Multi-Frequency Differential Absorption LIDAR (DIAL) System for Aerosol and Cloud Retrievals of CO₂/H₂O and CH₄/H₂O

Abstract

We discuss a remote sensing system that is used to simultaneously detect range-resolved differential absorption LIDAR (light detection and ranging; DIAL) signals and integrated path differential absorption LIDAR signals (IPDA LIDAR) from aerosol targets for ranges up to 22 km. The DIAL/IPDA LIDAR frequency converter consists of an OPO pumped at 1064 nm to produce light at 1.6 μm and operates at 100 Hz pulse repetition frequency. The probe light is free space coupled to a movable platform that contains one transmitter and two receiver telescopes. Hybrid photon counting/current systems increase the dynamic range for detection by two orders of magnitude. Range resolved and column integrated dry-air CO₂ and CH₄ mixing ratios are obtained from line shape fits of CO₂ and CH₄ centered at 1602.2 nm and 1645.5 nm, respectively, and measured at 10 different frequencies over ≈1.3 cm⁻¹ bandwidth. The signal-to-noise ratios (SNRs) of the IPDA LIDAR returns from cloud aerosols approach 1000:1 and the uncertainties in the mixing ratios weighted according to the integrated counts over the cloud segments range from 0.1% to 1%. The range-averaged DIAL mixing ratios are in good agreement with the IPDA LIDAR mixing ratios at the 1% to 2% level for both CO₂ and CH₄. These results can serve as a validation method for future active and passive satellite observational systems.

1. Introduction

Active remote sensing systems based on differential absorption LIDAR (DIAL) [1,2,3,4,5,6] and integrated path differential absorption LIDAR (IPDA LIDAR) [7,8,9,10,11] have been demonstrated and characterized on local scales for their measurement precision and accuracy to study greenhouse gas (GHG) concentrations in the atmosphere. At lower sensitivity, the mixing ratios of the GHGs have to be monitored using ground-based DIAL systems for small-area emission studies where precisions typically range from 1% to 2% [1,2,4,11,12,13]. Higher precision IPDA LIDAR systems based on airborne systems have been used to develop spaceborne platforms and to later calibrate ground-based and spaceborne sensors [9,14,15] at precisions as low as ±0.3% for CO₂ and ±0.5% for CH₄ over 10 s of km ranges and on short time frames (<10 s).

On global scales, however, ground based systems are no match for the wide scale, broadband spectral imaging capacities of passive satellite observational systems that use reflected sunlight from the planet surface to assess the state and dynamics of GHG source and sink fluctuations. Like IPDA LIDAR systems, the accuracy requirements are high for satellite systems that measure the column-averaged concentrations of GHGs because of the small contributions that surface fluxes make to these averages. Consequently, current satellite systems are specified with low uncertainties, <±1 ppm (<±0.25%) for CO₂ and <±10 ppb (<±0.5%) for CH₄. Future LIDAR-based satellite missions [16,17,18,19] are also planned for climate studies on a global scale to measure GHG mixing ratios and, in the case of CH₄, with statistical and systematic uncertainty requirements approaching 2 ppb (±0.1%) [20].

Current DIAL and IPDA LIDAR system development for remote sensing of CO₂ and CH₄ focuses on the near-infrared spectral regions (1572 nm (CO₂), 1602 nm (CO₂), 1645 nm (CH₄), 2050 nm (CO₂)). The main technological challenges are the selection of a suitable laser source and detector, both of which have been the subject of numerous studies for current research systems over the past decade (see the overview given in [11]). The latest developments for DIAL [4] and IPDA LIDAR [9,11,21] systems make use of multiple frequencies to sample the absorption line profile and use subsequent spectral line shape fitting to obtain the GHG mixing ratios. This method is found here to be particularly advantageous for complicated line shapes consisting of multiple underlying transitions (e.g., CH₄) and/or multi-species determinations (e.g., CH₄ and H₂O, denoted CH₄/H₂O or CO₂/H₂O,). Additional advantages have been realized in studies to and through clouds to enhance sensitivity [5,9,22]. For example, previous IPDA LIDAR studies from aircraft [9] measured 30 points across the CO₂ line shape to perform studies to cloud tops and through clouds to the surface at a reported precision of <±0.2% and a bias of <±0.25% in a 1 s average. Cloud targets have also found importance in the specification of the overall spectral purity of the DIAL frequency converter from comparisons of the online and offline frequency returns [23].

The work presented here demonstrates the application of a 10-frequency DIAL system to study GHG mixing ratios of CO₂/H₂O and CH₄/H₂O from aerosol and cloud aerosol returns using a laser based on an optical parametric oscillator (OPO) [22]. We present the data processing and mixing ratios obtained from IPDA LIDAR retrievals to clouds and make comparisons to the column-averaged DIAL mixing ratios to validate the methods. The primary objective of this work is to evaluate the precision and to investigate sources of bias of a standoff platform to simultaneously perform range-resolved DIAL measurements and IPDA LIDAR measurements to cloud targets for ranges up to 20 km and for averaging times of 10 min. One intended application is to perform future measurements in a mobile platform to cross validate the results against the GHG mixing ratios and uncertainties measured at Total Carbon Column Observing Network (TCCON) sites that are used worldwide to provide calibrations of passive satellite systems. The results we present here for cloud returns achieve the targeted precision of ±0.5% for dry-air mixing ratios of CH₄ but only partially achieve the goal of ±0.25% uncertainty for dry-air CO₂. We discuss additional methods for improving the measurement precision and accuracy that we expect are needed to reach these detection limits for both GHGs.

This paper is structured as follows: Section 2.1 presents the experimental 10-frequency OPO transmitter and movable transceiver platform used to perform DIAL and IPDA LIDAR studies of CO₂/H₂O and CH₄/H₂O. Section 2.2 discusses the DIAL and IPDA LIDAR data processing including the line shape fitting procedure to obtain GHG concentrations and mixing ratios. Section 2.3 reviews the latest spectroscopic improvements to the CH₄ line shape near 1645 nm. Section 3 presents the data processing for IPDA LIDAR retrievals from diffuse clouds (3.1), the IPDA LIDAR results for CO₂/H₂O and CH₄/H₂O (3.2), the DIAL results for CO₂/H₂O and CH₄/H₂O (3.3), and a comparison of the mixing ratios obtained from the DIAL and IPDA LIDAR retrievals. A discussion of the uncertainties relative to satellite calibration requirements and the possible sources of bias observed in the CH₄ line shapes is given in Section 4. The paper closes with conclusions and outlook in Section 5.

2. Experiment

2.1. Combined DIAL and IPDA LIDAR Hardware Description and Line Selection Criteria

A schematic of the overall system setup at NIST (Boulder, CO, USA) is shown in Figure 1 and the system specifications are given in

Table 1. Specifications of the OPO based DIAL/IPDA LIDAR system ^a [4,24].

