Development of Raman Lidar for Remote Sensing of CO 2 Leakage at an Artiﬁcial Carbon Capture and Storage Site

: We developed a Raman lidar system that can remotely detect CO 2 leakage and its volume mixing ratio (VMR). The system consists of a laser, a telescope, an optical receiver, and detectors. Indoor CO 2 cell measurements show that the accuracy of the Raman lidar is 99.89%. Field measurements were carried out over a four-day period in November 2017 at the Eumsong Environmental Impact Evaluation Test Facility (EIT), Korea, where a CO 2 leak was located 0.2 km from the Raman lidar. The results show good agreement between CO 2 VMR measured by the Raman lidar system (CO 2 VMR Raman LIDAR ) and that measured by in situ instruments (CO 2 VMR In-situ ). The correlation coefﬁcient (R), mean absolute error (MAE), root mean square error (RMSE), and percentage difference between CO 2 VMR In-situ and CO 2 VMR Raman LIDAR are 0.81, 0.27%, 0.37%, and 4.92%, respectively. The results indicate that Raman lidar is an effective tool in detecting CO 2 leakage and in measuring CO 2 VMR remotely.


Introduction
The average temperature of Earth has increased by 0.8 • C since the beginning of the twentieth century, garnering worldwide interest [1]. This global warming is reported to be caused by an increase in the concentrations of greenhouse gases (e.g., CO 2 , CH 4 , N 2 O, O 3 , and CFCs) produced by human activities such as fossil fuel combustion, industrial processes, and deforestation [2,3].
As concern about increasing greenhouse gas concentrations has increased, many countries have begun to regulate carbon emissions [4] according to the Kyoto Protocol [5]. Carbon capture and storage (CCS) is considered a promising technology to reduce atmospheric CO 2 [3,6]. However, leakage of CO 2 from large-scale CCS can affect the CO 2 reduction efficiency and have serious impacts on the surrounding ecosystem, such as reduced soil pH, damage to plant and microbial communities, and groundwater contamination [7]. Previous studies [8,9] have monitored CO 2 leaking from CCS sites. Hui et al. [8] assessed CO 2 leakage from a storage facility in Xuzhou city, Jiangsu province, China, using multiple in situ instruments and wireless sensor networks. Elio et al. [9] reported a technique z 0 α(λ, z)dz × β(λ, z) × exp − z 0 α(λ, z)dz S area z 2 × ξ(z) × η λ , (1) where P(λ, z) is the returned backscatter signal from distance z at the laser wavelength λ, E(λ) is the initial laser energy, α is the total (=aerosol + molecular) extinction coefficient, β is the total (=aerosol + molecular) backscatter coefficient, S area is the telescope receiver area, ξ describes the overlap function, and η λ is the receiver efficiency that reflects how many of the incoming photons are detected. Raman scattering refers to a process of inelastic scattering of light by molecules such as CO 2 in which the wavelength of the incident light is changed [14]. The Nd:YAG laser of wavelength 355 nm used in this study is shifted to a wavelength of 371.6 nm caused by the vibrational Raman process of CO 2 molecules. For reference, the Raman scattering wavelength of N 2 molecules in dry air in the lower troposphere is 386.7 nm. By using Equation (1), the backscattered signal from a Raman CO 2 channel at 371.6 nm, P CO 2 , can be expressed as where k CO 2 is a proportionality constant for the system optical efficiency, the telescope receiver area, photomultiplier tube (PMT) efficiency, and the laser output energy. σ CO 2 is the backscattered cross-section for CO 2 caused by Raman scattering, n CO 2 is the number density for CO 2 as a function of distance z, and α(λ) is the volume extinction coefficient at wavelength λ.
Remote Sens. 2018, 10, 1439 3 of 12 where k N 2 is a proportionality constant for the system optical efficiency, the telescope receiver area, photomultiplier tube (PMT) efficiency, and the laser output energy. σ N 2 is the backscattered cross-section for N 2 caused by Raman scattering, n N 2 is the number density for N 2 as a function of distance z, and α(λ) is the volume extinction coefficient at wavelength λ.
