*Communication* **A Fiber-Integrated CRDS Sensor for In-Situ Measurement of Dissolved Carbon Dioxide in Seawater**

**Mai Hu 1,2, Bing Chen <sup>1</sup> , Lu Yao <sup>1</sup> , Chenguang Yang <sup>3</sup> , Xiang Chen <sup>1</sup> and Ruifeng Kan 1,\***


**Abstract:** Research on carbon dioxide (CO<sup>2</sup> ) geological and biogeochemical cycles in the ocean is important to support the geoscience study. Continuous in-situ measurement of dissolved CO<sup>2</sup> is critically needed. However, the time and spatial resolution are being restricted due to the challenges of very high submarine pressure and quite low efficiency in water-gas separation, which, therefore, are emerging the main barriers to deep sea investigation. We develop a fiber-integrated sensor based on cavity ring-down spectroscopy for in-situ CO<sup>2</sup> measurement. Furthermore, a fast concentration retrieval model using exponential fit is proposed at non-equilibrium condition. The in-situ dissolved CO<sup>2</sup> measurement achieves 10 times faster than conventional methods, where an equilibrium condition is needed. As a proof of principle, near-coast in-situ CO<sup>2</sup> measurement was implemented in Sanya City, Haina, China, obtaining an effective dissolved CO<sup>2</sup> concentration of ~950 ppm. The experimental results prove the feasibly for fast dissolved gas measurement, which would benefit the ocean investigation with more detailed scientific data.

**Keywords:** seawater dissolved gas; carbon dioxide; optical cavity ring-down spectroscopy; in-situ measurement

#### **1. Introduction**

Marine carbon cycling is the result of a series of physical, geological and biological processes on a spatiotemporal scale [1,2]. The ocean absorbs one-third of the anthropogenic carbon emission, about 2 billion tons per year. As such, the ocean becomes one important place for carbon sequestration [3]. Furthermore, CO<sup>2</sup> is the main greenhouse gas, the main dissolved gas of seawater and the main fluid component of the deep-sea extreme window of cold spring and hydrothermal solution. Precise measurement on its spatiotemporal distribution is significant to investigate the biogeochemical material cycle and global climate change [4,5]. However, common methods based on sampling-laboratory analysis are not enough to support modern marine science. In-situ CO<sup>2</sup> sensors with high sensitivity, high fidelity, large dynamic range and fast response are highly needed in many cutting-edge research topics, such as the sea-air exchange flux of CO<sup>2</sup> [6], CO<sup>2</sup> concentration of deep-sea cold spring and hydrothermal fluid components [7–9], and isotope measurement [7,8,10,11].

Currently, optical technology, semiconductor gas sensing and mass spectrometry [8,9,11–13] are common methods in in-situ measurement of dissolved gas in seawater. Among them, laser-based in-situ optical spectrometer is suitable for greenhouse gas sensing in seawater due to its unique selectivity and sensitivity [12,14,15]. Using infrared spectroscopy technology, the German Hydro C company demonstrated dissolved CH<sup>4</sup> measurement in seawater [16]. Using the off-axis integrated cavity output spectroscopy (OA-ICOS), the LGR Company in the United States measured the dissolved CH4/CO<sup>2</sup> and

**Citation:** Hu, M.; Chen, B.; Yao, L.; Yang, C.; Chen, X.; Kan, R. A Fiber-Integrated CRDS Sensor for In-Situ Measurement of Dissolved Carbon Dioxide in Seawater. *Sensors* **2021**, *21*, 6436. https://doi.org/10.3390/ s21196436

Academic Editor: Jin Li

Received: 21 August 2021 Accepted: 24 September 2021 Published: 27 September 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

its isotope δ <sup>13</sup>CH<sup>4</sup> in seawater [11]. Using mid-infrared absorption spectroscopy technology, Zheng Chuantao et al. achieved the measurement of dissolved CO<sup>2</sup> and its isotope δ <sup>13</sup>CO<sup>2</sup> [12]. In addition, cavity ring-down spectroscopy (CRDS), proposed by O, Keefe and Deacon in 1988 [17], has ultra-high detection sensitivity, light intensity jitter immunity and free instrument calibration, making CRDS one of the best candidates for gas detection. The past years have witnessed its remarkable progresses in precision spectroscopy [18–20] and important applications in atmospheric trace gas measurement [21–23]. However, the application of this technology in the marine field for in-situ measurement remains unresolved due to the challenges of high stability resonant cavity, high precision/low power circuit and time-consuming dissolved gas concentration retrieval.

In this paper, we report the development of an in-situ CRDS based dissolved CO<sup>2</sup> sensor. A fast exponential regression model is proposed to retrieve the concentration of dissolved gas in seawater. In the implementation, we design high-pressure assembling and use polydimethylsiloxane (PDMS) membrane for water/gas separation and enrichment [24,25]. A long-time in-situ observation near the coast is carried out to prove the feasibility of in-situ separation, enrichment and measurement of dissolved CO<sup>2</sup> in seawater.

