Open Access
Add to BibSonomy
Abstract
We present a demonstration of a diode-laser-based high spectral resolution lidar. It is capable of performing calibrated retrievals of aerosol and cloud optical properties at a 150 m range resolution with less than 1 minute integration time over an approximate range of 12 km during day and night. This instrument operates at 780 nm, a wavelength that is well established for reliable semiconductor lasers and detectors, and was chosen because it corresponds to the D2 rubidium absorption line. A heated vapor reference cell of isotopic rubidium 87 is used as an effective and reliable aerosol signal blocking filter in the instrument. In principle, the diode-laser-based high spectral resolution lidar can be made cost competitive with elastic backscatter lidar systems, yet delivers a significant improvement in data quality through direct retrieval of quantitative optical properties of clouds and aerosols.
© 2017 Optical Society of America
1. Introduction
Lidar has a demonstrated benefit for aerosol and cloud monitoring due to its high sensitivity to small particulate scatters and direct time-of-flight measurement of target range. It can deliver the aerosol vertical distributions needed for accurate retrieval of aerosol optical depth from passive sensors [1]. It is also used for monitoring aerosol transport [2] and validation of transport models [3]. Such lidar observations are best performed with quantitative lidar systems. A conventional direct detection atmospheric elastic backscatter lidar (hereafter referred to as “elastic backscatter lidar” for brevity) does not capture sufficient information content to quantitatively and independently determine optical properties of clouds and aerosols in different atmospheric layers. This is because there are two unknown atmospheric optical terms (transmission and backscatter coefficient) but only one linearly independent detection channel. Additionally elastic backscatter lidar have no straightforward mechanism for calibration of instrument overlap functions and rely on knowledge of the instrument power stability. Algorithms from Klett [4] and Fernald [5] have been developed in attempt to retrieve quantitative optical properties of the atmosphere with such systems. However, they (1) require assumptions about the extinction to backscatter ratio, or lidar ratio, of the atmospheric particles which can vary considerably [6], (2) are prone to significant errors at low altitudes where geometric overlap functions contribute additional uncertainty, (3) depend on accurate monitoring of the lidar transmitted power and (4) couple and compound retrieval errors between different altitude layers of the atmosphere as shown in error comparisons between EarthCARE and CALIOP [7]. It is notable, that neither Klett nor Fernald algorithms include a geometric overlap term in their lidar equation definitions. Left unaccounted for, the effects of geometric overlap can produce very large biases in inverted data. At the same time, it is quite difficult to obtain an accurate calibration for geometric overlap with an elastic backscatter system.
Raman lidar [8] and high spectral resolution lidar (HSRL) [9–13] are generally regarded as the ideal lidar instruments for obtaining quantitative optical properties of clouds and aerosols [14]. Quantitative estimates of backscatter coefficient are obtained through a ratio of total backscatter signal to a molecular backscatter signal so the data is not dependent on absolute power stability, their ability to retrieve backscatter properties of aerosols at low altitudes is much less sensitive to the lidar overlap function, and their retrievals are independent of other altitudes so errors are decoupled between atmospheric layers. Direct retrievals of atmospheric extinction are also possible with these systems, though obtaining measurements with the desired accuracy and precision remains quite challenging [15]. While data products from these systems are clearly superior to elastic backscatter lidar, Raman lidar and HSRL are generally considered cost prohibitive for extensive unattended network deployment [16,17]. Furthermore, the high power of Raman systems and the visible wavelength of most HSRL (typically operating at 532 nm) can pose hazards to aircraft. Obtaining clearance to transmit on a continuous unattended basis poses a substantial bureaucratic challenge, particularly in the United States. Lidar that is not eye-safe generally requires human spotters or aircraft radar to act as transmitter interlocks and all regulated lidar (visible or class II or above) is categorically denied clearance to transmit if it is within 2 nautical miles (3.7 km) of an airport or within 5 nautical miles (9.3 km) of the runway ends (see Sections 4.5.6 and 4.6.1 in [18]).