where q(λ 0 , z 0 , z) is the atmospheric transmittance from Raman lidar at distance z 0 to z at wavelength 355 nm. We used Raman signals of N 2 as a reference to calculate the CO 2 mixing ratio. The CO 2 mixing ratio is the mass of CO 2 divided by the mass of dry air in a given volume [12]. Since N 2 is present at a constant rate in dry air, the N 2 Raman signal, P N 2 , is used as a measure of dry air [13]. The CO 2 mixing ratio can be derived from the normalization signal, which is the quotient of the CO 2 Raman signal divided by the N 2 Raman signal. The normalization signal can be expressed as The backscatter signals of CO 2 and N 2 were used to calculate the normalization signal. The Raman scattering signals of CO 2 and N 2 are obtained at different wavelengths (371.7 and 386.7 nm, respectively), and thus it has been assumed that there is no wavelength dependence with respect to extinction and consequently to atmospheric transmittance. Figure 1 shows a flow chart of the process used to calculate the CO 2 volume mixing ratio using the Raman lidar system. First, the normalized ratio is calculated using the CO 2 and N 2 inelastic signals caused by Raman scattering, as measured by the Raman lidar system.
In the right part of Figure 1, the regression coefficients are determined using the calculated normalized ratio and the CO 2 VMR measured by the in situ instrument. The regression coefficients are then used in the calibration equation. The surface CO 2 mixing ratios are calculated using a regression between the calculated normalized ratio and the CO 2 VMR measured by an in situ instrument. where ( 0 , 0 , ) is the atmospheric transmittance from Raman lidar at distance 0 to z at wavelength 355 nm. We used Raman signals of N2 as a reference to calculate the CO2 mixing ratio. The CO2 mixing ratio is the mass of CO2 divided by the mass of dry air in a given volume [12]. Since N2 is present at a constant rate in dry air, the N2 Raman signal, 2 , is used as a measure of dry air [13]. The CO2 mixing ratio can be derived from the normalization signal, which is the quotient of the CO2 Raman signal divided by the N2 Raman signal. The normalization signal can be expressed as The backscatter signals of CO2 and N2 were used to calculate the normalization signal. The Raman scattering signals of CO2 and N2 are obtained at different wavelengths (371.7 and 386.7 nm, respectively), and thus it has been assumed that there is no wavelength dependence with respect to extinction and consequently to atmospheric transmittance. Figure 1 shows a flow chart of the process used to calculate the CO2 volume mixing ratio using the Raman lidar system. First, the normalized ratio is calculated using the CO2 and N2 inelastic signals caused by Raman scattering, as measured by the Raman lidar system.
In the right part of Figure 1, the regression coefficients are determined using the calculated normalized ratio and the CO2 VMR measured by the in situ instrument. The regression coefficients are then used in the calibration equation. The surface CO2 mixing ratios are calculated using a regression between the calculated normalized ratio and the CO2 VMR measured by an in situ instrument.

Raman Lidar Setup
The Raman lidar system consists of the third harmonic of a Nd:YAG laser of wavelength 355 nm with 80 mJ energy and 20 Hz repetition rate, a telescope, an optical receiver, and detectors. The

Raman Lidar Setup
The Raman lidar system consists of the third harmonic of a Nd:YAG laser of wavelength 355 nm with 80 mJ energy and 20 Hz repetition rate, a telescope, an optical receiver, and detectors. The pulsed laser is emitted horizontally to the ground from the Nd:YAG laser. Since the laser is emitted horizontally, the diameter of the laser beam was expanded, using a beam expander, by a factor of five times to ensure eye safety. The eye safety distance of our Raman lidar system is 208 m. The Raman lidar is applied co-axially with respect to the optical system to detect CO 2 near the location of the lidar, with an overlap distance of 5-10 m. To observe CO 2 mixing ratios, the Raman lidar system measures backscatter signals caused by CO 2 Raman, N 2 Raman, and Rayleigh-Mie scattering. These signals are detected and recorded as a function of distance and wavelength. To achieve real-time monitoring, a three-channel signal collection system was designed to collect the three types of backscatter signals simultaneously ( Figure 2). pulsed laser is emitted horizontally to the ground