Frequency Converter	Parameters	Description/Manufacturer
Wavelength coverage	1600 nm to 1647 nm	New Focus, ECDL
Number of frequencies	10	Spectral coverage ≈ 40 GHz
PRF, SRF	100 Hz, 10 Hz	-
Microwave Synthesizer	20 GHz bandwidth	Anritsu, MG37022A
Pulse energy	5 mJ signal wave	Ep ≈ 170 mJ, Coherent Infinity
Pulse length (FWHM)	<3 ns	Pump pulse ≈ 3 ns FWHM
Spectral linewidth (FWHM)	190 MHz	Nearly transform limited
Two OPO KTA Crystals	10 × 10 × 15 mm3, θ = 67.4°, φ = 0°	Optical Platz, Inc. (NM, USA)
OPO	RISTRA, twisted rectangle	AS Photonics, Inc. (Albuquerque, MN, USA)
Telescopes and detectors
Tx beam diameter, divergence	254 mm, <300 to 400 μrad	Ritchey-Chrétien, Orion
NF Rx. diameter, FOV	279.4 mm, 3 mrad	Schmidt-Cassegrain, Celestron
FF Rx. diameter, FOV	406.5 mm, 2 mrad	Schmidt-Cassegrain, Meade
Bandpass filter, 1645.5 nm	1.9 nm (FWHM), T > 90%	>5 OD rejection, Alluxa
Bandpass filter, 1602.2 nm	2.0 nm (FWHM), T > 90%	>5 OD rejection, Materion
Neutral density filter	0.1, 0.3, 0.6, and 1 OD	Thorlabs
Rx to PMT fiber core diameters	1 mm	1 m cable length
NF PMT	~3% QE	Hamamatsu H12397-75
FF PMT	~8% QE	Hamamatsu H12397-75 SEL
Data acquisition
Transimpedance amp	5 kV/A, τ3dB ≈ 1 ns	Femto HCA-400M-5K-C
Digitizer	8 bits, 2 GS/s, 2 channels	GaGe CobraMax CSE24G8
Raw data storage	10 s average	NetCDF4 file format
Range bin resolution	250 m to 500 m	Defined in post-processing
Temporal resolution	10 min	DIAL concentration average

^a NF Rx: near-field receiver telescope; FF Rx: far-field receiver telescope; FOV: field of view; OPO: optical parametric oscillator; PRF: pulse repetition frequency; SRF: scan repetition frequency; E_p: 1064 nm pump pulse energy; FWHM: full width at half maximum; PMT: photomultiplier tube; QE: quantum efficiency; T: transmittance; OD: optical density; Tx: transmitter; Rx: receiver.

[24]. The frequency converter that includes the optical parametric oscillator (OPO) seed system is nearly identical to the system described in refs. [4,11]. Briefly, we sequentially generate 10 seed frequencies of high spectral purity (>99.9%) over a spectral bandwidth of ≈40 GHz using a filter cavity and a broadband electro-optic phase modulator driven by a fast scan microwave (MW) synthesizer. The different frequencies are generated at the 100 Hz pulse repetition frequency (PRF) of the injection-seeded 1064 nm Nd:YAG pump laser. A PZT in the OPO cavity is controlled by the Pound–Drever–Hall lock signal (PDH lock method [25]) to rapidly bring each seed frequency into cavity resonance prior to the arrival of the pump pulse. The frequency sequence is continuously repeated every 100 ms to give a 10 Hz overall scan repetition frequency (SRF). The KTA crystals of the OPO are cut so that the signal wave can be angle-tuned for type I phase matching over the spectral region from 1600 nm to 1650 nm. The angles of the two crystals were adjusted for the optimal signal wave pulse energies and flatness over the 40 GHz spectral bandwidth. To ensure eye-safe operation, the signal wave is separated from the pump and idler waves using a prism beam separator containing four SF₆ 60° prisms in a folded wavelength-compensating-path configuration. In the two spectral regions investigated in this work (near 1602 nm and 1645 nm), the signal pulse energies generated were ≈5 mJ at a pump energy of ≈170 mJ and have measured nearly transform-limited linewidths of ≈190 MHz [4].

The remaining aspects of the system that includes the movable platform (91 cm × 152 cm) weighing more than 300 kg are new to this work and shown in Figure 1. The platform pivots (about the tilt axis) along the front edge near the laboratory exit port to provide a 0° to 30° of inclination using a pneumatic cylinder equipped with an internal slide wire for position encoding. The encoder signal is used for active feedback to a proportional–integral–differential (PID) controller of the air pressure to the cylinder. A stepper motor stage is used for lateral rotation (about the rotation axis) of the platform by ±30°. The signal wave output is propagated along a Coudé path that uses two galvanometers, Galvo-X and Galvo-Y, for X and Y beam steering, respectively, (where X is parallel, and Y is normal to the platform surface) and a quadrant photodetector for locking the position of the transmit beam as the platform is moved. After reflecting from the Galvo-X and Galvo-Y mirrors, the beam is reflected along the center axis of the rotation stage and then to a 45° mirror located on the center axis and fixed to the platform. This configuration keeps the X position of the transmit beam fixed as the platform is rotated. The beam is then directed to a 254 mm expanding telescope (Tx) for Class I level transmission with a beam divergence of <400 μrad. To keep the pointing direction of the transmit beam fixed along the Y direction as the platform is tilted, the rotation axis of Galvo-Y is aligned colinear with the tilt axis of the platform. The encoded signal from the cylinder is gain-corrected and used as the input signal (feedforward) to Galvo-Y to keep of the transmit beam parallel to the platform surface as it is tilted. To remove the remaining pointing errors in the galvanometers’ beam steering relative to the platform motion, a small portion of the transmit beam illuminates the quadrant photodetector mounted on the platform. The detector’s differential error signal in the Y direction is electronically added to the Galvo-Y input signal for fine level feedback to lock the transmit beam to the center of the detector along Y. Likewise, the differential error signal in the X direction is used as input to Galvo-X to lock the transmit beam to the detector’s center along the X direction.

Two receiver telescopes (Rx) are positioned on either side of Tx and consist of 279 mm near-field (NF-Rx) and 406 mm far-field (FF-Rx) telescopes having fields of view (FOV) of 3 mrad and 2 mrad, respectively (only the FF data are discussed in these studies). Both are equipped with altitude/azimuth mount for alignment with the fixed transmitted beam. The collected light is coupled into 1 mm multi-mode fibers (MMF) and then free-space-coupled to near-IR photomultiplier tubes (PMT_A and PMT_B) housed in a temperature-cooled box. Optical bandpass filters inserted between the MMF and the PMT are centered at 1602.2 nm for CO₂/H₂O and 1645.5 nm for CH₄/H₂O and have peak transmissions of >90% and a full width at half maximum (FWHM) bandpass of 2 nm. To aid in reducing background light during daytime viewing, filter wheels are available with optical densities (OD) from 0 to 3. The NF and FF channels are amplified with a transimpedance gain of 5 kV/A (AMP) and digitized separately at 2 × 10⁹ samples (2GS) per second (0.5 ns/point) with 8 bits of resolution (ADC). The digitized signals are processed in real time to (i) discriminate and photon count the individual pulses (Δt_FWHM ≈ 3.5 ns) and (ii) accumulate the digital values (which are proportional to the PMT current) for each receiver. The counts and current were stored separately for each transmitted frequency and receiver resulting in 40 data segments that are stored every 10 s in Network Common Data Form file format (NetCDF). The range for each retrieval was truncated at 48 km.