#### **2. Principle of CRDS-Based Seawater Dissolved Gas Measurement**

#### *2.1. Water/Gas Separation and Enrichment, and Dissolved Gas Retrieval*

A PDMS membrane is one common tool, as the gas-liquid interface with a thickness of l, to separate and enrich seawater dissolved gases [26,27]. The concentration difference between both sides enables the dissolved gas pass through the membrane and blocks liquid water molecule (H2O)n. Thus, small gas molecules, such as CH4, CO<sup>2</sup> and O<sup>2</sup> can be separated from seawater. The water/gas separation of PDMS membrane is typically described by the "dissolution-diffusion" model [24,26]. When concentration difference between both sides exists, gas molecules diffuse into the membrane and realize gas exchange.

In the case of a stable situation, the dissolved gas concentration remains stable along the direction of film thickness. The diffusion flux on the side of the gas chamber can be expressed by Fick's first law [24,26,27]:

$$F\_G = \frac{D\_G S\_G A (P\_{G1} - P\_{G2})}{l} \,\tag{1}$$

where *F<sup>G</sup>* (cm<sup>3</sup> ·cm<sup>2</sup> (cm2polymer)−<sup>1</sup> ·s −1 ) is the diffusion flux of gas component *G* per unit time, *D<sup>G</sup>* (cm<sup>2</sup> ·s −1 ) is the diffusion coefficient of gas component *G* in the membrane, *S<sup>G</sup>* (cm<sup>3</sup> ·(cm2polymer)−<sup>1</sup> ·Pa−<sup>1</sup> ) is the solubility coefficient of gas component *G* in the membrane, *PG*<sup>1</sup> (Pa) is the partial pressure of seawater dissolved gas *G*, *PG*<sup>2</sup> (Pa) is the partial pressure of gas component *G* in the gas chamber, *A* (cm<sup>2</sup> ) is the film area, *l* (cm) is the film thickness. While gas diffusion flux can be expressed by Fick's second law in the case of unstable situation [25,26]:

$$F\_{\rm G,f} = F\_{\rm G,ss} \left( 1 + 2 \sum\_{n=1}^{\infty} (-1)^n \exp\left\{ \frac{-n^2 \pi^2 D\_{\rm G} t}{l^2} \right\} \right), \tag{2}$$

where, *FG*,*<sup>t</sup>* is the gas flux at time *t*, *FG*,*ss* is the gas flux in the stable situation.

The diffusion coefficient of CO<sup>2</sup> in PDMS membrane is about 1.5 <sup>×</sup> <sup>10</sup>−<sup>5</sup> cm2/s, *<sup>l</sup>* 1 cm, *<sup>t</sup>* = 1 s, then exp<sup>n</sup> −*n* 2*π* <sup>2</sup>*DGt l* 2 o is almost zero and *FG*,*<sup>t</sup>* ≈ *FG*,*ss*. Thus, the diffusion flux in this case can also be described by Fick's first law. Therefore, the concentration change of the gas component *G* is as follows:

$$P\_{\rm G2} \* V = \int \frac{D\_{\rm G} S\_{\rm G} A (P\_{\rm G1} - P\_{\rm G2})}{l} d\_{t\nu} \tag{3}$$

where *V* is the volume of the gas chamber. From Equation (3), we can obtain: CH4, the value of <sup>−</sup> ಸௌಸ is independent from partial pressure.

where *V* is the volume of the gas chamber. From Equation (3), we can obtain:

ௗುಸమ ௗ

*Sensors* **2021**, *21*, x FOR PEER REVIEW 3 of 12

$$\frac{d\_{\rm F\_{G2}}}{d\_t} \* V = \frac{D\_{\rm G} S\_{\rm G} A (P\_{\rm G1} - P\_{\rm G2})}{l} \tag{4}$$

With the boundary condition, *t* → ∞, *PG*<sup>2</sup> = *PG*<sup>1</sup> , it will be extrapolated that:

ீଶ ∗= ಸௌಸ(ಸభିಸమ)

With the boundary condition, → ∞, ீଶ = ீଵ, it will be extrapolated that:

where *K* is related to the initial pressure. For non-condensable gases, such as CO2 and

ீଶ = ∗ exp ቀ− ಸௌಸ

∗= ಸௌಸ(ಸభିಸమ)

௧, (3)

, (4)

ቁ + ீଵ, (5)

$$P\_{G2} = K \* \exp\left(-\frac{D\_G S\_G A}{lV}t\right) + P\_{G1} \tag{5}$$

where *K* is related to the initial pressure. For non-condensable gases, such as CO<sup>2</sup> and CH4, the value of − *DGS<sup>G</sup> A lV* is independent from partial pressure. decays to *Iout* due to gas absorption, which can be described by the Beer-Lamber law [10,17]:

After measuring the value of *PG*<sup>2</sup> over time in an unbalanced situation, the exponential regression is used to retrieve *PG*<sup>1</sup> . *PG*<sup>1</sup> equals to *PG*<sup>2</sup> in balanced situation. ௨௧ = ∗ exp(− ∗ ), (6)

#### *2.2. Optical Measurement Using CRDS* where, *α* (cm−1) is the spectral absorption coefficient, *L* (cm) is the interaction distance.

