Wireless laser spectroscopic sensor node for atmospheric CO₂ monitoring—laboratory and field test

Abstract

We developed a low-power, portable, wireless laser spectroscopic sensor for atmospheric CO₂ monitoring. The sensor is based on tunable diode laser absorption spectroscopy with a 2-μm wavelength VCSEL as a source and wavelength modulation technique for spectroscopic signal detection. The sensor allows measurement of CO₂ concentration changes with a 1σ sensitivity of 0.14 ppmv Hz^−1/2. This sensor was both laboratory and field tested under varying environmental conditions. It was used to measure a soil respiration rate of topsoil in the lab and of forest floors in the field. Measurement results are compared with those of commercial non-dispersive infrared sensors and very good agreement is found.

1 Introduction

Carbon dioxide (CO₂) is one of the most important greenhouse gases in the atmosphere with a significant impact on the global climate. Monitoring of environmental CO₂ levels would benefit greatly from the use of highly sensitive, yet portable CO₂ sensors. Deployment of wireless sensors in networks for local monitoring of CO₂ sources and sinks, or on lightweight airborne platforms for atmospheric profiling requires compact, low-power, autonomous, wireless sensors. In this study a compact semiconductor laser based CO₂ sensor has been developed and laboratory and field tested in the context of soil respiration monitoring. Indeed, the carbon balance of the world’s ecosystems and the response of these systems to disturbances, such as changing climate, remain uncertain. A major source of this uncertainty is the unknown rate of carbon transformations below ground [1–10]. The contribution of soil to ecosystem carbon flux is of particular interest during transient events (e.g., rain), yet rarely measured [3, 10]. Due to the high spatiotemporal variability of the soil ecosystem a CO₂ monitoring network consisting of compact, sensitive and robust sensors would revolutionize soil respiration measurements and thus give a better estimate of the soil’s contribution to carbon cycling. Additionally, soil respiration measurements are essential in remote ecosystems such as high latitude tundra or boreal forests. Here it is very important that a CO₂ sensor is autonomous and robust enough to maintain its sensitivity in harsh and changing environments [1, 3, 8].

The developed compact CO₂ sensor employs a vertical cavity surface emitting laser (VCSEL) operating at a 2-μm wavelength and a 3.5-m path Herriott multi-pass cell with a physical mirror spacing of about 12 cm (see Fig. 1(a)). The sensor system is based on tunable diode laser absorption spectroscopy (TDLAS) and it uses an optimized version of wireless laser spectroscopic sensor node electronics developed earlier by So et al. [11, 12]. The TDLAS sensor provides adequate sensitivity for various environmental monitoring applications, and it is intended to be both highly portable and robust for use in the field. Its wireless interface enables networking of multiple sensor nodes for large area environmental monitoring. The control electronics used in this work have previously been tested in a simple 3-sensor-node network for point source location of molecular oxygen (O₂) [13].

Fig. 1

(a) The CO₂ sensor as seen from the top. The total size is 16 cm × 24 cm × 9 cm. (b) A schematic representation of the different components of the CO₂ sensor

For soil respiration monitoring experiments, the TDLAS CO₂ sensor was compared with a commercial Vaisala GMP343 non-dispersive infrared (NDIR) CO₂ sensor. The Vaisala GMP343 was chosen for performance comparisons because it is of a popular and robust technology which has been employed in many experiments before, including soil respiration monitoring [4, 5, 14–16]. Section 2 of this paper details the functionality and design of the TDLAS CO₂ sensor. Long-term laboratory tests and sensor stability analysis are also presented in this section. In Sect. 3, laboratory and field CO₂ sensing tests of the TDLAS sensor performance together with cross-comparison with the commercial sensor are presented in an application focused on soil respiration studies.