Two spectral regions targeted for these studies are illustrated in Figure 2a,b and consist of the R(18e) CO₂ line of the 30013-0001 band at 1602.205 nm (6241.40 cm⁻¹) and the R(6) manifold of the 2ν₃ band of CH₄ centered at 1645.534 nm (6077.05 cm⁻¹). The choice for the line selection is based on several factors. First, the CO₂ and CH₄ lines selected are among the strongest transitions in the bands over the temperature range of these studies (320 K to 285 K). The temperature sensitivity of the CO₂ and CH₄ lines causes an ≈11% and ≈13% increase in the optical depths, respectively, over this high-to-low temperature range. Second, for CH₄ at atmospheric conditions, all transitions in the 2ν₃ band consist of a series of unresolved pressure broadened rovibrational lines. The multiplet arises from Coriolis coupling which severely complicates the dependence of the line shape function and integrated intensity on temperature, pressure, line mixing and speed-dependent effects. Given the importance of this manifold to the MERLIN mission [16], high-accuracy cavity ring down studies have recently been performed on the R(6) manifold [26,27] to determine the line-mixing and speed-dependent parameters over the wide range of atmospherically relevant conditions. As we discuss below, these refined parameters are necessary to minimize the residuals in the fits. Third, both selected regions in Figure 2 contain a nearby water vapor line or lines. Dry-air mixing ratios derived from [X]_dry = [X]/(1 − 0.01·[H₂O]), where X = CO₂ or CH₄ and [H₂O] is percent mole fraction (PCM), are the universally comparable data products needed from atmospheric GHG studies. For relative humidities from 20% to 50%, the dry air correction can increase the dry air concentrations between 0.6% and 1.5%, so the water vapor concentrations are needed to remove this bias.

For the CO₂/H₂O region shown in Figure 2a, five frequencies are selected on the CO₂ line and two frequencies on the H₂O line pair at 1602.133 nm and 1602.102 nm (6241.68 cm⁻¹ and 6241.81 cm⁻¹), and the remaining 3 frequencies sample the baseline regions at both ends of the spectrum. For the CH₄/H₂O region shown in Figure 2b, six frequencies sample across the doublet structure of the CH₄ manifold, one frequency on the H₂O line at 1645.471 nm (6077.28 cm⁻¹) and the remaining three frequencies sample the baseline regions on both ends. Over a range of 250 m (500 m path length), the optical densities of CO₂ and CH₄ are near 4% for ambient levels of CO₂ (≈420 ppm) and CH₄ (≈2 ppm).

The choice of frequencies chosen to sample the CO₂ and CH₄ line shapes is based on the following reasons. In general, the best SNR is expected for frequencies that sample across the maximum difference in absorbance between the online and offline regions which is analogous to 2-frequency DIAL. Therefore, maintaining this large absorbance difference in multi-frequency DIAL requires localizing the additional frequencies in these two regions. An added advantage of the multi-frequency approach is the ability to select more online vs. offline frequencies to offset the lower count rates of the online absorption to improve the SNR of the line shape measurement. The multiple offline frequencies that sample regions on both sides of the line help to minimize frequency-dependent effects in the detection system that impact the baseline absorption profile. Moreover, multiple frequencies that sample near the peak of the absorption enable the accurate determination of the line center frequencies from least square fits. This determination is particularly important in our strongly pumped OPO system to remove the offset and drifts associated with the seed–signal mismatch [28] that has been shown to range from 100 to 150 MHz in our system [4]. Finally, while adding more frequencies to the scan may be useful to better evaluate line shapes and even the temperature and pressure (T&P) profiles in the retrievals, there is a tradeoff between the resulting slower scan rate and SNR. These tradeoffs are currently under investigation.

2.2. Raw Data Processing and Concentration Fitting Procedure

The pre-processing of the raw data follows the procedure discussed previously [4]. Briefly, the hybrid photon counter/current detection system consists of a dual-channel 8-bit digitizer that samples the amplified output voltages of the PMTs at a rate of 2 GS/s. This is fast enough to resolve the individual photons detected at the PMT with 7 samples across the full width of each count pulse. Double counting is prevented by skipping 6 samples following threshold detection. In software (JB64, v6.05.09), the accumulated photon counts and the direct sum over the digital signals which is proportional to the PMT current are saved in separate records on 10 s intervals for both the NF and FF receivers and each frequency in NetCDF format.

For each of the 10 frequencies, the processing begins with the accumulation of the photon counts and the current signals over 10 consecutive ADC samples which reduces the range resolution to ≈1.5 m. Then, the 10 min average intervals of counts and current are determined for each frequency by averaging over 60 consecutive 10 s intervals. To splice together the current and counts intervals, a series of non-linear least square fits (NLLSFs) are first performed to determine the saturation parameters to correct the counts for pulse pileup error [29]. The individual parameters are typically fit over a region where the count rates range between N = (1 − 5) × 10⁶ counts/s. The interval-averaged saturation parameter at each frequency is then fixed and the NLLSF repeated to fit the current to the count intervals using the best fit scalar and offset parameters for each interval. This hybrid detection scheme extends the linearity of the photon counting system by two orders of magnitude. These same saturation parameters at each frequency are used for the fitting of splices over the enhanced backscatter regions from cloud returns when the saturated count threshold levels are exceeded.

To fit the concentrations of the DIAL and IPDA LIDAR profiles at each altitude (i.e., at each T&P), we first generate line shape templates for CO₂ and CH₄ at 400 ppm and 2 ppm, respectively. Given that the CO₂ line is a single transition and that we use only 5 online frequencies near the peak of the absorption to measure the line, a higher accuracy line shape function that includes line mixing and speed dependent effects was not necessary [30]. Therefore, the template line shape function used was a Voigt profile based on the line broadening and shift parameters reported in the HITRAN 2020 database [31]. However, for CH₄, the most accurate line shape function was required and discussed separately below. Voigt functions for the nearby water vapor lines were added to the overall profile at the 1% (10,000 ppm) level.