When laser with an intensity of *Iin* passes through uniform gas substance, the laser decays to *Iout* due to gas absorption, which can be described by the Beer-Lamber law [10,17]: In the configuration of CRDS, a couple of high reflectivity mirrors, usually higher than 99.99%, enable a significant interaction length extension inside a limited physical

$$I\_{\rm out} = I\_{\rm in} \* \exp(-\mathfrak{a} \* L),\tag{6}$$

where, *α* (cm−<sup>1</sup> ) is the spectral absorption coefficient, *L* (cm) is the interaction distance. Its working principle is shown in Figure 1. When laser beam resonates with one resonant

In the configuration of CRDS, a couple of high reflectivity mirrors, usually higher than 99.99%, enable a significant interaction length extension inside a limited physical space [28], and become capable of detecting minor absorption of trace gas concentration. Its working principle is shown in Figure 1. When laser beam resonates with one resonant cavity mode, the laser power inside the cavity will be rapidly built up. With a trigger that the inside laser power reaches a certain threshold, the incident laser is quickly cut off to generate a free intracavity ring down. The leaked laser intensity *Iν*(*t*) after each ring down is successively recorded at the exit to obtain optical intensity that decays with time as follows, cavity mode, the laser power inside the cavity will be rapidly built up. With a trigger that the inside laser power reaches a certain threshold, the incident laser is quickly cut off to generate a free intracavity ring down. The leaked laser intensity ఔ() after each ring down is successively recorded at the exit to obtain optical intensity that decays with time as follows, ఔ(t) = ఔ ∗ exp ቆ− ௧ (1−+ఔ௦)ቇ, (7)

$$I\_{\nu}(\mathbf{t}) = I\_{0\nu} \* \exp\left(-\frac{\text{tc}}{L\_{\text{mirror}}}(\mathbf{1} - \mathbf{R} + a\_{\nu}L\_{\text{mirror}})\right),\tag{7}$$

where, *c* is the speed of light, *t* (µs) is the time, *Lmirrors* (cm) is the physical distance between the two mirrors, *R* is the reflectivity of the optical cavity mirror, *α<sup>ν</sup>* (cm−<sup>1</sup> ) is the spectral absorption coefficient of the specific wavelength; *I*0*<sup>ν</sup>* is the initial laser intensity when the laser is cut off. the two mirrors, *R* is the reflectivity of the optical cavity mirror, ఔ (cm−1) is the spectral absorption coefficient of the specific wavelength; ఔ is the initial laser intensity when the laser is cut off.

**Figure 1.** Schematic of the cavity ringdown technique. M1 and M2: cavity mirrors with high reflectivity (>99.99%). **Figure 1.** Schematic of the cavity ringdown technique. M<sup>1</sup> and M<sup>2</sup> : cavity mirrors with high reflectivity (>99.99%).

The time for the initial laser intensity reduces to 1/e in the measurement is defined as the ring-down time [17]. According to Equation (7), in the case of no gas absorption, the empty cavity ring-down time *τ*<sup>0</sup> is:

$$\pi\_0 = \frac{L\_{\text{mirror}}}{\mathbb{C}(1-R)'} \tag{8}$$

In the presence of gas absorption, the ring-down time *τ<sup>ν</sup>* is:

$$\pi\_{\nu} = \frac{L\_{\text{mirror}}}{\mathcal{C}(1 - R + \alpha\_{\nu} L\_{\text{mirror}})} \,\text{}\,\text{}\tag{9}$$

Combining (8) and (9), we obtain the intracavity spectral absorption coefficient as:

$$\mathfrak{a}\_{\nu} = \frac{1}{\mathbb{C}\mathfrak{r}\_{\nu}} - \frac{1}{\mathbb{C}\mathfrak{r}\_{0}},\tag{10}$$

In the absorption spectrum [29], the absorption coefficient can be expressed as:

$$\alpha\_{\boldsymbol{\nu}} = \mathcal{S}(\boldsymbol{T}) \* P\_{\text{total}} \* X \* \boldsymbol{\psi}(\boldsymbol{\nu}), \tag{11}$$

where *S*(*T*) (cm−<sup>2</sup> ·atm−<sup>1</sup> ) is the intensity of the absorption line, *Ptotal* (atm) is the total pressure of the gas, *X* is the molecular concentration, *ψ*(*ν*) (cm) is the absorption line shape function.

Since integral of *ψ*(*ν*) over *ν* is equals 1, i.e., R <sup>+</sup><sup>∞</sup> −∞ *ψ*(*ν*)*d<sup>ν</sup>* = 1, *S*(*T*) relates to the temperature for a specific absorption line, and *Ptotal* can be measured using a commercial pressure sensor. Therefore, after fitting the absorption coefficient curve to obtain the integrated area *A*, the partial pressure of the gas can be calculated by:

$$P = XP\_{total} = \frac{A}{S(T)}.\tag{12}$$

Thus, the partial pressure of dissolved gas (*PG*<sup>1</sup> ) can be retrieved by measuring the partial pressure of gas (*PG*<sup>2</sup> ) in the cavity and fitting formula (5).