2 CO₂ TDLAS sensor performance

The TDLAS sensor uses wavelength modulation spectroscopy (WMS) to detect the molecular concentration [17]. The injection current of the VCSEL is modulated at ∼10 kHz and the amplitude of the current modulation corresponds to a WMS modulation index of m=2.2 (where m is the ratio of laser wavelength modulation depth to the half-width at half-maximum HWHM of the target absorption line) [18]. The modulated laser radiation passes through a custom Herriott multi-pass gas sample cell and the transmitted beam is incident on an uncooled InGaAs photodetector. The photodetected signal is then amplified and digitized for further digital signal processing. The system performs lock-in demodulation of the first three harmonics of the WMS signal. The bandwidth of the lock-in amplifier is set to 1 Hz. The control of laser temperature and bias current as well as the required signal processing are all performed by a custom electronics that can transmit data to the central computer either via USB or wirelessly [19]. A simple block diagram of the sensor system and electronics is shown in Fig. 1(b).

For field operation the sensor prototype was enclosed in a 24 cm × 16 cm × 9 cm watertight NEMA enclosure with an inlet and outlet for the gas sample (i.e., ambient air) (Fig. 1(a)). To protect the optics from possible condensation and to measure the dry air CO₂ concentration, the sensor node is also equipped with a dryer (Perma Pure DM Series Desiccant Membrane Dryer) and a small 600 mW pump providing 600 mL/min flow that can be used to deliver the sample gas to the sensor head. The system was powered by a ∼10-Ah, 3.7-V Li-ion battery and could run continuously for 30 hours with the pump on. The sensor reported in this paper was primarily operated without the pump. Instead, the enclosure lid was removed and the surrounding gas sample was delivered through gas diffusion and/or convection. The power consumption of the sensor without the pump (including laser current and temperature control) is approximately 350 mW, and a total power consumption of less than 1 W was achieved for the entire sensor system (with a running pump). The VCSEL frequency can be varied from ∼4985 to ∼4990 cm⁻¹ by changing laser temperature, which provides access to multiple CO₂ absorption lines. The sensor is designed to primarily operate in an absorption line-locked mode using an active feedback that controls laser wavelength by adjusting its temperature. The capability of laser current scanning was not included in this version of the control electronics. All measurements in this work have been performed using the R12 CO₂ line at 4987 cm⁻¹.

First the sensor was evaluated in a scan mode in which a temperature scan was performed to acquire the CO₂ spectrum. A calibration gas containing 285 ppmv (parts-per-million by volume) of CO₂ balanced with N₂ was introduced to the system and a spectrum was collected with 1 s averaging per spectral point. The signal-to-noise ratio (SNR) of 2530 was estimated using the 2f WMS peak signal amplitude observed in the scan and a noise measured away from the absorption line with the laser temperature kept at a constant value (Fig. 2). This corresponds to 1σ detection limit of ∼0.12 ppmv in 1 second averaging time. To perform long-term measurements, the sensor is operated in a line-locked mode. In this mode an active feedback loop is implemented to set the laser frequency at the line center. The WMS 3f signal is used as the error signal because at the line center the 3f spectrum is approximately a linear function of frequency with a zero-crossing coinciding with the 2f spectrum peak. After sensor calibration the peak value of the WMS 2f signal is used for continuous measurement of the CO₂ concentration in the sample.

Fig. 2

A 2f WMS spectrum acquired using a temperature scan of the VCSEL over the CO₂ line at 4987 cm⁻¹. Noise is estimated from the standard deviation of the 2f signal away from the absorption line

Conversion of the peak 2f signal value to CO₂ concentration is done during post-processing using predetermined calibration data obtained from measurements of a calibrated CO₂ gas mixture. Since the total atmospheric CO₂ absorption over the optical path of 3.5 m in this wavelength region is less than 2 %, the Beer–Lambert’s law can be considered linear with molecular number density. At constant temperature and pressure the 2f signal, S _2f , resulting from WMS measurement can also be considered proportional to number density:

$$ S_{2f} \propto n = \frac{CP}{k_{B}T} $$

(1)

where C is CO₂ concentration, P is sample pressure, k _B is the Boltzmann’s constant, and T is temperature. After calibration with a certified gas mixture, the measured CO₂ concentration can be expressed as

$$ C = C_\mathit{ref}\frac{S_{2f}}{S_{2f,\mathit{ref}}}\frac{T}{T_\mathit{ref}}\frac{P_\mathit{ref}}{P} $$