from the Nd:YAG laser. Since the laser is emitted horizontally, the diameter of the laser beam was expanded, using a beam expander, by a factor of five times to ensure eye safety. The eye safety distance of our Raman lidar system is 208 m. The Raman lidar is applied co-axially with respect to the optical system to detect CO2 near the location of the lidar, with an overlap distance of 5-10 m. To observe CO2 mixing ratios, the Raman lidar system measures backscatter signals caused by CO2 Raman, N2 Raman, and Rayleigh-Mie scattering. These signals are detected and recorded as a function of distance and wavelength. To achieve real-time monitoring, a three-channel signal collection system was designed to collect the three types of backscatter signals simultaneously ( Figure 2). A Schmidt-Cassegrain telescope with a diameter of 15.24 cm is used as an optical receiver. First, the scattered light is collected by the telescope. After passing through a pinhole, it is narrowed by the collimating lens (L1). A dichroic beam splitter (D.M1) reflects the light at wavelengths above 355 nm. Since the Raman scattering signal is weaker than Rayleigh-Mie scattering [12], a notch filter (F1) is used which reflects light at 355 nm and transmits it at all other wavelengths, thus reflecting the elastic signals caused by Rayleigh-Mie (total) scattering and detecting the inelastic signal caused by Raman scattering. The CO2 filter (F2) transmits light with a wavelength of 371.7 nm to detect the Raman scattering signals of CO2 and reflects light at all other wavelengths. The CO2 filter (F3) transmits light at 386.7 nm and reflects all other wavelengths to detect the Raman scattering signals of N2 needed to calculate the CO2 VMR. The 355 nm filter (F4) transmits light at 355 nm to detect the elastic scattering signal used in aligning the lidar signal. The lenses L2, L3 and L4 are used to focus the CO2 Raman, N2 Raman, and elastic signals into the PMT. The signals detected by the three-channel collecting system are analyzed to determine the CO2 mixing ratio. Table 1 summarizes the main components of the CO2 Raman lidar system.  A Schmidt-Cassegrain telescope with a diameter of 15.24 cm is used as an optical receiver. First, the scattered light is collected by the telescope. After passing through a pinhole, it is narrowed by the collimating lens (L1). A dichroic beam splitter (D.M1) reflects the light at wavelengths above 355 nm. Since the Raman scattering signal is weaker than Rayleigh-Mie scattering [12], a notch filter (F1) is used which reflects light at 355 nm and transmits it at all other wavelengths, thus reflecting the elastic signals caused by Rayleigh-Mie (total) scattering and detecting the inelastic signal caused by Raman scattering. The CO 2 filter (F2) transmits light with a wavelength of 371.7 nm to detect the Raman scattering signals of CO 2 and reflects light at all other wavelengths. The CO 2 filter (F3) transmits light at 386.7 nm and reflects all other wavelengths to detect the Raman scattering signals of N 2 needed to calculate the CO 2 VMR. The 355 nm filter (F4) transmits light at 355 nm to detect the elastic scattering signal used in aligning the lidar signal. The lenses L2, L3 and L4 are used to focus the CO 2 Raman, N 2 Raman, and elastic signals into the PMT. The signals detected by the three-channel collecting system are analyzed to determine the CO 2 mixing ratio. Table 1 summarizes the main components of the CO 2 Raman lidar system.

Indoor CO 2 Cell Measurement
Indoor CO 2 cell measurements were carried out to quantify the accuracy of the lower detection limit of the Raman Lidar system. Figure 3 shows a schematic diagram of the indoor CO 2 cell setup using the Raman lidar. The cell measurement setup consists of a CO 2 gas vessel, a mass flow controller (MFC), a CO 2 gas cell, a vacuum pump, and the Raman lidar system. CO 2 in the CO 2 gas vessel is injected into the CO 2 cell using the MFC, which controls the CO 2 mixing ratio in the cell. A vacuum pump is connected to the cell to release CO 2 gas from the cell into the air.
To measure the VMR of CO 2 gas using Raman lidar, the inside of the CO 2 cell is evacuated and CO 2 gas is injected. The CO 2 VMR of the cell is set to range from 10% to 100%, as the lowest measurement unit of the vacuum gauge used in this study is 1%. The inelastic signals of CO 2 and N 2 caused by Raman scattering are subsequently measured by the Raman lidar and are hereafter referred to as the normalized ratio.