For the fits of the range-resolved DIAL concentrations, the raw data are processed at the 10 frequencies using a numerical procedure described in [4]. The differential absorption optical depth, DAOD(R,ν_i), at each of the 10 frequencies, ν_i, is first determined from the integrated photon count data, N_i, at range R = (R₁ + R₂)/2 with range resolution ΔR = (R₂ − R₁) from

$$DAOD\left(R,{\nu }_i\right)=ln\left(\frac{N_i\left(R_1\right)N_b\left(R_2\right)}{N_b\left(R_1\right)N_i\left(R_2\right)}\right)$$

(1)

where N_b is the one of the background frequencies chosen as the first frequency (i.e., $DAOD\left(R,{\nu }_1\right)$ = 0) and R = 0 is where the pulse leaves the transmitter telescope. Determinations of the DAOD(R,ν_i) using the other offline frequencies gave results within the reported uncertainties of the concentrations. From our prior DIAL study of CO₂ [4], the OPO signal-wave frequency is known to be shifted by 100 MHz to 150 MHz relative to the input seed frequency. To account for this shift prior to performing fits of the DIAL concentrations, the frequency offset of each 10 min interval was determined from fits to the averaged profile of the first 6 range bins. Minor adjustments to this offset were made based on the small model-based pressure shift that results from changes in altitude with range.

The final fits of the optical densities at the 10 frequencies were obtained by varying a baseline offset, slope and the scaling factors of the line shape templates. Because of the much higher SNR of the IPDA LIDAR data, the frequency offset was simultaneously varied in the fit of each profile. All template components of a given species were scaled together to give concentrations in μmol/mol (ppm) for CO₂ and CH₄, and in percent mol/mol (PCM) for water vapor. We also note that the linewidth of the laser (190 MHz) was added in quadrature to the Gaussian component of the profile and the T&P used at each altitude were derived from the U.S. Standard Atmosphere model [32] after correction for the weather station data at the base altitude of the transmitter in Boulder (1655 m). Further details of the IPDA LIDAR processing procedure are given in Section 3.2.

2.3. Methane Line Shape Profile

The line shape function of the R(6) manifold of the 2ν₃ band of CH₄ is complicated given that it is an unresolved profile under atmospheric conditions primarily composed of the overlap of 6 individual pressure broadened lines. This complexity has likely contributed to the significant changes in the cross sections and Voigt line shape parameters reported in the HITRAN databases over the past 20 years. Moreover, the line shape is further complicated by line-mixing and speed-dependent effects which cannot be modeled by a simple Voigt profile used in HITRAN. Given the importance of the R(6) line for the upcoming active satellite IPDA measurements from MERLIN [16,17], several cavity ringdown studies have been performed to (i) measure the precise line centers of the 6 lines using saturation spectroscopy [33,34] and (ii) fit the T&P dependence of these transitions [26,27] to the Hartmann–Tran profile (HTP) [35] that includes line-mixing and speed-dependent effects. As a result of the high sensitivity of the IPDA LIDAR returns, we found it necessary to make use of this model to minimize the residuals. It is further noted that the overall integrated intensity of the HTP R(6) profile is scaled to that of HITRAN 2020 [27] and that additional studies are underway at NIST to link the intensity scale to the International System of Units (SI).

3. Results

3.1. IPDA LIDAR Studies of Signal Returns from Diffuse Clouds

The measurements of the CO₂/H₂O region were performed over the 5 h nighttime period on 24 November 2021 beginning at 8:08 PM local time. The telescope was at an inclination of 2.5° above the horizon and pointed ≈−10° off North across the city of Boulder. The measurements of the CH₄/H₂O region were performed over the 5 h nighttime period on 28 April 2022, beginning at 10:05 PM local time. The telescope was at an inclination of 10.5° and pointed ≈+2.4° off north across the city of Boulder.

The logarithm (base 10) of the range- and background-corrected LIDAR map for the first offline frequency (ν₁) in the CO₂ region is shown for the full measurement period in Figure 3b. Cross-sections of the LIDAR returns are shown in Figure 3a at each of the 10 frequencies for the first 10 min interval. In the first 2 h period, normal aerosol returns extend to about 10 km, at which point the scattering is significantly enhanced upon passing through a ≈2 km cloud segment centered at ranges from 10 km to 16 km. The offline cloud signal returns (shown in blue shades) exceed 200 × 10⁶ count/s and are of the same magnitude as seen near the peak of the signal return near 0.5 km. The inset in Figure 3a shows an expanded region of the first 10 min interval over the cloud segment. The signal returns are plotted on a logarithmic scale to enhance their separation at each frequency. From a comparison of the online (orange/red shades) and offline (blue shades) returns in the cloud region, the CO₂ absorption signal over this 20+ km pathlength represents ≈85% fractional absorption.

The logarithm (base 10) of the range- and background-corrected LIDAR map for the first offline frequency (ν₁) in the CH₄ region is shown for the full measurement period in Figure 3d and a cross section cut of the raw returns at all 10 frequencies is shown in Figure 3c for ranges up to 22 km. The enhanced backscatter from the more diffuse clouds extends over a 6 km range beginning near 15 km and is 50 times weaker than the CO₂ cloud region shown in Figure 3a. The inset in Figure 3c shows an expanded region of the first 10 min interval over the cloud segment. Again, the signal returns are plotted on a logarithmic scale to enhance their separation at each frequency. With the pathlength being nearly double that of the CO₂ region, the online fractional absorption in the CH₄ cloud region is ≈95% of the offline returns. Furthermore, the larger returns for the CO₂ region are above the saturated count threshold and required the use of the PMT’s current channel, while the returns in the CH₄ region remained linear in photon counts. Therefore, these two studies represent near limiting cases for backscatter studies from clouds.

The range dependence of the optical depth gives a better representation of the signal-to-noise ratio of the returns. The natural logarithm of the ratios for each frequency, k, relative to the first offline color gives the optical depth, OD(R), as a function of range, R, from

$${OD}_k\left(R\right)=\mathrm{l}\mathrm{n}\left(\frac{N_k\left(R\right)}{N_1\left(R\right)}\right)$$

(2)

where $N_k\left(R\right)$ are the accumulated counts in each 10 min interval. The ${OD}_k\left(R\right)$ values for the nine ratios are shown in Figure 4a,b for the CO₂/H₂O and CH₄/H₂O regions, respectively. The near linear increases observed for the online ratios in both regions result from the following reasons. First, the returns represent the path average over the column, so deviations from linearity caused by changes in the T&P profiles with altitude will be reduced in the average over the line shapes. In general, for an isolated pressure broadened transition in the atmosphere at constant mixing ratio, the peak line intensity (OD) remains unchanged with decreasing pressure (i.e., with increasing altitude) since the line width scales linearly with pressure. Furthermore, temperature also decreases with increasing altitude, and for the selected CO₂ and CH₄ lines in this work, the shifts in the partition functions increase their intensity with altitude. For the five online CO₂ frequencies, non-linearities in ${OD}_k\left(R\right)$ from temperature changes (<5 K obtained from Appendix A, Figure A2a) are small because of the nearly horizontal path used in this study (i.e., a 0.44 km change in altitude at a 10 km range). For the six overlapping lines in the CH₄ manifold, the narrowing of each width leads to an overall intensity decrease across the line profile. This decrease in ${OD}_k\left(R\right)$ is, however, approximately offset by the decreasing temperature of ≈−35 K over a 3.5 km altitude change (see Appendix A, Figure A2a).