#### **3. In-Situ Dissolved CO<sup>2</sup> Sensor Configuration**

Figure 2 depicts the schematic diagram of the optical dissolved-CO<sup>2</sup> sensor, which comprises three parts, one pressured chamber, one water/gas separation and enrichment unit and one gas measurement unit. The first part, i.e., the pressure chamber, is a dry titanium alloy chamber with an inner diameter of 128 mm, a length of 750 mm and a wall thickness of 10 mm. It can withstand a pressure as high as 57 MPa, which is suitable for experiments at about 4500 m under water. The second part, i.e., the water gas separation and enrichment unit consists of a water pump (SEA-BIRD SBE-5T), a PDMS membrane module (membrane thickness 50 µm, diameter 5 cm, gas separation efficiency 0.034 mL/min at 296 K and 1 atm pressure difference), a drying box, an air pump (KNF NMP05), a filter (Swagelok, SS-2F-05), two needle valves (Swagelok, SS-ORS2) and the gas chamber (optical resonant cavity). With the water pump, the seawater continuously flows through the surface of PDMS membrane at a flow rate of 0.8 L/min, forming a stable dissolved gas concentration field on the surface of the membrane. The dissolved gas permeates into the PDMS membrane. On the other side of the membrane, the permeated gas is desiccated by the drying box, then enters the gas chamber through the needle valve 1 and the filter, and finally returns to the PDMS membrane module through the filter, the needle valve 2 and the gas pump. Thus, water/gas separation and enrichment can be completed. The needle valve permits a flow rate of approximately 50 mL/min.

The third part, i.e., the gas detection unit, includes a control circuit (DSP, TMS320C6748, Texas Instruments, Dallas, TX, USA), a DFB laser (NLK1L5GAAA, NEL, Yokohama, Japan), a semiconductor optical amplifier (SOA, BOA1080P, Throlabs, USA), an isolator (PIISO-

1600-D-L-05-FA, Qinghe Photonics, Shenzhen, China), an optical resonator and a photoelectric detector (GPD, GAP1000FC, GPD Optoelectronics, Salem, MA, USA). Sawtooth signal and step signal generated by the control circuit are combined to drive the laser to achieve laser beam resonance enhancement inside the optical resonator. When the enhanced intracavity laser power, monitored by the detector, exceeds a certain threshold, the incident laser is rapidly cut off using the TTL driven SOA to generate ring-down signal. With the ring down signal recorded, a fast single exponential fitting is performed to calculate the ring down time. After 20 recordings of ring-down time for a single longitudinal mode, the step voltage is reset to match the laser to next longitudinal mode until the whole absorption line of CO<sup>2</sup> is covered. With the averaged absorption spectra, the CO<sup>2</sup> absorbance and concentration are calculated by spectral fitting algorithm. Finally, the concentration of the seawater side gas is retrieved according to the exponential fitting in the unbalanced situation. *Sensors* **2021**, *21*, x FOR PEER REVIEW 5 of 12

**Figure 2.** Framework of in-situ underwater instrument for dissolved gas measurement. **Figure 2.** Framework of in-situ underwater instrument for dissolved gas measurement.

#### *3.1. Absorption Line Selection*

The third part, i.e., the gas detection unit, includes a control circuit (DSP, TMS320C6748, Texas Instruments, Dallas, TX, USA), a DFB laser (NLK1L5GAAA, NEL, Yokohama, Japan), a semiconductor optical amplifier (SOA, BOA1080P, Throlabs, USA), an isolator (PIISO-1600-D-L-05-FA, Qinghe Photonics, Shenzhen, China), an optical resonator and a photoelectric detector (GPD, GAP1000FC, GPD Optoelectronics, Salem, MA, USA). Sawtooth signal and step signal generated by the control circuit are combined to drive the laser to achieve laser beam resonance enhancement inside the optical reso-Appropriate absorption line with high absorption intensity and less background interference can benefit the dissolved CO<sup>2</sup> measurement with high signal-to-noise ratio. Considering the application condition in seawater dissolved gas measurement, absorption spectra of CO2, <sup>13</sup>CO2, H2O and CH<sup>4</sup> within 1599.4–1599.7 nm is simulated based on the HITRAN database [30] (temperature: 296 K, gas pressure: 1.01 <sup>×</sup> <sup>10</sup><sup>5</sup> Pa, CO<sup>2</sup> = 400 ppm, <sup>13</sup>CO<sup>2</sup> =, H2O = 2% and CH<sup>4</sup> = 2 ppm). The simulation results, shown in Figure 3, illustrate that CO<sup>2</sup> absorption coefficient is 2.5 <sup>×</sup>10−<sup>7</sup> cm−<sup>1</sup> with negligible interference. Thus absorption transition R(36) at 6251.761 cm−<sup>1</sup> is selected for the following CO<sup>2</sup> measurement.

#### nator. When the enhanced intracavity laser power, monitored by the detector, exceeds a certain threshold, the incident laser is rapidly cut off using the TTL driven SOA to gen-*3.2. Optical Resonator Design*

exponential fitting in the unbalanced situation.

