013). Spectral transmittance of the sea ice (i.e., 1.75 m sea ice covered with 0.1 m snow) was also calculated by using the downwelling irradiance below the ice bottom (line 7 in Fig. 15a) to divide that above the snow surface (line 1 in Fig. 15a). It indicated that spectral transmittance of the sea ice generally decreased with longer wavelengths with the biggest value of 0.021 at 640 nm. The transmittance was very close to 0 for wavelengths longer than 710 nm because there is almost no light within these bands transmitted through the sea ice and detected by the spectrometer (Fig. 15c).

In this study, the feasibility of employing a miniature near-infrared spectrometer in a cold sea ice environment has been assessed and verified. The temperature dependence and integration time dependence of the spectrometer output and dark output have been measured over the entire operating temperature range of –50°C to 30°C, well below the stated operating range of this spectrometer. The results indicate that the offset voltage decreases linearly with increasing temperature and dark voltage increases exponentially with increasing temperatures at temperatures above –10°C. A dark output prediction model is developed to represent the dark output as a function of temperature and integration time with an uncertainty of ±0.6% for all wavelengths of interest. The saturation output and upper limit of the linear operating region of the spectrometer increase with increasing temperature, at a rate of 42.16 counts/°C and 55.53 counts/°C, respectively. Signal output increases with increasing temperature in an approximately parabolic manner when the spectrometer is operated in the linear operating range. However, it increases linearly with longer integration times over the entire operating temperature of –50°C to 30°C. Two signal output correction models are developed and compared to convert the signal output measured at any temperature within the operating range and integration time to that measured at the reference temperature T_ref and integration time t_ref, with respective model uncertainties of ±1.5% and ±0.5% for the spectral range considered. The overall performance of the spectrometer module is testified in an ice pack in the Arctic. The results denote that the general shape of the solar irradiance above the snow surface measured by the spectrometer agrees well with that at the sea level, as measured by other commercially available oceanographic spectroradiometers. Additionally, the calculated optical properties of the sea ice where the spectrometer is deployed are identical to those pertaining to similar sea ice calculated from measurements by other researches. Power consumption of the spectrometer together with its driver electronics is ~265 mW in working mode and ~10 mW in idle mode in its current configuration. Given that the spectrometer makes measurement once a day with the measurement time of 15 min, a battery with a capacity of at least 9 800 mAh is needed to maintain the normal operation of spectrometer for 12 months. Future work will focus on longer-term change assessment in stability or sensitivity of the spectrometer, field verification of the spectrometer performance, and comparative measurements between this spectrometer and oceanographic spectrometers to further determine the feasibility of the spectrometer.