Indoor CO2 Cell Measurement
Indoor CO2 cell measurements were carried out to quantify the accuracy of the lower detection limit of the Raman Lidar system. Figure 3 shows a schematic diagram of the indoor CO2 cell setup using the Raman lidar. The cell measurement setup consists of a CO2 gas vessel, a mass flow controller (MFC), a CO2 gas cell, a vacuum pump, and the Raman lidar system. CO2 in the CO2 gas vessel is injected into the CO2 cell using the MFC, which controls the CO2 mixing ratio in the cell. A vacuum pump is connected to the cell to release CO2 gas from the cell into the air.  Figure 4 shows the correlation between the normalized ratio measured by CO 2 Raman lidar and the CO 2 VMRs from the CO 2 cell. The x-axis represents the CO 2 VMR inside the CO 2 cell (CO 2 VMR CELL ) and the y-axis represents the normalized ratio, calculated from the inelastic signals of CO 2 and N 2 caused by Raman scattering (see Equation (5)). The correlation coefficient (R) between the CO 2 VMR Raman LIDAR and the normalized ratio is 1, showing excellent agreement. The deviation from the regression line observed at CO 2 VMR of 10% in Figure 4 is due to the inaccuracy of the vacuum gauge used to set the CO 2 concentration in the vacuum cell. The vacuum gauge has a concentration error of 1%. Next, the CO 2 VMR in the cell is retrieved using the normalized ratio and the regression equation obtained in Figure 4 based on the method displayed in Figure 1. To measure the VMR of CO2 gas using Raman lidar, the inside of the CO2 cell is evacuated and CO2 gas is injected. The CO2 VMR of the cell is set to range from 10% to 100%, as the lowest measurement unit of the vacuum gauge used in this study is 1%. The inelastic signals of CO2 and N2 caused by Raman scattering are subsequently measured by the Raman lidar and are hereafter referred to as the normalized ratio. Figure 4 shows the correlation between the normalized ratio measured by CO2 Raman lidar and the CO2 VMRs from the CO2 cell. The x-axis represents the CO2 VMR inside the CO2 cell (CO2 VMRCELL) and the y-axis represents the normalized ratio, calculated from the inelastic signals of CO2 and N2 caused by Raman scattering (see Equation (5)). The correlation coefficient (R) between the CO2 VMRRaman LIDAR and the normalized ratio is 1, showing excellent agreement. The deviation from the regression line observed at CO2 VMR of 10% in Figure 4 is due to the inaccuracy of the vacuum gauge used to set the CO2 concentration in the vacuum cell. The vacuum gauge has a concentration error of 1%. Next, the CO2 VMR in the cell is retrieved using the normalized ratio and the regression equation obtained in Figure 4 based on the method displayed in Figure 1.  Figure 5 shows the correlation between CO2 VMRCELL and CO2 VMR as measured by the Raman lidar system (CO2 VMRRaman LIDAR). CO2 VMRRaman LIDAR obtained from CO2 Raman Lidar shows a good agreement with CO2 VMRCELL. Both R and the slope between CO2 VMRCELL and CO2 VMRRaman LIDAR are 1.00. An R value close to 1 indicates more stable and consistent laser power and higher repeatability of the detector than an R value lower than 1, because CO2 is the only variable that changes during the training and retrieval of indoor cell measurements of CO2. If the R value departs from 1, then the output power of the laser is not consistent or the repeatability of the detector is not sufficient for CO2 retrieval. The CO2 in the cell was measured 20 times by Raman lidar under constant conditions. We calculated the error of the value measured by Raman lidar when the CO2 VMR in the cell was the same as that during CO2 measurements. The CO2 measurement accuracy of our Raman lidar is 99.89% based on the indoor CO2 cell test.  Figure 5 shows the correlation between CO 2 VMR CELL and CO 2 VMR as measured by the Raman lidar system (CO 2 VMR Raman LIDAR ). CO 2 VMR Raman LIDAR obtained from CO 2 Raman Lidar shows a good agreement with CO 2 VMR CELL . Both R and the slope between CO 2 VMR CELL and CO 2 VMR Raman LIDAR are 1.00. An R value close to 1 indicates more stable and consistent laser power and higher repeatability of the detector than an R value lower than 1, because CO 2 is the only variable that changes during the training and retrieval of indoor cell measurements of CO 2 . If the R value departs from 1, then the output power of the laser is not consistent or the repeatability of the detector is not Remote Sens. 2018, 10, 1439 7 of 12 sufficient for CO 2 retrieval. The CO 2 in the cell was measured 20 times by Raman lidar under constant conditions. We calculated the error of the value measured by Raman lidar when the CO 2 VMR in the cell was the same as that during CO 2 measurements. The CO 2 measurement accuracy of our Raman lidar is 99.89% based on the indoor CO 2 cell test.

Field Test
To examine the Raman lidar system's capability in measuring spatially resolved CO2 and N2 Raman signals remotely, it was used to detect signals at a distance of 900 m from the lidar. The test was carried out on 30 October, 2017, at a study site in Daejeon, South Korea, as shown in Figure 6a. We detected the Raman signals of CO2 and N2 for 1 h during the night. The effective spatial resolution of the Raman lidar was 250 ns (37.5 m). CO2 and N2 inelastic signals caused by Raman scattering were measured between the lidar system and a mountain location. During the daytime, the background signal from sunlight is stronger than the Raman scattering signal, so measurements were made only at night. Figure 6b shows the Raman signals of CO2 and N2 as a function of distance from the lidar system and the motorway. The increased CO2 Raman signals found at 400 and 750 m from the lidar system represent the locations of a local motorway (A) and the Gapcheondosi Expressway (B), respectively. The enhanced Raman signals of CO2 at locations A and B in Figure 6b imply that the lidar system is capable of detecting spatially resolved Raman signals of CO2 emitted from motor vehicles. The Raman lidar obtained clear and stable Raman scattering signals of N2 and CO2 at a distance of 900 m from the lidar system.