(2)

where C _ref is the concentration of CO₂ in the calibrated mixture used as a reference, and S _2f,ref is the average 2f signal value measured during a long-term (>12 hours) measurement of that calibration mixture at a given pressure P _ref and temperature T _ref . S _2f in Eq. (2) is a measured 2f signal value for the sample, and P and T are the pressure and temperature recorded during the sample measurement. Since Eq. (1) relies on the fact that in the low absorption region the Beer–Lambert’s law can be approximated as linear, the response of the CO₂ sensor for varying calibrated concentrations of CO₂ has been evaluated to ensure that its performance follows this approximation. Figure 3 shows the sensor 2f signal as a function of CO₂ concentration. The CO₂ concentration was varied by using the Environics 4040 gas dilution system to dilute a 4980-ppmv CO₂ in N₂ mixture with a pure N₂ gas. As can be seen from the graph, over a range of CO₂ concentrations between ∼200 and ∼900 ppmv CO₂ that could be reliably generated with the dilution system, the sensor shows a linear response to concentration. This range is also relevant to the concentration levels observed in soil respiration experiments investigated in this work.

Fig. 3

The 2f signal of the CO₂ sensor as the concentration of CO₂ in N₂ mixture is changed from 200 to 900 ppmv with a linear fit of the experimental data points

Since the TDLAS sensor system will eventually be used for long-term measurements, stability over long period of time is critical. To test the system stability, two different long-term measurements were carried out in laboratory conditions and Allan deviation analysis has been performed (Fig. 4(a)) [20]. In both tests the Allan deviation analysis shows lower performance below 1 second averaging. This is a result of output signal averaging that is shorter than a limit set by the 1 Hz bandwidth of the lock-in amplifier. In the first test the VCSEL temperature in the TDLAS sensor was set to a constant value and in the second test the TDLAS sensor was operated in an active line-locking mode (as described above). In both tests the TDLAS sensor system was operated for 12–14 hours with a calibrated mixture of 299 ppmv CO₂ balanced with N₂ continuously flushing the system in order to evaluate long-term instrumental drift. The resulting time series showing the calculated CO₂ concentration from the measured 2f signal for both experiments can be seen in Fig. 4, (b) and (c).

Fig. 4

(a) Allan deviation of the CO₂ absorption signal for 299 ppmv CO₂ in N₂ mixture for a fixed laser temperature and for a line-locked mode. The temporal plots of concentration measurements in both modes of operation are shown in (b) and (c)

In the constant temperature test, the initial TDLAS sensor parameters (VCSEL drive current and temperature) were set to yield the highest SNR, i.e., the VCSEL emission wavelength coincided with the peak of the CO₂ absorption line. A constant drive current was set by the built-in current driver, and the VCSEL temperature was controlled by an active feedback loop, which uses a thermistor in the laser package as a laser temperature sensor. The calculated Allan deviation (Fig. 4(a)) of this test showed white noise performance up to 25 seconds after which drift started dominating. The 1 second detection limit of 0.12 ppmv is consistent with the spectral scan estimated value. The ultimate minimum detectible absorption in this operation mode was 2.5×10⁻⁶, which corresponds to a 0.055-ppmv CO₂ detection limit. After ∼20,000 seconds of averaging, system drift starts dominating and the sensitivity deteriorates to 0.66 ppmv.

In the second test, an active line-locking of the VCSEL operating frequency to the CO₂ transition at 4987 cm⁻¹ has been performed. This active control allowed better stability of the system as demonstrated by the calculated Allan deviation shown in Fig. 4(a). When using active line-locking, the sensor was white noise limited up to ∼100 seconds while the minimum detectible absorption of 1.25×10⁻⁶ (0.029 ppmv CO₂ concentration detection limit) was nearly two times lower than in the first test. The short-term (one second) sensitivity of 0.14 ppmv was slightly deteriorated (from 0.12 ppmv observed in the first test), which suggests that the line locking feedback loop introduces some additional noise in the system that slightly affects short-term performance. This effect is only noticeable at short integration times and at long integration times (∼20,000 seconds averaging) the sensor performance is actually improved with the accuracy better than 0.5 ppmv. Thus the performance of this low power TDLAS sensor technology is comparable to the current state-of-the-art in trace gas sensing [21, 22].