*3.1. Absorption Line Selection* 

ment.

erate ring-down signal. With the ring down signal recorded, a fast single exponential fitting is performed to calculate the ring down time. After 20 recordings of ring-down time for a single longitudinal mode, the step voltage is reset to match the laser to next longi-The optical resonator is shown in Figure 4. Two identical cavity mirrors M1 and M2 (Layertec) have a diameter of 12.7 mm, a radius of curvature of 1000 mm and a reflectivity of higher than 99.99% (@1500~1700 nm). Lens1 and Lens2 are adjustable focus aspherical collimators (CFC-8X-C, Throlabs, Newton, NJ, USA). To suppress the high-order modes,

tudinal mode until the whole absorption line of CO2 is covered. With the averaged ab-

Appropriate absorption line with high absorption intensity and less background interference can benefit the dissolved CO2 measurement with high signal-to-noise ratio. Considering the application condition in seawater dissolved gas measurement, absorption spectra of CO2, 13CO2, H2O and CH4 within 1599.4–1599.7 nm is simulated based on the HITRAN database [30] (temperature: 296 K, gas pressure: 1.01 × 105 Pa, CO2 = 400 ppm, 13CO2=, H2O=2% and CH4 = 2 ppm). The simulation results, shown in Figure 3, illustrate that CO2 absorption coefficient is 2.5 ×10−7 cm−1 with negligible interference. Thus absorption transition R(36) at 6251.761 cm−1 is selected for the following CO2 measurevapor and CH4.

the laser beam is spatially filtered using a single-mode fiber before illuminating on the photodetector, achieving a high-order modes suppression ratio of better than 100:1. It should be noted that, most of the optical and mechanical components of the optical cavity are fixed with structural adhesive to adapt to the extreme marine environment and improve the system stability. *Sensors* **2021**, *21*, x FOR PEER REVIEW 6 of 12

*Sensors* **2021**, *21*, x FOR PEER REVIEW 6 of 12

**Figure 3.** Simulation of CO2 absorption feature and background interference from of 13CO2, H2O vapor and CH4. **Figure 3.** Simulation of CO<sup>2</sup> absorption feature and background interference from of <sup>13</sup>CO<sup>2</sup> , H2O vapor and CH<sup>4</sup> . ment and improve the system stability.

modes, the laser beam is spatially filtered using a single-mode fiber before illuminating on the photodetector, achieving a high-order modes suppression ratio of better than **Figure 4.** Framework of the optical resonant cavity, SMF: Single mode fiber; M1 and M2: cavity mirrors. Lens1 is a resonant cavity mode matching lens for mode matching of a laser-beam to a **Figure 4.** Framework of the optical resonant cavity, SMF: Single mode fiber; M1 and M2: cavity mirrors. Lens1 is a resonant cavity mode matching lens for mode matching of a laser-beam to a high finesse optical cavity. Lens2 is the light converging lens for coupling light into fiber.

#### 100:1. It should be noted that, most of the optical and mechanical components of the optical cavity are fixed with structural adhesive to adapt to the extreme marine environhigh finesse optical cavity. Lens2 is the light converging lens for coupling light into fiber. *3.3. Longitudinal Mode Matching and Spectral Scanning*

ment and improve the system stability. **Laser Source SMF Optical cavity M1 M2 SMF Photodetector Lens1 Lens2 Figure 4.** Framework of the optical resonant cavity, SMF: Single mode fiber; M1 and M2: cavity mirrors. Lens1 is a resonant cavity mode matching lens for mode matching of a laser-beam to a *3.3. Longitudinal Mode Matching and Spectral Scanning*  The physical distance between the two cavity mirrors is 420 mm, corresponding to a free spectral range (FSR) of 0.0119 cm−1. For each measurement using the continuous-wave CRDS system, the laser frequency is adjusted to resonate with one cavity mode. Since the cavity length remains stable during the measurement, the longitudinal cavity modes are used as the frequency reference to depict spectral absorption curve. To avoid potential spectral distortion from the longitudinal mode leakage during the spectral scanning process, we propose a strategy of longitudinal mode matching in a stepwise manner shown in Figure 5, and the step size is set as 1/5 FSR. Ideally, when the laser resonates with the longitudinal cavity mode q, 5 steps can rematch laser and the resonant cavity. However, it can be hardly realized due to the laser wavelength drift and cavity The physical distance between the two cavity mirrors is 420 mm, corresponding to a free spectral range (FSR) of 0.0119 cm−<sup>1</sup> . For each measurement using the continuous-wave CRDS system, the laser frequency is adjusted to resonate with one cavity mode. Since the cavity length remains stable during the measurement, the longitudinal cavity modes are used as the frequency reference to depict spectral absorption curve. To avoid potential spectral distortion from the longitudinal mode leakage during the spectral scanning process, we propose a strategy of longitudinal mode matching in a stepwise manner shown in Figure 5, and the step size is set as 1/5 FSR. Ideally, when the laser resonates with the longitudinal cavity mode q, 5 steps can rematch laser and the resonant cavity. However, it can be hardly realized due to the laser wavelength drift and cavity vibration. Differently, a high frequency sawtooth modulation (amplitude, 1/4FSR ≤ M ≤ 2/5FSR) is simultaneously superimposed on the step signal to ensure the resonance of each cavity mode with the laser. In the implementation of this strategy, each longitudinal mode resonates at least once in 5-step scanning, and at least one step does not resonate. The discrete spectrum can be obtained by taking the non-resonant step as a marker.

high finesse optical cavity. Lens2 is the light converging lens for coupling light into fiber.

free spectral range (FSR) of 0.0119 cm−1. For each measurement using the continuous-wave CRDS system, the laser frequency is adjusted to resonate with one cavity mode. Since the cavity length remains stable during the measurement, the longitudinal cavity modes are used as the frequency reference to depict spectral absorption curve. To avoid potential spectral distortion from the longitudinal mode leakage during the spectral scanning process, we propose a strategy of longitudinal mode matching in a stepwise manner shown in Figure 5, and the step size is set as 1/5 FSR. Ideally, when the laser resonates with the longitudinal cavity mode q, 5 steps can rematch laser and the resonant cavity. However, it can be hardly realized due to the laser wavelength drift and cavity vibration. Differently, a high frequency sawtooth modulation (amplitude, 1/4FSR ≤ M ≤ 2/5FSR) is simultaneously superimposed on the step signal to ensure the resonance of each cavity mode with the laser. In the implementation of this strategy, each longitudinal mode resonates at least once in 5-step scanning, and at least one step does not resonate.