Field Test
To examine the Raman lidar system's capability in measuring spatially resolved CO 2 and N 2 Raman signals remotely, it was used to detect signals at a distance of 900 m from the lidar. The test was carried out on 30 October 2017, at a study site in Daejeon, South Korea, as shown in Figure 6a. We detected the Raman signals of CO 2 and N 2 for 1 h during the night. The effective spatial resolution of the Raman lidar was 250 ns (37.5 m). CO 2 and N 2 inelastic signals caused by Raman scattering were measured between the lidar system and a mountain location. During the daytime, the background signal from sunlight is stronger than the Raman scattering signal, so measurements were made only at night. Figure 6b shows the Raman signals of CO 2 and N 2 as a function of distance from the lidar system and the motorway. The increased CO 2 Raman signals found at 400 and 750 m from the lidar system represent the locations of a local motorway (A) and the Gapcheondosi Expressway (B), respectively. The enhanced Raman signals of CO 2 at locations A and B in Figure 6b imply that the lidar system is capable of detecting spatially resolved Raman signals of CO 2 emitted from motor vehicles. The Raman lidar obtained clear and stable Raman scattering signals of N 2 and CO 2 at a distance of 900 m from the lidar system.

Field Test
To examine the Raman lidar system's capability in measuring spatially resolved CO2 and N2 Raman signals remotely, it was used to detect signals at a distance of 900 m from the lidar. The test was carried out on 30 October, 2017, at a study site in Daejeon, South Korea, as shown in Figure 6a. We detected the Raman signals of CO2 and N2 for 1 h during the night. The effective spatial resolution of the Raman lidar was 250 ns (37.5 m). CO2 and N2 inelastic signals caused by Raman scattering were measured between the lidar system and a mountain location. During the daytime, the background signal from sunlight is stronger than the Raman scattering signal, so measurements were made only at night. Figure 6b shows the Raman signals of CO2 and N2 as a function of distance from the lidar system and the motorway. The increased CO2 Raman signals found at 400 and 750 m from the lidar system represent the locations of a local motorway (A) and the Gapcheondosi Expressway (B), respectively. The enhanced Raman signals of CO2 at locations A and B in Figure 6b imply that the lidar system is capable of detecting spatially resolved Raman signals of CO2 emitted from motor vehicles. The Raman lidar obtained clear and stable Raman scattering signals of N2 and CO2 at a distance of 900 m from the lidar system.

CO 2 -Leakage Measurement Campaign
To examine the CO 2 -leakage measurement capability of the lidar system, measurements were carried out using both the CO 2 Raman lidar system and an in situ CO 2 probe (VAISALA, GMP343). The measurements took place between 20 and 24 November 2017, at the Eumseong Environmental Impact Evaluation Test Facility (EIT) on Seepage of Geologically Stored CO 2 , an artificial CO 2 gas-leakage site (36 • 96 N, 127 • 47 E). As shown in Figure 7, CO 2 is injected at 0.5 m below the surface and left to leak back out. CO 2 gas was injected into the ground at 12 L/min for 12 h on 22 November, 2017. The Raman lidar system is located 200 m from the in situ CO 2 measurement device, which is located near the CO 2 injection location. The Raman lidar line-of-sight is located 0.5 m above the surface of the CO 2 injection inlet. The effective spatial resolution of the Raman lidar is 250 ns (37.5 m) and the beam diameter is about 29.98 cm at the target. The in situ instrument measured the CO 2 VMR at a distance of 1 m from the Raman lidar line-of-sight. A linear regression equation between the normalized ratio and CO 2 VMR measured by the in situ instrument is derived based on the measurements on 24 November, 2017. Six hours after stopping the CO 2 gas injection, the measurement of CO 2 on the surface was started using the CO 2 Raman lidar. The change in Raman scattering signals with respect to the change in the CO 2 mixing ratio on the surface was measured as CO 2 gas was being injected into the ground at 12 L/min. During the campaign period, the in situ instrument with its inlet located 0.5 m from the surface often failed to measure CO 2 leakage near the surface. This is thought to be associated with low CO 2 concentrations even at 0.5 m from the surface, probably due to a negligible exit velocity of CO 2 molecules. To evaluate the performance via comparison with the in situ measurement, the point of view of the lidar system was focused at the in situ inlet location to increase the lidar's sensitivity to CO 2 molecules.