Also apparent in Figure 4a,b, the SNR decreases with range since the accumulated counts decrease by the inverse square of the range (i.e., SNR(R) ≈ $\sqrt{N_k/R^2}$ in a uniform aerosol field). However, beyond 10 km in Figure 4a, the SNR increases significantly from the enhanced backscatter from the cloud aerosol returns. Furthermore, the near linear increase in ${OD}_k\left(R\right)$ across the cloud region indicates sufficient transparency to perform integrated path measurements. A similar trend is observed in Figure 4b for CH₄/H₂O, although the enhancement in SNR from cloud aerosols is not as pronounced because the 2-fold increase in range results in significant decreases in the total backscatter signals.

In contrast to DIAL processing that is laser-energy-independent (see Equations (1) and (2)), the IPDA LIDAR processing of the integrated cloud returns requires knowledge of the relative energy in each pulse. The integrated path optical depth, ${IPOD}_k\left(R\right)$, of point k can be expressed as

$${IPOD}_k\left(R\right)=\mathrm{l}\mathrm{n}\left(\frac{N_1\left(0\right)}{N_k\left(0\right)}\right)+\mathrm{l}\mathrm{n}\left(\frac{N_k\left(R\right)}{N_1\left(R\right)}\right)$$

(3)

where the first term on the right represents the log ratio of the transmitted laser pulse energies (LPE). The different log ratios of the LPEs are the reason for the offset curves shown in Figure 4a,b and must be accurately evaluated before the ${IPOD}_k\left(R\right)$ can be used in the line shape fits of concentration. Typically, for IPDA LIDAR studies, the LPEs are measured independently in a reference channel by sampling the output beam before expansion and transmission [11,36]. However, unlike the output from single-mode fiber-based sources [11], the position of our free-space OPO beam depends slightly on frequency because of small changes in the PZT mirror position at each cavity resonance condition. Because of the secondary mirror obscuration in the beam-expanding telescope, the transmission efficiency was found to depend slightly on these beam shifts relative to the energies measured prior to the telescope using an integrating sphere and photodetector. Consequently, the energy ratios of the signal and reference channels at the different frequencies were not sufficiently proportional for normalization. Other factors such as the imperfectly matched bandpass filters and differences in the detectors’ frequency response functions also contributed to the observed variations in these ratios. As previously discussed [36], careful designs and detailed calibrations are required to proportionally match the relative energies in the signal and reference channels.

The simultaneous measurement of DIAL and IPDA LIDAR returns has enabled us to use an alternative method to determine the received LPEs that include the transmitter’s, detector’s, and bandpass filter’s frequency response functions. The method takes advantage of the linear range dependence of the optical depths following small corrections for changes in the range resolved concentrations and altitude of the data shown in Figure 4a,b. From Equation (3), the received LPEs can be obtained from the linear fits of these corrected data. By extrapolating the curves back to when the laser (OPO) pulse leaves the transmitter (R = 0), the received LPE terms in Equation (3) are obtained directly. The curve fits are performed over the region from 0.625 km to 5 km (see vertical dashed lines in Figure 4a,b) where the intercepts at R = 0 represent the received LPEs for each 10 min average. The uncertainties of the intercepts are typically <0.01 in optical density and therefore have a minor impact on the fitted concentrations. The early portions of the curve fits are shown with solid black lines from R = 0 to R = 0.625 km where full overlap begins.

3.2. IPDA LIDAR Studies of CO₂/H₂O and CH₄/H₂O

Unlike hard-target backscatter where the integrations are effectively performed over the width of the laser pulse, the diffuse nature of cloud aerosols leads to an extended source of the enhanced backscatter that ranges from 2 km for the CO₂/H₂O region to > 6 km ranges for the CH₄/H₂O region. The inclination angle for the CO₂/H₂O study was 2.5° which corresponds to clouds at an altitude of ≈0.44 km at a range of 10 km from the transmitter. The inclination angle for the CH₄/H₂O study was 10.5°, which corresponds to clouds at an altitude of ≈3.5 km at a distance of 19 km from the transmitter. Since the model line shape function depends on the T&P profile, it will change with altitude over the extended region of the cloud. As a result, separate calculations are required over multiple and contiguous range bin slices to determine the average model line shape to fit to obtain the path integrated concentrations.

The path integrated concentrations to cloud targets were determined on 10 min intervals, and fits were performed over consecutive 250 m slices. Furthermore, because of variations of T&P over the elevated path particularly in the CH₄ study, the model line shape used in the fits for each slice was calculated by dividing the path length into five sections and performing an average over the predicted line shapes for each of these sections. For the CH₄ study, each section corresponds to an atmospheric layer 0.7 km in altitude. The line shape for each layer was calculated at the T&P predicted from the U.S. Standard Atmosphere model following correction for the base altitude. The parameters that varied in the NLLSF included the frequency offset, baseline offset, slope and the concentration scalars of CO₂/H₂O or CH₄/H₂O where all lines of the given species were scaled together as a group. This approach is very similar to the multi-frequency IPDA LIDAR method described in ref. [37] to extract column-averaged dry-air CO₂ concentrations to cloud tops and to the surface from near-nadir airborne measurements. One exception is that we perform our own layered profile calculations that include the HTP line profile for CH₄ and the HITRAN 2020 database for CO₂ rather than using the line-by-line radiative transfer model [38] and the HITRAN 2008 database for CO₂ used in the airborne studies.

The series of best-fit line shapes to the absorption data over the CO₂/H₂O and CH₄/H₂O regions are shown on 250 m intervals in Figure 5a,b, respectively. The absorption data were obtained for a single 10 min interval near the beginning of both measurement periods after processing using Equation (3). The fit residuals shown in the sub-panels increase with range as expected for a ≈1/R decrease in SNR. However, in the cloud aerosol region of CO₂/H₂O between 12 km and 13.5 km, the SNR improves significantly, leading to an obvious reduction in the scatter of the residuals. The improvement in the corresponding cloud aerosol region of CH₄/H₂O between 18 km and 21 km is not as apparent because of the much weaker signal returns.

More detailed views of the fits and residuals of Figure 5a,b are illustrated in Figure 6a–d at both a short range from normal aerosol returns (1.5 km to 1.75 km) and at a long range from cloud aerosol returns (near 12.5 km for CO₂ and 18.75 km for CH₄). In both short and long range, the relative magnitude of the residual scatter for CO₂/H₂O is on the order of 1% relative to the peak online optical density. For the CH₄/H₂O region, residuals are again near 1% at short ranges but increase to ≈3% at long ranges. This increase may arise from the very low online count rates at 19 km (see Figure 3c). However, some systematic structure appears to remain in the residuals. The possible sources of bias in the measurements are discussed in Section 4. We note that relative to CO₂, the CH₄ cloud region was measured at more than eight times the altitude (0.43 km vs. 3.5 km), resulting in significant gradients in the T&P profile over the range. The model line shape narrows at this altitude and contributions to the average line shape lead to a more well-resolved doublet profile as the data in Figure 6d reveal.