The discrete spectrum can be obtained by taking the non-resonant step as a marker.

*3.3. Longitudinal Mode Matching and Spectral Scanning* 

The discrete spectrum can be obtained by taking the non-resonant step as a marker.

vibration. Differently, a high frequency sawtooth modulation (amplitude, 1/4FSR ≤ M ≤ 2/5FSR) is simultaneously superimposed on the step signal to ensure the resonance of

mode resonates at least once in 5-step scanning, and at least one step does not resonate.

*Sensors* **2021**, *21*, x FOR PEER REVIEW 7 of 12

**1/4FSR<M<2/5FSR**

**Figure 5.** Schematic of the stepwise spectral scanning. **Figure 5.** Schematic of the stepwise spectral scanning. **4. Experimental Results** 

#### **4. Experimental Results** *4.1. Sensor Performance*

#### **4. Experimental Results**  *4.1. Sensor Performance* After assembling the sensor, calibrated CO2 sample with a certain concentration of

*4.1. Sensor Performance*  After assembling the sensor, calibrated CO2 sample with a certain concentration of 885 ppm (uncertainty, 1%) was sealed inside the gas chamber. The absorption spectrum of CO2 is obtained using the method described in Section 3.3. Voigt spectrum fitting was performed using a Python LMFIT-based program. Figure 6 presents a direct comparison After assembling the sensor, calibrated CO<sup>2</sup> sample with a certain concentration of 885 ppm (uncertainty, 1%) was sealed inside the gas chamber. The absorption spectrum of CO<sup>2</sup> is obtained using the method described in Section 3.3. Voigt spectrum fitting was performed using a Python LMFIT-based program. Figure 6 presents a direct comparison between raw spectrum data of and the fitting curve, and a residual error of 1.5 <sup>×</sup> <sup>10</sup>−<sup>9</sup> cm−<sup>1</sup> . The SNR is calculated to be 500, corresponding to a detection limit of 1.8 ppm. 885 ppm (uncertainty, 1%) was sealed inside the gas chamber. The absorption spectrum of CO2 is obtained using the method described in Section 3.3. Voigt spectrum fitting was performed using a Python LMFIT-based program. Figure 6 presents a direct comparison between raw spectrum data of and the fitting curve, and a residual error of 1.5 × 10−9 cm−1. The SNR is calculated to be 500, corresponding to a detection limit of 1.8 ppm.

**Figure 6.** Comparison of the measured CO2 spectral data and Voigt fitting curve. **Figure 6.** Comparison of the measured CO<sup>2</sup> spectral data and Voigt fitting curve.

**Figure 6.** Comparison of the measured CO2 spectral data and Voigt fitting curve. CO2 samples with different concentration of 2000 ppm, 1600 ppm, 1000 ppm, 500 ppm and 250 ppm were generated by diluting the calibrated 10000 ppm CO2 with pure CO2 samples with different concentration of 2000 ppm, 1600 ppm, 1000 ppm, 500 ppm and 250 ppm were generated by diluting the calibrated 10000 ppm CO2 with pure N2. The uncertainty for above samples is 1%, mainly introduced by the dilution system. The mass flow meter was used to control the inlet flow rate at 50 mL/min. A continuous CO<sup>2</sup> samples with different concentration of 2000 ppm, 1600 ppm, 1000 ppm, 500 ppm and 250 ppm were generated by diluting the calibrated 10000 ppm CO<sup>2</sup> with pure N2. The uncertainty for above samples is 1%, mainly introduced by the dilution system. The mass flow meter was used to control the inlet flow rate at 50 mL/min. A continuous measurement of each CO<sup>2</sup> sample was performed over 45 min and the experimental results are shown in Figure 7a.

N2. The uncertainty for above samples is 1%, mainly introduced by the dilution system. The mass flow meter was used to control the inlet flow rate at 50 mL/min. A continuous

measurement of each CO2 sample was performed over 45 min and the experimental re-

sults are shown in Figure 7a.

sults are shown in Figure 7a.

data.

data.