To further enable atmospheric aerosol studies, Wiegner et al. [16] and Madonna et al. [17] suggest that more lidar observation sites are needed to fill in the spatial and temporal gaps between high performance Raman lidar and HSRL distributed throughout Europe in the European Aerosol Lidar Network (EARLINET). The use of commercially available non-quantitative elastic backscatter lidar technology (e.g. Vaisala Ceilometer, Sigma Space Micropulse Lidar) are generally assumed to be the tools to fill these gaps because they are affordable and readily available. However, since these instruments provide such marginal information content, we propose a more worthwhile effort is in the development of affordable quantitative approaches to lidar. In this line of thinking, we began investigating the possibility of developing a low-cost and simplified version of HSRL with promise for unattended operation that could also deliver quantitative data products, albeit at a lower signal quality than high performance HSRL systems.
Early HSRL used etalons to separate the spectrally narrow aerosol backscatter from the spectrally broad molecular backscatter components [10, 11]. However etalons present an engineering challenge because their bandwidth is angle dependent, so the minimum achievable notch bandwidth is dictated by the cone of angles incident on the etalon. Therefore obtaining the narrow bandwidths required for aerosol notch filtering requires a large etalon in order to conserve étendue. However large etalons are difficult to manufacture, particularly for high optical depth where very flat pieces of glass are required. In effect, the requirements for high optical depth and narrow bandwidth work against each other. In addition, the etalon’s absolute notch wavelength will depend on environmental conditions, so an environmental control system or laser reference locking system (to lock the laser to the etalon) are needed.
It was later proposed by Shimizu et al. [19] and then demonstrated by She et al. [12] that HSRL could be achieved using molecular or atomic vapor cells as filters. The barium atomic filter used in [12] has relatively low environmental sensitivity, the absorption line is stable, with no angular dependence and very high optical depths are achievable. In effect, absorption cells were shown to be alignment free narrowband filters. While the use of atomic and molecular filters for aerosol signal rejection eliminated the complexity of etalons, in many cases it made things more complicated on the transmitter end. Sources were not necessarily available at molecular or atomic absorption wavelengths. For example dye lasers were used in both Shimizu et al. (molecular iodine) and She et al. (barium vapor cell). Later Piironen and Eloranta [13] demonstrated HSRL using an iodine cell as a notch filter while operating at 532 nm, the first harmonic of Nd:YAG lasers. This was a considerable improvement in ease of operation over dye lasers and the iodine/Nd:YAG combination has more or less become the standard for ground based HSRL. Still, Nd:YAG lasers with narrow linewidths remain expensive and their visible wavelength corresponding to the iodine absorption line is regulated for aircraft safety, making them relatively poor candidates for general unattended continuous operation.
A major challenge in developing a low-cost HSRL is finding a receiver and transmitter pair that work well together. In the transmitter, we require a reliable, cost effective laser source that meets HSRL technical requirements (e.g. frequency stability and linewidth). In the receiver, we need a narrow linewidth filter (presumably an atomic or molecular vapor cell) to block aerosol returns. While many species of atomic and molecular filters have been investigated, it is only recently that sources meeting the requisite technical requirements have become available at some of these wavelengths. Perhaps nowhere is this more true than at the rubidium D2 absorption lines near 780 nm, which are frequently used for laser cooling and trapping. The considerable growth of research in this area has made available a significant amount of semiconductor optical components in the rubidium D2 wavelength region.
In collaboration with Montana State University (MSU), the National Center for Atmospheric Research developed a low-cost, micro-pulse differential absorption lidar that profiles water vapor in the lower to mid troposphere (WV-DIAL) [20, 21]. The diode-laser-based lidar technology employed in this system is well suited to unattended remote deployments, where there is limited access to the instrument for routine maintenance. To date, it has been deployed on four field projects (three domestic one international) and been shown to deliver quality data products during day and night, clear and cloudy conditions [22]. The instrument’s laser radiation is invisible and rated class 1M, so it is eye-safe and poses no visual interference at night. The instrument has undergone a number of revisions to improve its long term robustness and versatility and a test network of five units is currently under construction [23].