The enhanced backscatter from the extended cloud aerosol regions affords an opportunity to reduce the measurement uncertainties of the column-averaged mixing ratios. The column integrated concentrations may be obtained from the mean over all slices, ${\overline{x}}_{wtd}$, each weighted by the integrated counts for that slice relative to the total returns (i.e., total weight). The uncertainties, $s_{wtd}$, can then be evaluated from the weighted standard deviation of the mean of the slices as given by

$${\overline{x}}_{wtd}=\frac{{\sum }_{i=1}^n{w}_i{x}_i}{{\sum }_{i=1}^n{w}_i}$$

(4)

$${var\left(x\right)}_{wtd}=\left[\frac{{\sum }_{i=1}^n{w}_i{x}_i^2}{{\sum }_{i=1}^n{w}_i}-{\left({\overline{x}}_{wtd}\right)}^2\right]\frac{n}{n-1}$$

(5)

$$s_{wtd}=\sqrt{\frac{{var\left(x\right)}_{wtd}}{n}}$$

(6)

where $w_i$ are the weights of the individual measurements, $x_i$, n is the number of slices and ${var\left(x\right)}_{wtd}$ is the weighted variance.

The averaging procedure has been applied over a large enough range to capture the cloud features in both regions. Signals below a certain threshold in the slices were excluded in the weighted average. The weighted dry-air mixing ratios are shown for 10 min intervals in Figure 7a,b for the CO₂/H₂O and CH₄/H₂O regions, respectively.

For dry-air CO₂ mixing ratios, the uncertainties range from about ±0.2% to ±1% over the initial 3+ h measurement period. Interestingly, while the first 120 min showed enhanced backscatter from the clouds and therefore improved residuals, the weighted averages show little in the way of improvement relative to the cloudless regions after the first 120 min. We attribute the lack of improvement to difficulties in splicing the current to the counts for the signal returns that vary significantly over the initial 10 km range (see Figure 3b) and are near the upper dynamic range limit of the PMT as a current source. In contrast, the weighted mixing ratios for CH₄ in Figure 7b show substantial improvement in the initial 2.5 h when clouds were present relative to the final 2.5 h when they were absent or too distant. We find the relative uncertainties over the cloud period range from ±0.1% to ±0.5% for CH₄ and increase to 2% to 3% in the cloudless period.

3.3. DIAL Studies of CO₂/H₂O and CH₄/H₂O and Comparison with the IPDA LIDAR Results

The range-resolved dry-air concentration maps for CO₂ and H₂O obtained from the DIAL fits using Equation (1) are shown in Figure 8a,b, respectively, and those for CH₄ and H₂O are shown in Figure 8d,e, respectively. The data are shown on 10 min intervals but have an effective temporal resolution of ≈20 min because of additional surface smoothing. Because of technical issues, the results are shown only from the far-field receiver where full overlap begins near 0.625 km. In both cases, the range resolution is 250 m beginning at 0.75 km and is linearly increased to a range resolution of ≈1.5 km at a range of 9.25 km to help offset the ≈1/R drop in SNR. The corresponding range- and background-corrected LIDAR maps are shown for the first offline frequency (ν₁) in the lower panels on a logarithmic scale to enhance the variations in the aerosol backscatter over the range. The regions with an absence of color are below the detection sensitivity.

For the CO₂/H₂O region, the inclination angle is only 2.5° and, at 1.25 km, the beam path is only ≈50 m above two emission stacks from a natural gas burning utility plant on the University of Colorado campus that was running at full capacity this time of year. As evident in Figure 8a near this range, particularly after the first hour of the measurement period, additional contributions from the plant’s plume emission result in a (10–30) ppm increase above the background levels of ≈425 ppm. From NOAA wind LIDAR data, the initial increase is likely a result of a change in wind speed and direction from <3 m/s SE in the first hour to (8–10) m/s S for the remaining 4 h. Furthermore, the DIAL data suggest the increase in the column-averaged CO₂ concentrations in Figure 7a that also occur in the first hour may arise, in part, from a stronger plume addition to the column average. Also apparent in Figure 8c at this range is a sharp and distinct increase (spike) in the LIDAR signals from the plume condensate. A recently reported DIAL study of ours using a fiber amplifier system made use of the spike to measure the emission flux from this utility plant [13]. We also note, in Figure 8c, the gradual decrease in the LIDAR signal intensity after the first 3 h which reduces the SNR at the longer ranges. This led to some missing data in regions beyond 7 km as indicated by the absence of color in Figure 8a.

For the CH₄/H₂O region, the inclination angle is 10.5° and, at a range of 9.2 km, the relative altitude above the transmitter is 1.69 km. The CH₄ concentrations in the first 4 km appear to be slightly more uniform relative to the higher variations (1.8 ppm to 2.1 ppm) that appear at longer ranges. In some cases, the bands of enhanced concentration appear to extend across the 5 h measurement period. The reasons for these variations are not clear. We note, however, that the CH₄ measurements made with the cavity ringdown point sensor located near the transmitter also displayed an unusual broad variation of CH₄ concentrations that ranged from 2.0 ppm to 2.4 ppm during this period (see Figure A3b). While some transients are common (e.g., see Figure A3a for CO₂) and likely arise from a nearby power plant located on the Boulder NIST campus; the observed variations on this day far exceed those typically observed.

A cross check of the DIAL concentration maps can be obtained through comparisons with the IPDA LIDAR results reported in Section 3.2. Given that all data were acquired with the same transceiver system, filter and detector, and that the LPEs are not needed in the DIAL retrievals, differences in the column average mixing ratios from these two independent techniques can provide insight into any systematic errors associated in the data processing and/or assumptions in the models to retrieve concentration. For this, the mixing ratios of the range-averaged DIAL (RA-DIAL) data displayed in Figure 8a,b are needed. In contrast to the weighted mean method performed over multiple slices in the cloud region, the range resolved mixing ratios need to be weighted equally and the overall uncertainties must be added in quadrature. For a fixed range bin size of 250 m, the uncertainties would be dominated by the largest uncertainties associated with the weakest returns at long range. As discussed above, the range bin size was linearly increased to largely “equalize” these uncertainties over range. It is further noted that this comparison is not complete in the sense that the DIAL mixing ratios are missing in the first 0.75 km and, for the CH₄/H₂O region, the “gap” between the DIAL data and the IPDA LIDAR data is too large. Nevertheless, barring any unknown sources and sinks in these regions, the path-averaged values should largely agree with one another.