**2000×10–6**

*Sensors* **2021**, *21*, x FOR PEER REVIEW 8 of 12

**Figure 7.** (**a**) Measured results of CO2 samples ranging from 250 ppm to 2000 ppm and (**b**) a linear fit to the measured **Figure 7.** (**a**) Measured results of CO<sup>2</sup> samples ranging from 250 ppm to 2000 ppm and (**b**) a linear fit to the measured data. A linear fit, shown in Figure 7b, was thus performed to the measured data and the

A linear fit, shown in Figure 7b, was thus performed to the measured data and the R-square value was calculated to be 0.9999, illustrating a good linear response to the CO2 concentration from 250 ppm to 2000 ppm. The relative error of the measured values of the five groups of standard gases is less than 2.66%, which is consistent with the uncertainty of the standard gases. A linear fit, shown in Figure 7b, was thus performed to the measured data and the R-square value was calculated to be 0.9999, illustrating a good linear response to the CO<sup>2</sup> concentration from 250 ppm to 2000 ppm. The relative error of the measured values of the five groups of standard gases is less than 2.66%, which is consistent with the uncertainty of the standard gases. R-square value was calculated to be 0.9999, illustrating a good linear response to the CO2 concentration from 250 ppm to 2000 ppm. The relative error of the measured values of the five groups of standard gases is less than 2.66%, which is consistent with the uncertainty of the standard gases. To prove the feasibility, a long-term (8 h) comparison experiment was carried out

To prove the feasibility, a long-term (8 h) comparison experiment was carried out between this sensor and one commercial land-based instrument CRDS instrument (G2201-i, Picarro). The gas pipelines of G2201-i and this sensor were connected together to guarantee the same the analyte was simultaneously and separately measured. The comparison result, shown in Figure 8, demonstrates that the measured CO2 concentration and its variation trend are consistent with each other and the maximum relative differ-To prove the feasibility, a long-term (8 h) comparison experiment was carried out between this sensor and one commercial land-based instrument CRDS instrument (G2201-i, Picarro). The gas pipelines of G2201-i and this sensor were connected together to guarantee the same the analyte was simultaneously and separately measured. The comparison result, shown in Figure 8, demonstrates that the measured CO<sup>2</sup> concentration and its variation trend are consistent with each other and the maximum relative difference is within 1.3%. between this sensor and one commercial land-based instrument CRDS instrument (G2201-i, Picarro). The gas pipelines of G2201-i and this sensor were connected together to guarantee the same the analyte was simultaneously and separately measured. The comparison result, shown in Figure 8, demonstrates that the measured CO2 concentration and its variation trend are consistent with each other and the maximum relative difference is within 1.3%.

**Figure 8.** Comparison of this sensor to one commercial CRDS instrument (G2201-i, Picarro, Santa **Figure 8.** Comparison of this sensor to one commercial CRDS instrument (G2201-i, Picarro, Santa Clara, CA, USA). **Figure 8.** Comparison of this sensor to one commercial CRDS instrument (G2201-i, Picarro, Santa Clara, CA, USA).

Clara, CA, USA).

#### *4.2. Verification of Exponential Fit Retrieval Method for Dissolved CO<sup>2</sup> Concentration* vue, WA, USA) is used to continuously cycle the solution through the PDMS module to

*4.2. Verification of Exponential Fit Retrieval Method for Dissolved CO2 Concentration* 

Sample solutions are prepared by mixing 2000 ppm CO2 and pure N2 (purity:

99.99%) using two mass flow meters (AST10-DLCMX-100C-025-A2B2-4VE, Asert, Franklin, MA, USA), as shown in Figure 9a. A submersible pump (SBE-5T, SEA-BIRD, Belle-

*Sensors* **2021**, *21*, x FOR PEER REVIEW 9 of 12

Sample solutions are prepared by mixing 2000 ppm CO<sup>2</sup> and pure N<sup>2</sup> (purity: 99.99%) using two mass flow meters (AST10-DLCMX-100C-025-A2B2-4VE, Asert, Franklin, MA, USA), as shown in Figure 9a. A submersible pump (SBE-5T, SEA-BIRD, Bellevue, WA, USA) is used to continuously cycle the solution through the PDMS module to separate dissolved CO<sup>2</sup> to be measured. PDMS module is connected to the chamber of in-situ measurement system via a stainless tube. After 12-h measurement, the exponential model combined with Levenberg Marquardt (LM) algorithm is performed on the measured data. Figure 9b depicts the results with a R-square of 99.84%, when gas-phase CO<sup>2</sup> concentration in the chamber is lower than the dissolved CO2. Figure 9c depicts the results with a R-square of 98.71% when gas-phase CO<sup>2</sup> concentration in the chamber is higher than the dissolved CO2. The coefficient *<sup>D</sup>GS<sup>G</sup> <sup>A</sup> lV* , independent from the gas concentration, inside and outside the membrane are 0.1389 and 0.1444, respectively. The coefficients of two different situations are consistent with each other with a small difference of 3.8%. Thus, the method of exponential fit for dissolved gas concentration proves to be validated. separate dissolved CO2 to be measured. PDMS module is connected to the chamber of in-situ measurement system via a stainless tube. After 12-h measurement, the exponential model combined with Levenberg Marquardt (LM) algorithm is performed on the measured data. Figure 9b depicts the results with a R-square of 99.84%, when gas-phase CO2 concentration in the chamber is lower than the dissolved CO2. Figure 9c depicts the results with a R-square of 98.71% when gas-phase CO2 concentration in the chamber is higher than the dissolved CO2. The coefficient ಸௌಸ , independent from the gas concentration, inside and outside the membrane are 0.1389 and 0.1444, respectively. The coefficients of two different situations are consistent with each other with a small difference of 3.8%. Thus, the method of exponential fit for dissolved gas concentration proves to be validated.