CO2-Leakage Measurement Campaign
To examine the CO2-leakage measurement capability of the lidar system, measurements were carried out using both the CO2 Raman lidar system and an in situ CO2 probe (VAISALA, GMP343). The measurements took place between 20 and 24 November 2017, at the Eumseong Environmental Impact Evaluation Test Facility (EIT) on Seepage of Geologically Stored CO2, an artificial CO2 gas-leakage site (36°96′N, 127°47′E). As shown in Figure 7, CO2 is injected at 0.5 m below the surface and left to leak back out. CO2 gas was injected into the ground at 12 L/min for 12 h on 22 November, 2017. The Raman lidar system is located 200 m from the in situ CO2 measurement device, which is located near the CO2 injection location. The Raman lidar line-of-sight is located 0.5 m above the surface of the CO2 injection inlet. The effective spatial resolution of the Raman lidar is 250 ns (37.5 m) and the beam diameter is about 29.98 cm at the target. The in situ instrument measured the CO2 VMR at a distance of 1 m from the Raman lidar line-of-sight. A linear regression equation between the normalized ratio and CO2 VMR measured by the in situ instrument is derived based on the measurements on 24 November, 2017. Six hours after stopping the CO2 gas injection, the measurement of CO2 on the surface was started using the CO2 Raman lidar. The change in Raman scattering signals with respect to the change in the CO2 mixing ratio on the surface was measured as CO2 gas was being injected into the ground at 12 L/min. During the campaign period, the in situ instrument with its inlet located 0.5 m from the surface often failed to measure CO2 leakage near the surface. This is thought to be associated with low CO2 concentrations even at 0.5 m from the surface, probably due to a negligible exit velocity of CO2 molecules. To evaluate the performance via comparison with the in situ measurement, the point of view of the lidar system was focused at the in situ inlet location to increase the lidar's sensitivity to CO2 molecules.  Figure 8 shows the time series of CO2 VMRIn-situ measured by the in situ instrument (blue line) and CO2 VMRRaman LIDAR measured by the Raman lidar system (red lines) at the Eumseong EIT site during the field campaign. The CO2 gas was artificially injected into the ground and left to leak from the surface. An increasing trend in both CO2 VMRIn-situ and CO2 VMRRaman LIDAR is expected with continued CO2 injection. In Figure 8, the CO2 VMR measured by the CO2 Raman lidar increases steadily over time, while the CO2 VMR measured by the in situ instrument shows an increasing trend with large variability over a short time interval. The reason for this difference between the steadily increasing CO2 VMRRaman LIDAR and the increasing CO2 VMRIn-situ trend with large fluctuations could be associated with differences in measurement coverage. The spatial effective resolution of the Raman lidar is 37.5 m, which is different from the in situ instrument inlet area with a diameter of 1.5 cm. It is clearly shown that CO2 VMRRaman LIDAR tends to be lower than CO2 VMRIn-situ obtained from the CO2 leakage spot in Figure 9, which implies a certain influence of the ambient CO2 VMR on the CO2 VMRRaman LIDAR since the ambient CO2 VMR is likely to be lower than the CO2 VMR of the leakage and also does not change rapidly for a short time period [15]. In the present study, if the CO2 concentration of the leak is lower than the concentration of CO2 leaking from the field campaign in Figure 7. Schematic representation of the Eumseong Environmental Impact Evaluation Test Facility for the seepage of geologically stored CO 2 . Figure 8 shows the time series of CO 2 VMR In-situ measured by the in situ instrument (blue line) and CO 2 VMR Raman LIDAR measured by the Raman lidar system (red lines) at the Eumseong EIT site during the field campaign. The CO 2 gas was artificially injected into the ground and left to leak from the surface. An increasing trend in both CO 2 VMR In-situ and CO 2 VMR Raman LIDAR is expected with continued CO 2 injection. In Figure 8, the CO 2 VMR measured by the CO 2 Raman lidar increases steadily over time, while the CO 2 VMR measured by the in situ instrument shows an increasing trend with large variability over a short time interval. The reason for this difference between the steadily increasing CO 2 VMR Raman LIDAR and the increasing CO 2 VMR In-situ trend with large fluctuations could be associated with differences in measurement coverage. The spatial effective resolution of the Raman lidar is 37.5 m, which is different from the in situ instrument inlet area with a diameter of 1.5 cm. It is clearly shown that CO 2 VMR Raman LIDAR tends to be lower than CO 2 VMR In-situ obtained from the CO 2 leakage spot in Figure 9, which implies a certain influence of the ambient CO 2 VMR on the CO 2 VMR Raman LIDAR since the ambient CO 2 VMR is likely to be lower than the CO 2 VMR of the leakage and also does not change rapidly for a short time period [15]. In the present study, if the CO 2 Remote Sens. 2018, 10, 1439 9 of 12 concentration of the leak is lower than the concentration of CO 2 leaking from the field campaign in the case of a small leak, our current Raman lidar system with a spatial resolution of 37.5 m is barely able to detect the CO 2 leak due to the long spatial resolution of the system. Therefore, the spatial resolution of the lidar system is an important factor in improving the CO 2 precision and detection limit, especially for small leaks. A high-speed photon counter, which provides fine spatial resolution, is required to enhance the precision and detection limit. The LDL and precision of the Raman lidar are also thought to affect the overall efficiency and capability of the lidar to detect such small variations in low CO 2 levels. The trend difference in Figure 8 between CO 2 VMR In-situ and CO 2 VMR Raman LIDAR under windy conditions could be associated with the wind direction between a CO 2 leak and the measurement location of each instrument, as the closest distance between the Raman lidar line-of-sight and the in situ instrument is 1 m. However, the effect of wind direction and speed could not be quantified due to the unavailability of wind data during the campaign period.