The comparison of the dry-air mixing ratios of the range-averaged DIAL (RA-DIAL) and the IPDA LIDAR measurements are shown in Figure 9a,b for CO₂ and CH₄, respectively. At most of the 10 min intervals, the RA-DIAL uncertainties for CO₂ and CH₄ range from ±1% to ±2% which are 5- to 10-fold larger than those from the IPDA LIDAR data. More importantly, IPDA LIDAR data are well-centered on the RA-DIAL mixing ratios for most of the measurement periods with perhaps a slight bias for lower IPDA LIDAR values at a few points. Nevertheless, for nearly every interval, the corresponding uncertainties are well within those of the RA-DIAL values.

4. Discussion

The accuracy requirements for satellite observing systems that measure the column-averaged dry-air concentrations of GHG over paths to and from the planet surface are quite severe because even sizable contributions from source and sink fluxes near the surface make minor changes to the column average. For this reason, passive satellites such as GOSAT, MethaneSAT, OCO-2, etc., typically have targeted uncertainties of <±1 ppm (<±0.25%) for CO₂ and <±10 ppb (<±0.5%) for CH₄. To reduce the uncertainties to such levels requires a worldwide network of ground-based calibration sites known as the Total Carbon Column Observing Network (TCCON). Like passive satellites, TCCON uses sunlight, typically a Fourier transform infrared spectrometer and spectral fitting software to obtain dry-air mixing ratios for fly-over satellite calibrations. These sites, in turn, require the use of in situ World Meteorological Organization (WMO) calibrated instruments on board aircraft as well as other bias corrections for the site’s certified validation. Moreover, for the future MERLIN mission, the accuracy requirements are even further reduced to near 2 ppb (±0.1%) for the active sensing of CH₄ [20]. To complement the TCCON network, one intended use of our ground-based approach is to provide a calibration-free platform for passive and active satellites by developing a new portable 200 Hz OPO-based system in a mobile platform for vertical and slant path range-resolved DIAL and IPDA LIDAR measurements of clouds. While satellite retrievals depend on a cloud-free field-of-view (FOV), we expect to operate the IPDA LIDAR system on days when some clouds are present to enable retrievals over a near coincident path with the satellite FOV.

The measurement uncertainties of the IPDA LIDAR mixing ratios from cloud aerosol retrievals shown in Figure 7 represent two examples of the currently measured precision of this approach and reflect partial success in the achieving our targeted goals. For the dry-air CO₂ mixing ratios in Figure 7a (note that the y-axis tics in both panels represent ≈1% uncertainty), only roughly a third of the measurements are at or below the ±0.25% level. However, for dry-air CH₄ mixing ratios, all measurements within the first 2.5 h when clouds were present are at or below ±0.5% uncertainty and some even approach ±0.1%.

One of the primary goals in these ground-based spectroscopic studies is to assess possible biases in the method that would lead to increases in these uncertainties for the absolute mixing ratios. As mentioned in Section 2.3, systematic errors in the form of repeating patterns in the residuals from fits of the CH₄ data to the HITRAN database required us to implement the more refined Hartmann–Tran profile (HTP) [26,27]. From comparisons of the HTP CH₄ line shape with the HITRAN 2008, 2012, 2016 and 2020, the overall best shape relative to HTP was found for HITRAN 2008 (after scaling the overall peak intensity by 0.98). It is clear from the fit residuals to HITRAN 2008 shown in Figure 10a that the systematic errors are even more pronounced relative to those for HTP in Figure 10b. The HTP-integrated line intensity in ref. [27] was adjusted to that of the HITRAN 2020 database which corresponds to a peak intensity reduction of ≈0.91 of the HITRAN 2020 profile because of the reduced line broadening in HTP. Since our fits are to peak intensity; this led to a >8% increase in the best fit mixing ratios of CH₄ relative to HTP. From these comparisons, there is a clear need for an SI standard that is traceable to the (WMO) scale for the absolute intensities of the CH₄ lines. We again note that while the CO₂ profile is also impacted by line-mixing and speed-dependent effects, the effects on the profile are much smaller relative to CH₄ and are not revealed in the residuals in this study because of the smaller spectral coverage of the online CO₂ measurements.

Another unknown that impacts the absolute uncertainties, particularly for the IPDA LIDAR results of CH₄ at an altitude of ≈3.5 km, is in the calculated T&P profile using the U.S. Standard Atmosphere model [32]. While the CO₂ study was performed over a nearly horizontal path where T&P changes along the column are expected to be minimal, some evidence for systematic residuals in the CH₄ line shape residuals remain in Figure 10b. To investigate the impact of the T&P profile, we have refitted these data using two different inclination angles, θ, of 6.5° and 12.5°. The results are shown in Figure 10c,d, respectively. In contrast to the substantial increase in residuals for θ = 12.5°, the reduction seen for θ = 6.5° is significant. This angle difference of −4° corresponds to an altitude decrease of ≈1.4 km at a range of 20 km.

Given the strong pressure dependence of the shape of the CH₄ doublet on the residuals, we have performed pressure fits over two series of IPDA LIDAR line shape profiles. The 10 min intervals chosen are shown in Figure 11c as vertical black lines superimposed on the LIDAR map. The best-fit path-averaged pressures and uncertainties (Type A, k = 1 or 1σ) are shown in Figure 11a,b for the 2nd and 7th intervals, which represent high- and low-altitude cloud aerosol regions, respectively. The trends expected for the path-averaged T&P profile of the U.S. Standard Atmosphere are superimposed on the plots with yellow and red lines for inclination angles θ = 6.5° and 10.5°, respectively. As found above, the profile for θ = 6.5° is in much better agreement with the observed pressure changes. Finally, we also perform pressure fits to profile differences that are separated by 6 km as illustrated by the yellow boxes in Figure 11c. This 6 km separation corresponds to the difference between the cloud ranges in the two columns. We note that, as a result of the improved SNR in the cloud regions at 19 km and 13 km in Figure 11a,b, respectively, the relative uncertainties of the pressure fits near the 16 km average range in Figure 11d are significantly reduced. The results shown in Figure 11d are again in much better agreement with the T&P profile calculated at an inclination angle of θ = 6.5°.

There may be several possible reasons why the predicted inclination angle from the pressure fits in Figure 11 is about 4° less than the actual inclination angle determined from the platform pitch sensor (which has a confirmed accuracy of ±0.5°). Any effect on the CH₄ line shape that tends to fill in the central dip region will result in fits to higher pressures that will correlate to a lower inclination angle (and lower altitude). One possible mechanism that will fill in the dip region is from additional line broadening by H₂O vapor. The pressure broadening in the HTP model used here is for dry air only. Previous studies [39] have found that CH₄ dry-air broadening coefficients increase by 0.34 times the vapor pressure. As shown in the DIAL map of water vapor concentrations (Figure 8e), the relative humidity in this study is between 11% and 22% and, therefore, the additional CH₄ broadening by water vapor is only between 0.1% and 0.2% which will not account for the observed angle difference of −4°.