**Figure 9.** (**a**) Schematic of sample solution preparation and dissolved CO2 measurement system. Real-time concentration measurements when CO2 concentration in the cavity is (**b**) lower and (**c**) higher than the dissolved CO2, respectively. **Figure 9.** (**a**) Schematic of sample solution preparation and dissolved CO<sup>2</sup> measurement system. Real-time concentration measurements when CO<sup>2</sup> concentration in the cavity is (**b**) lower and (**c**) higher than the dissolved CO<sup>2</sup> , respectively.

#### *4.3. In-Situ Detection of Dissolved CO<sup>2</sup> in Seawater*

*4.3. In-Situ Detection of Dissolved CO2 in Seawater*  From 6 to 7 March 2021, an in-situ observation was carried out near the coast of Sanya Institute of Deep Sea, Chinese Academy of Sciences, Sanya, Hainan Province. From 6 to 7 March 2021, an in-situ observation was carried out near the coast of Sanya Institute of Deep Sea, Chinese Academy of Sciences, Sanya, Hainan Province. Figure 10 shows the observation location.

Figure 10 shows the observation location. At the beginning of measurement, about 1100 ppm CO<sup>2</sup> was filled in the cavity to quickly balance the concentration inside and outside the membrane. Figure 11 shows the observation result, the whole measurement is divided into two unstable parts and one stable part. In the first unstable part (3 h), the measured concentration was higher than that of seawater dissolved CO<sup>2</sup> due to the presence of 1100 ppm CO<sup>2</sup> in the measurement chamber, and the CO<sup>2</sup> diffused from the measurement chamber into seawater. The concentration of seawater dissolved CO<sup>2</sup> is calculated to be 850 ppm. In the second stable part (9 h), the

gas concentration inside and outside the membrane remained equal, and the dissolved gas concentration fluctuated between 950 ppm to 980 ppm. In the third unstable part (8 h), the concentration of dissolved CO<sup>2</sup> decreased and the CO<sup>2</sup> in the cavity continued to diffuse into seawater. The concentration of dissolved CO<sup>2</sup> in the water is calculated to be 808 ppm. Interestingly, these measured dissolved CO<sup>2</sup> concentrations are much higher than the atmospheric CO<sup>2</sup> concentration, about 400 ppm, mainly because of a number of yachts cruising near the coast and the tide phenomenon. *Sensors* **2021**, *21*, x FOR PEER REVIEW 10 of 12 Latitude : 18.217° Longitude: 109.487°

*Sensors* **2021**, *21*, x FOR PEER REVIEW 10 of 12

**Figure 10.** The in-situ test site is located at 18.217° north latitude and 109.487° east longitude. **Figure 10.** The in-situ test site is located at 18.217◦ north latitude and 109.487◦ east longitude. a number of yachts cruising near the coast and the tide phenomenon.

a number of yachts cruising near the coast and the tide phenomenon. **Figure 11.** The result of In-situ measurements near coast. **Figure 11.** The result of In-situ measurements near coast.

#### **5. Conclusions**

**High tide Low tide –192.559× Exp(–0.1421t)+850 Data1 978× Exp(–0.13976t)+808** We develop a fiber-integrated in-situ dissolved CO<sup>2</sup> sensor using CRDS, in which a PDMS membrane is employed for water/gas separation and enrichment, and an exponential regression model is proposed for fast dissolve CO<sup>2</sup> retrieval. The model feasibility is verified by performing a comparison test under two different situations, i.e., measurement

00:00 05:00 10:00 15:00 20:00

**Figure 11.** The result of In-situ measurements near coast.

chamber CO<sup>2</sup> concentration is higher and lower than the dissolved CO<sup>2</sup> concentration. Both their R-square of fitting are better than 98.5% and the difference of the two individually measured PDMS membrane coefficient *<sup>D</sup>GS<sup>G</sup> <sup>A</sup> lV* is only 3.8%. The whole absorption spectrum of CO<sup>2</sup> can be obtained within 90 s and a detection sensitivity of 1.8 ppm has been achieved. A near coast in-situ measurement has been implemented over 24 h and provided regular fluctuation of dissolved CO<sup>2</sup> concentrations, which is due to the tide phenomenon. Future efforts will be made to improve the corrosion resistance ability by using titanium alloy stainless steel as the pressured chamber material and to improve the gas separation efficiency by increasing the membrane surface area. Therefore, the developed senor could act as a promising tool to achieve high-precision detection of dissolved gas in seawater and then support the investigation on the ocean, such as vertical dissolved CO<sup>2</sup> profile as deep as 4500 m and long-term dissolved CO<sup>2</sup> monitoring under deep-sea extreme environment window, e.g., hydrothermal and cold spring.

**Author Contributions:** Conceptualization, M.H. and B.C.; methodology, M.H. and R.K.; software, M.H. and L.Y.; validation, M.H. and C.Y.; formal analysis, X.C.; investigation, X.C.; resources, C.Y.; writing—original draft preparation, M.H.; writing—review and editing, R.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by the Scientific Instrument Developing Project of the Chinese Academy of Sciences (YJKYYQ20190037), the Second Comprehensive Scientific Investigation of the Qinghai-Tibet Plateau(2019QZKK020802), the National Key Research and Development Project (2019YFB2006003) and the National Natural Science Foundation of China (61805286).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data that support the plots within this paper are available from the corresponding author on request basis.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