3 Laboratory and field testing

Since this TDLAS sensor has been designed for atmospheric sensing applications, it is important to perform comparison with other commercially available CO₂ sensors. To that end several tests were performed comparing the TDLAS CO₂ sensor to the Vaisala GMP343 CO₂ sensor. Soil respiration monitoring is used as a test application. In all tests, the TDLAS CO₂ sensor was run in line-locked mode and the Vaisala GMP343 was run in the default mode with 50 % humidity, temperature, and pressure correction.

The GMP343 is an NDIR-based sensor that provides CO₂ measurements with ±($5~\mbox{ppmv/}\sqrt{\mathrm{Hz}} +2$ % of reading) accuracy and requires yearly calibration for most environments [23]. These parameters are adequate for atmospheric/soil respiration monitoring. While operating, the total power dissipation of the GMP343 is 1 W without optics heating and 3.5 W with optics heating. Additionally, there is a required 30-minute warm-up time to achieve full accuracy. This actually prevents the NDIR technology from being reliably used in sensor network applications that require lower operating power for battery powered deployments and short warm-up time for convenient duty cycling in power constrained environments.

The TDLAS CO₂ sensor reported in this paper has a significant advantage of being low power with fast duty cycling capability. Figure 5 shows a typical startup sequence. The device was put in a slip mode and has been continuously flushed with a 349-ppmv CO₂ in N₂ mixture and then a startup sequence was initiated (at relative time of 0 s in Fig. 5). Within 5 seconds of startup the TDLAS sensor has already set the VCSEL drive current and temperature to the preset levels, yielding a 348-ppmv measurement. The line-locking is activated approximately at the 10-second mark and the control loop is settled by the 30-second mark (as seen in Fig. 5 in the 3f signal recording used as a feedback error). After the line-lock the system reads the true 349 ppmv concentration level. Thus after 30 seconds total startup time the CO₂ sensor is ready for continuous CO₂ concentration monitoring.

Fig. 5

The CO₂ sensor startup sequence. When line-locking is employed, a stable operation is reached within 30 seconds of startup

While the linearity and stability tests in the preceding section were performed using calibrated CO₂ in N₂ gas mixtures, the soil respiration experiments were performed directly in the open air with dynamically changing concentration levels. To account for a discrepancy between the pressure broadening coefficients of the target transition in air and in N₂, both the TDLAS sensor and the Vaisala sensor were calibrated in a 445.5-ppmv CO₂ in air mixture (the temperature during calibration was set to 24 ^∘C). Once the Vaisala and TDLAS sensors were individually calibrated they were placed in a vessel which was then flushed with the same calibration gas (Fig. 6(a)) and the responses of both sensors were recorded during the gas exchange. Due to spatial constraints, the measurement volumes of the two sensors were separated vertically by approximately 15 cm (with Vaisala mounted on the top of the vessel). The gas inlet and outlet were also located at the top of the vessel, which caused the Vaisala sensor to respond faster to the change in CO₂ concentration. As Fig. 6(a) shows, the Vaisala sensor began to measure a change in CO₂ concentration at the 52-second mark, while the TDLAS sensor began to measure a change at the 84-second mark. This time delay between sensors was corrected in the post-processing by time shifting the TDLAS data by 32 s. A plot of TDLAS measured concentration vs. Vaisala measured concentration is shown in Fig. 6(b) (for both sensors only data above 100-second mark were used). The data were fitted by a linear function yielding a slope of 1.00 and an intercept of −7.23 ppmv. It should be noted that due to nonlinear change in concentration over time, the slope and the intercept values of the sensor correlation line might vary depending on the actual time shift applied to the TDLAS data. This shift is a result of the different locations of the sensors and it depends on many other factors, such as measurement volume or diffusion rate through the protective enclosures of the sensor, which make precise measurement of this parameter difficult. Nevertheless, the measurements performed with both sensors are in excellent agreement, which confirms the capability of the TDLAS sensor to perform reliable measurement of the rate of CO₂ concentration change in soil respiration studies. To further verify that the sensors were appropriately calibrated, each sensor was again independently exposed to the 445.5-ppmv CO₂ in air calibration mixture and they measured 445.5 ppmv (for TDLAS) and 446.1 ppmv (for Vaisala). Additionally, when placed side-by-side in ambient room conditions, the sensors consistently measured the ambient CO₂ concentration to be between 637 and 642 ppmv, with their readings always within approximately 5 ppmv of each other, which is within the specified accuracy of the GMP343.