A second possibility that will decrease the dip depth is from averaging over the Doppler line shifts associated with changes in wind shear over the integrated path [40]. While a Doppler wind LIDAR system is planned for future measurement campaigns, the only atmospheric wind profile measurements available in this study are from two Denver International Airport (DIA) soundings performed several hours before and after the IPDA LIDAR study. The wind profile measurements, shown in Appendix A, Figure A1a,b, vary between 11 m/s and 18 m/s at altitudes >4 km relative to near-ground-level speeds between 1 m/s and 5 m/s. Over this range, the predicted Doppler frequency shifts for winds parallel to the beam path (i.e., worst case scenario) would be in the range from ±3 MHz to ±10 MHz. Tests on the integrated HTP line shapes over the largest of these shifts indicates the impact on the CH₄ dip depth is small. We also note that the maximum laser frequency shift from the thermal drift the filter cavity over periods longer than these 10 min averaging times is expected to be of this magnitude or less [4].

A final possible reason is that the U.S. Standard Atmosphere T&P profile is not an accurate approximation in this study. Some evidence for this is found from comparisons between the two DIA soundings shown in Appendix A, Figure A2a,b. We note especially the large differences in temperature profiles and an overall pressure shift of ≈15 hPa over this 12 h period. Significant differences are also seen relative to the T&P profile of the U.S Standard Atmosphere shown in Figure A2c. Given the impact of such factors on the online/offline measurements planned for the MERLIN mission [16,17] where such effects will not be evident in the data, current efforts are underway to perform simultaneous 3D wind profile measurements using a Doppler wind LIDAR system.

Improvements in the accuracy of the T&P profiles are also required to achieve the low absolute measurement uncertainties specified above. This is particularly important for the CO₂ and CH₄ lines chosen in this work given the high sensitivity to changes in temperature where the OD increases by 11% and 13%, respectively, for a ΔT of −35 K (320 K to 285 K). One possible way to improve the T&P model is to use data from the Global Forecast System weather prediction model and perform an atmospheric reanalysis [41] which should significantly improve the accuracy compared to the U.S. Standard Atmosphere used in these studies. Furthermore, efforts are underway to extend the measurement bandwidth of the frequency converter to measure additional lines. As we have shown previously [42], the scan bandwidth can be increased by as much as 4 cm⁻¹ (120 GHz) by using the higher order sidebands of the electro-optic phase modulator for frequency generation [11]. For the CH₄ line, the extension to 1.75 cm⁻¹ (53 GHz) is all that is required to capture a second water vapor line, (1₀₁–2₁₂), at 6078.25 cm⁻¹. The temperature dependence of the optical depth of the (1₀₁–2₁₂) transition is opposite to that of the (6₂₅–7₁₆) transition measured here. We estimate that a column-averaged temperature uncertainty of ≈±1 K can be obtained from OD measurements of these two water lines at a precision of ±1%. Based on the OD change of +13% for a −35 K change, the ±1 K translates to an uncertainty of <±0.4% for the CH₄ concentration. To measure the three additional CO₂ lines, R(20e), R(22e) and R(24e), adjacent to the R(18e) line measured here, will require a bandwidth extension to 3.7 cm⁻¹ (111 GHz). We estimate that a column-averaged temperature uncertainty of ±1.3 K can be obtained from the OD measurements of these lines at a precision of ±0.25%. Given the R(24e) line of CO₂ has about one half the temperature sensitivity of the R(18e) line (a 5.5% increase in OD for a −35 K change), the ±1.3 K translates to a ≈±0.2% uncertainty for the CO₂ concentration. Measurements are currently underway to assess these estimates.

Finally, we note that while the position of the transceiver platform is computer-controlled, real-time feedback to steer the platform’s transmitted beam to continuously measure cloud regions has not yet been implemented. Consequently, the cloud backscatter returns in Figure 3b,d were limited in duration (i.e., 2 h for CO₂ and 2.5 h for CH₄). More importantly, the overall backscatter signal returns were far from optimal; in the CO₂/H₂O region, the offline returns were near the saturation limit of the PMT detection system, which is in sharp contrast to the online returns in the CH₄/H₂O region that were near the noise floor. It would clearly be advantageous to work well between these two limiting cases. We expect that the SNR ratios will improve significantly once an independent LIDAR-based cloud profiling system under development is completed.

5. Conclusions and Outlook

In this work, we have demonstrated a combined DIAL/IPDA LIDAR system to measure range-resolved and integrated path-mixing ratios of dry-air CO₂ and CH₄ above the city of Boulder, Colorado. The integrated path measurements are performed to cloud targets up to 20 km in range and up to 3.5 km in altitude. The technique uses 10 frequencies to sample across ≈1.3 cm⁻¹ spectral regions and to measure portions of the absorption line profiles of CO₂/H₂O and CH₄/H₂O as a continuous function of range. A hybrid photon counting/current detection system enhanced the dynamic range to measure range-resolved backscatter signals from aerosols as well as distant cloud targets.

The DIAL/IPDA LIDAR multi-frequency method addresses two challenging tasks associated with standoff methods. For IPDA LIDAR studies, one important task is the measurement of the pulse energies with high precision and accuracy at each of the transmitted frequencies. This challenge is overcome through fitting the range-resolved DIAL data to extrapolate received pulse energies back to the time of emission thereby eliminating the difficulties associated with the uniform unbiased sampling of the transmitted pulses and with the calibration of multiple detectors’ responsivities [36]. A second challenge of two-frequency online/offline DIAL and IPDA LIDAR measurements is with potential biases that impact the optical densities and the frequency offset of the model fitting function arising from unknown environmental variations over the column average. Such factors may include added broadening contributions from foreign gases, poor models of the temperature and pressure (T&P) profiles and/or Doppler effects from altitude-dependent wind shear. From the multi-frequency sampling across the doublet structure of the R(6) manifold of CH₄, a clear bias signature was observed in this work and while its specific origin remains unclear, the nature of the observed deviations from the model have indicated the need to perform simultaneous wind LIDAR and T&P profile measurements. An added benefit of the line shape fits is the determination of the frequency offset that removes systematic errors associated with the OPO seed–signal frequency shift and other potential thermal drift factors.

Two independent studies were performed in the CO₂/H₂O and CH₄/H₂O regions to simultaneously measure both DIAL and IPDA LIDAR mixing ratios. Over the cloud aerosol regions, the uncertainties in the IPDA LIDAR retrievals were found to vary between ±0.2% and ±1% for CO₂ and ±0.1% and ±1% for CH₄, while the uncertainties in the range-averaged DIAL measurements varied between ±1% and ±2% in both regions. Moreover, the mixing ratios from the IPDA LIDAR measurements were found to be well-centered and within the error bars of the DIAL data.

Finally, we note that because of the current inability to adjust in real time the transmitter location, the overall signal returns were found to be far from optimal; the backscatter signal intensities were near the PMT detector’s saturation level in the CO₂/H₂O region while the online returns in the CH₄/H₂O region approached the detector’s noise floor. Current efforts are directed towards the development of a higher pulse repetition rate LIDAR system operating at 1.6 μm to survey the cloud base and provide active feedback for better positioning of the DIAL/IPDA LIDAR transceiver system.