Remote Sens. 2018, 10, x FOR PEER REVIEW 9 of 12 the case of a small leak, our current Raman lidar system with a spatial resolution of 37.5 m is barely able to detect the CO2 leak due to the long spatial resolution of the system. Therefore, the spatial resolution of the lidar system is an important factor in improving the CO2 precision and detection limit, especially for small leaks. A high-speed photon counter, which provides fine spatial resolution, is required to enhance the precision and detection limit. The LDL and precision of the Raman lidar are also thought to affect the overall efficiency and capability of the lidar to detect such small variations in low CO2 levels. The trend difference in Figure 8 between CO2 VMRIn-situ and CO2 VMRRaman LIDAR under windy conditions could be associated with the wind direction between a CO2 leak and the measurement location of each instrument, as the closest distance between the Raman lidar line-of-sight and the in situ instrument is 1 m. However, the effect of wind direction and speed could not be quantified due to the unavailability of wind data during the campaign period.   Figure 9 shows the results of linear regression for CO2 VMRIn-situ and CO2 VMRRaman LIDAR at the Eumseong EIT site, revealing a correlation. The R value and slope between CO2 VMRIn-situ and CO2 VMRRaman LIDAR are 0.81 and 0.15, respectively. The mean absolute error (MAE), the root mean square error (RMSE) and the percentage difference between CO2 VMRIn-situ and CO2 VMRRaman LIDAR are the case of a small leak, our current Raman lidar system with a spatial resolution of 37.5 m is barely able to detect the CO2 leak due to the long spatial resolution of the system. Therefore, the spatial resolution of the lidar system is an important factor in improving the CO2 precision and detection limit, especially for small leaks. A high-speed photon counter, which provides fine spatial resolution, is required to enhance the precision and detection limit. The LDL and precision of the Raman lidar are also thought to affect the overall efficiency and capability of the lidar to detect such small variations in low CO2 levels. The trend difference in Figure 8 between CO2 VMRIn-situ and CO2 VMRRaman LIDAR under windy conditions could be associated with the wind direction between a CO2 leak and the measurement location of each instrument, as the closest distance between the Raman lidar line-of-sight and the in situ instrument is 1 m. However, the effect of wind direction and speed could not be quantified due to the unavailability of wind data during the campaign period.   Figure 9 shows the results of linear regression for CO2 VMRIn-situ and CO2 VMRRaman LIDAR at the Eumseong EIT site, revealing a correlation. The R value and slope between CO2 VMRIn-situ and CO2 VMRRaman LIDAR are 0.81 and 0.15, respectively. The mean absolute error (MAE), the root mean square error (RMSE) and the percentage difference between CO2 VMRIn-situ and CO2 VMRRaman LIDAR are Figure 9. CO 2 VMR In-situ versus CO 2 VMR Raman LIDAR diagram. Figure 9 shows the results of linear regression for CO 2 VMR In-situ and CO 2 VMR Raman LIDAR at the Eumseong EIT site, revealing a correlation. The R value and slope between CO 2 VMR In-situ and CO 2 VMR Raman LIDAR are 0.81 and 0.15, respectively. The mean absolute error (MAE), the root mean square error (RMSE) and the percentage difference between CO 2 VMR In-situ and CO 2 VMR Raman LIDAR are 0.27%, 0.37%, and 4.92%, respectively. The errors due to a decrease in backscattered Raman signals that follow a Poisson distribution are 4.9% (112.5 m), 5.4% (150.0 m), or 6.3% (187.5 m). The uncertainty due to the regression coefficients, as calculated from the residuals between the regression line and measured values, is 5.5%. Thus, the total error in CO 2 VMR Raman LIDAR , which is calculated by error propagation, is estimated to be 7.4% (112.5 m), 7.7% (150.0 m), or 8.4% (187.5 m).