Fig. 6

(a) TDLAS sensor (in open-path configuration with no pump) and Vaisala GMP343 measurements of changing CO₂ concentration over time at ambient temperature of 24 ^∘C. The gray lines represent the precision of the Vaisala sensor. (b) A scatter plot with linear fit (black line) of the TDLAS sensor measurements vs. the Vaisala sensor measurements for the region starting at 100 seconds. The linear fit has a slope of 1.00 and an intercept of −7.23 ppmv

We have also performed soil respiration measurements similar to those in [8] both in the natural conditions simulated in laboratory and then in the field. In order to measure soil respiration rates in the lab, a rubber bin was filled with topsoil and moistened with water to activate the microbes contained within the soil. A custom-built clear plastic chamber was put on top of the soil and both CO₂ sensors were placed inside to measure the rate of CO₂ increase (see inset in Fig. 7). Several soil respiration measurements were performed in the lab at approximately 24 ^∘C. This was done by first placing the TDLAS sensor on one or more collars above the soil. To trap the emitted CO₂ the plastic chamber with the Vaisala sensor was then placed over the soil and the TDLAS sensor. It should be noted that in soil respiration studies the source of CO₂ is at the bottom of the container and the geometry and thus the time shift between sensors is different than in the previous experiment. It was found that the placement of the Vaisala sensor in relation to the TDLAS sensor was very important to obtain consistent CO₂ measurements. Because there is a significant vertical concentration gradient, differences of >100–200 ppmv are observed at any point in time if the sensors are placed too far apart. To mitigate this problem we have placed the sensors as close as possible with the lid of the TDLAS removed, and a fan was used inside the container to mix the air.

Fig. 7

An in-lab CO₂ soil respiration experiment. The TDLAS and Vaisala sensor measure the increase in CO₂ concentration over time due to respiration from within the soil. The gray lines represent the precision of the Vaisala sensor

Figure 7 shows the experimental results. The soil emits a very large concentration of CO₂ and both devices within 130 seconds measure levels up to 900 ppmv of CO₂. While the two sensors were placed as closely together as possible, the TDLAS was still closer to the soil than the Vaisala. Given the high concentration of CO₂ in the soil and the resulting concentration gradient, it is reasonable that the TDLAS measured a slightly faster increase in concentration (138.32 ppmv/min for TDLAS vs. 125.00 ppmv/min for Vaisala). Nevertheless, all concentration measurements are within the precision of the Vaisala. It should also be noted that in the comparison of the calibrated sensors shown in Fig. 6, the two sensors are exposed to similar conditions before the calibrated gas is introduced to the vessel. However in the soil respiration experiment, in which a chamber is placed above soil continuously emitting CO₂, such well-controlled initial conditions are difficult to attain. This can be clearly seen by the discrepancy in the first 30 seconds in Fig. 7, where the TDLAS sensor that is closer to the soil experiences larger fluctuations of the measured concentration. Independently of those small issues observed, the TDLAS sensor shows adequate performance for soil respiration studies.

The TDLAS sensor and the Vaisala sensor were then taken to a field location at Johns Hopkins University for performance comparison of forest floor respiration measurements. For these experiments, the temperature was approximately 22 ^∘C. Several experiments were performed and the change in CO₂ concentration was measured with both sensors simultaneously. The CO₂ levels observed in the field experiment were significantly lower than those in the lab experiment performed with topsoil. To assure that both sensors measured only soil respiration, a black plastic measurement chamber had to be used in order to suppress photosynthesis (i.e. stop CO₂ uptake). A fan was used in the chamber to mix the air so that a comparison could be made between the two sensors. An example measurement is shown in Fig. 8. Since the soil in the field was a weaker source of CO₂, both sensors show similar results despite being in different locations above the soil (the gradient of CO₂ concentration above the soil is smaller). Both sensors show a similar rate of CO₂ concentration increase of 7.01 ppmv/min measured with the TDLAS sensor and 7.22 ppmv/min measured with the Vaisala sensor. The measurements are clearly within the precision specified by the manufacturer for the GMP343 sensor.