The HSRL design described herein draws from the diode-laser-based lidar architecture used in the WV-DIAL. The approach delivers a compromise between network deployability – cost, simplicity, safety, reliability, transportability (often associated with elastic backscatter lidar) – and accurate quantitative data products associated with higher performance Raman lidar and HSRL. In this work we describe the design of the diode-laser-based high spectral resolution lidar (DLB-HSRL) and present some initial results from the demonstration that suggest this technology has a clear potential to deliver quantitative retrievals of cloud and aerosol optical properties.
2. System description
The lidar architecture of the DLB-HSRL is derived from the WV-DIAL system described in [21]. A schematic of the DLB-HSRL architecture is shown in Fig. 1. The DLB-HSRL operates at 780 nm where semiconductor laser and detector components are available as commercial-off-the-shelf parts. The particular wavelength was selected because it corresponds to the rubidium D2 absorption line. This allows us to use a rubidium cell as the narrowband notch filter needed for the HSRL technique.
Fig. 1 Schematic of the DLB-HSRL. DBR stands for distributed Bragg reflector, TSOA is tapered semiconductor optical amplifier, NBF is narrow band filter, BS is beam splitter, Rb is the rubidium cell, MM Fiber is multimode fiber and SPCM is single photon counting module.
The lidar transmitter starts with a CW distributed Bragg reflector (DBR) laser (Photodigm PH780DBR120T8). The output of the DBR seed laser is collimated with an aspheric lens, and then circularized using an anamorphic prism pair. To avoid optical feedback from affecting the spectral properties of the DBR, the beam is isolated using a Faraday isolator. The beam is then coupled into a single mode fiber with an aspheric lens. A fiber tap directs 10% of the light to a self-calibrating wavelength meter (Bristol 671A) used to lock the seed laser wavelength to the center of the Rb absorption line to within ±0.16 pm (±77 MHz). The remaining 90% of the fiber-coupled light is collimated with an aspheric lens to a free-space beam, passed through a Faraday isolator, and polarization aligned with a half-wave plate. The polarization aligned beam is focused into a TSOA (Eagleyard/XSoptix EYP-TPA-0780-03000-4006-CMT04-0000) which is overdriven in pulsed-current mode operation [24,25]. A 12 mW input seed power results in a peak output power of 8W over a 1 us pulse width with an 8.0 A current pulse supplied to the TSOA. An extinction ratio of greater than 50dB was measured between the CW seed leakage light and the amplified pulsed output. At 7 kHz, this delivers an average transmit power of 55 mW with a pulse energy of 8 μJ. After the amplifier, the laser light is collimated, expanded and passed through a pair of axicon prisms to create an annular beam. This beam shaping helps avoid loss of power from the obscuration of the telescope secondary. The losses due to the beam shaping result in a 5 μJ transmitted energy at telescope exit. Finally, the laser light enters the transceiver through a 10 mm hole bored through an elliptical mirror that acts as the transmit/receive (T/R) switch.
The specifications for the DBR laser include 30 dB side mode suppression. These side modes effectively allow aerosol backscatter signals to pass unattenuated through the rubidium cell, thereby contaminating the molecular signal. In order to obtain accurate retrievals, side mode suppression is highly desirable and can be quantified by performing spectral scans of the cell at high temperatures such that the attenuation will saturate, where the primary mode is blocked more than the side mode leakage (see the receiver scan shown in Fig. 2). Our spectral scans of rubidium indicate approximately 5 × 10−5 saturated transmission through the cell, which indicates that the laser was doing much better than the 30 dB specification. This may be in part due to the two etalons and interference filters in the receiver that could also attenuate laser side modes. The TSOA adds some additional amplified stimulated emission, so the return signal from the atmosphere is not as spectrally pure as what we observe with the seed DBR. The mode coupling seems to depend on filter tuning and there have been instances where the aerosol leakage is quite large (a few percent) and other instances when it was quite small (less than a tenth of a percent). The exact reason for this behavior is still under investigation.