Discussion
Hui et al. [8] monitored CO 2 leakage from a geological CO 2 storage site in Xuzhou, Jiangsu, China, using multiple in situ instruments and wireless sensor networks, and Elio et al. [9] applied a technique for measuring CO 2 soil flux at a CO 2 injection site at Hontomin, Spain. While these in situ instruments have high CO 2 measurement accuracy, a large number of such instruments are needed to monitor large CCS sites. To enhance monitoring efficiency over a large area, remote sensing techniques are desired. Therefore, as part of efforts to develop remote sensing techniques for CO 2 measurements at the Earth's surface, we developed the first compact, lightweight, and portable Raman lidar system for remote sensing of surface CO 2 leakage. To examine the performance of the lidar, it was used to detect strong Raman signals of CO 2 at two roads located 400 and 750 m from the instrument. In the case of measurements at an artificial CO 2 leakage site, the CO 2 VMRs at 0.5 m from the surface and 200 m from the Raman lidar were measured successfully and show a good correlation with data collected by an in situ instrument. It is also necessary to understand the longest remote sensing distance with the highest possible accuracy. In order to quantify the longest remote sensing distance of our Raman lidar system, it should be measured in the area without any obstacles on the line of sight. To obtain a relative accuracy from the comparison between CO 2 VMRs obtained from in situ instrument and the lidar, multiple in situ sensors need to be deployed at several distances on the line of the lidar sight.
During the campaign period, the CO 2 VMRs were detected successfully only near the surface, probably due to the low CO 2 exit velocity from the ground. In the case of such low velocities, the altitude of the lidar line of sight needs to be lowered to measure CO 2 , since this compound is likely to exist near the surface. However, lowering the measurement altitude may cause noise such as fluorescence effect in the backscatter signals if the laser encounters obstacles near the surface or the ground itself, arising from the lowered altitude of the line of sight. Our current CO 2 retrieval algorithm needs to be improved to account for fluorescence effects. In addition, Elio et al. [9] proposed a CCS area of 3 km × 3 km that requires a Raman lidar scan that can provide sufficiently high spatial resolution to detect CO 2 leakage from a small site. In the present study, a laser with 80 mJ power was used in the lidar. A laser with stronger power and a more efficient receiver is needed to generate the Raman signal of CO 2 for sites far from the laser and to detect weak CO 2 Raman signals. In addition, in the case of elevated terrain along the lidar line of sight, the measurement distance on the line of sight is reduced, meaning that a lidar located at a single spot cannot complete two-dimensional horizontal scanning over a large CCS site with elevated terrain.

Conclusions
We developed a Raman lidar system that remotely detects CO 2 leakage and CO 2 VMR. The system consists of a laser, a telescope, an optical receiver, and detectors. In terms of indoor CO 2 cell measurement, CO 2 Raman lidar shows very high accuracy. Field measurements were carried out using Raman lider at the Eumseong EIT site where a CO 2 leak is located 0.2 km from the Raman lidar system. There is good agreement between CO 2 VMR Raman LIDAR as measured by the Raman lidar system and CO 2 VMR In-situ as measured by an in situ instrument at the Eumseong EIT site. The R value and percentage difference between CO 2 VMR In-situ and CO 2 VMR Raman LIDAR are 0.81 and 4.92%, respectively. Discrepancies between the CO 2 values measured by Raman lidar and by the in situ instrument could be due to differences between measurement coverage and the measurement geometry of the two approaches. In the case of low CO 2 exit velocities, it is necessary to measure the CO 2 by lowering the altitude of both lidar and in situ instruments. In the case of lidar, lowering the measurement altitude may cause noise signals such as fluorescence to be included in the backscatter signals when the laser hits the ground due to the terrain. Therefore, the lidar algorithm and measurement methods need to be studied further. The CO 2 VMR measurement resolution and detection limit also requires further study at low levels of CO 2 . In a future experiment, wind data will be used to assess the optimal measurement height for the detection and measurement of CO 2 leaks.