Fig. 8

A soil respiration experiment in the field. CO₂ respiration over time is measured by the TDLAS and Vaisala sensors simultaneously. A black chamber is used to suppress photosynthesis. The gray lines represent the precision of the Vaisala sensor

As has been shown, equivalent scientific results of in-lab and in-field measurements can be obtained with both the commercial NDIR technology and the developed TDLAS technology. The main advantages of using the TDLAS sensor are its capability of duty cycling (i.e., accurate readings after only 30 s from turn-on time), modular design, low-power consumption, and autonomy. For example, the Vaisala sensor in low-power mode (i.e., no mirror heating) consumes 1 W. The equivalent power consumption of the TDLAS sensor is 350 mW (if configured as an open-path sensor with no pump). If further power conservation is desired, the sensor can be programmed to sample at regular intervals with low duty cycle and still provide nearly instantly accurate sampling of gas. Such a mode would drastically reduce the power consumption, based on the sleep and wake intervals. For a Vaisala sensor, with a warm-up time of 30 minutes, duty cycling with time intervals shorter than 1 hour becomes impractical. The modularity and autonomy of the TDLAS sensor design allows for better adaptation to a desired application. For instance, because the TDLAS sensor is wirelessly controlled, several wireless sensor nodes can be placed in a network around a base station that has easier access to power. The sensors could then be controlled wirelessly and remotely via an Internet connection, or an opportunistic scheduling of sensor operation could be administered automatically by the base station (e.g., if a rain event is detected, the CO₂ sensors can be activated to monitor the soil respiration triggered by an increase in soil moisture). There would be minimal need for environmentally infringing infrastructure to maintain the sensor nodes over the long term. The scientist would only be required to visit the sensors at intervals needed to replace the batteries, which could be weeks or months, depending on the battery used and duty cycle employed. Because of these advantages, the TDLAS technology is a considerable step forward towards truly autonomous environmental trace-gas monitoring.

4 Summary

We demonstrated a low-power, portable CO₂ sensor for use in atmospheric and soil respiration monitoring. The sensor shows white noise limited performance up to 100 seconds and ultimate minimum detectable absorption of 1.25×10⁻⁶. The 1σ minimum detection limit for CO₂ concentration is 0.14 ppmv in 1 second averaging time and it deteriorates to 0.45 ppmv for averaging times longer than 10,000 seconds due to system drift.

The performance of the TDLAS sensor was compared with the commercial Vaisala GMP343 CO₂ sensor. In the laboratory and field tests focused on soil respiration monitoring the two sensors yielded well correlated measurements in both CO₂ concentration levels, as well as in respiration rates (small discrepancies originated mainly from different geometry and locations of both sensors within the measurement chamber). In summary, the TDLAS sensor was found to have better short-term sensitivity than commercially available sensors based on NDIR technology while having higher sampling rate, approximately the same size, consuming less power, allowing for efficient measurement duty cycling, and having wireless networking capabilities that allow for measurements of larger areas using distributed sensor network configurations. Additionally, while NDIR has excellent performance for CO₂ monitoring applications, the number of different target molecules that could be measured with this technology is limited. TDLAS has an advantage of being a flexible technology that can easily be adapted (via changing the laser source) to other molecular targets such as CH₄, CO, H₂O, NH₃, etc. Further improvements to this TDLAS technology will be focused on algorithms to dynamically account for temperature and pressure changes, as well as on improved TDLAS sensor optomechanical design to minimize thermally induced beam walking and drift. This should significantly improve stability of the instrument in the field applications and provide long-term drifts near or below 1 ppmv of CO₂, which is required for atmospheric sensing applications.