Fig. 2 Scan of the DLB-HSRL molecular channel (black) with a simulated backscatter spectrum from approximately 3 km (blue) and the resultant spectrum after passing through the molecular channel (red).
For stability, the instrument uses a coaxial transceiver with the telescope acting as both beam expander and receiver aperture stop. The hole in the T/R elliptical mirror corresponds to the transmitter and the light reflected by the mirrored surface corresponds to the receiver. This mirror is conjugate to the primary mirror of the telescope. A 60 mm focal length lens matched to the telescope (400 mm diameter, f/3 Newtonian) allows for a configuration such that the inner 20 cm of the telescope primary diameter is used for the transmitter and the outer portion is used for the receiver.
After the telescope collects and collimates the backscattered light, the elliptical mirror directs it to a beam reducing pair of lenses and a focusing lens that couples the light into a shielded 1 meter length, 105 μm diameter, 0.22 NA step index multimode fiber, limiting the lidar field-of-view to approximately 115 μrad and therefore minimizing the amount of multiple scattering observed by the system. This fiber serves two functional purposes. First, it makes the receiver modular, so its alignment can be performed on a bench using a reference source and later plugged into the system. We can also perform receiver frequency scans by substituting the multimode fiber with the fiber from the DFB. Second, the fiber, to some extent, scrambles the spatial modes of the system and reduces the differences in overlap function between the molecular and combined channel detectors. No major effort is made to obtain complete scrambling of spatial modes in the fiber (e.g. tight bends, long fiber length) so this approach cannot be regarded as perfect, but with the fiber in place, we do not apply any differential overlap correction to the data during processing.
The exit port of the fiber directs light into the receiver module. The light is collimated then filtered using a pair of narrow band filters (NBF) – 10 nm FWHM ThorLabs (70% transmission) and 1.2 nm FWHM from Alluxa (85% transmission) – and two solid fused silica etalons (custom parts from LightMachinery) to minimize solar background. The narrow etalon has a 5.7 GHz bandpass with a 45 GHz free spectral range (FSR) and peak transmission of 70% that was initially designed for operation at 828 nm (i.e., it was not optimized for use at 780 nm). In May 2017, we added the wide etalon with a 15 GHz bandpass and 250 GHz FSR and peak transmission of 70%. The combination of a narrowband etalon with low FSR and a wideband etalon with large FSR reduces the number of passband fringes allowed onto the detector. We then split the filtered light using a 70:30 beam splitter so that the 30% portion is focused onto the combined channel detector and the 70% portion passes through a heated 19 mm diameter x 75 mm cell of isotopic rubidium 87 (ThorLabs GC19075-RB87) controlled to approximately 65° C. The beam splitter ratio roughly corresponds to the reciprocal of the difference in efficiency between the two channels so that they both operate in approximately the same dynamic range. The transmitter wavelength is tuned to the rubidium 87 D2, F=2 line so the spectrally narrow aerosol/cloud backscatter is attenuated and the wings of the molecular return pass through the cell and make it onto the molecular channel detector.
A spectral scan of the molecular channel is shown in Fig. 2. It shows the rubidium 87 spectrum with the F=2 line centered in the filtered pass band (the F=1 line can be seen on the left side of the figure). The effect of the narrowband sunlight blocking etalon can also be seen, but the wideband etalon and NBF pair are too broad to appear in this scan. The rubidium cell can attenuate aerosol backscatter by more than four orders of magnitude which is more than adequate for the measurements presented here. The impact of imperfect notch filter blocking on backscatter coefficient retrievals will be discussed in the next section.
During alignment of the molecular channel, a pair of rare earth, NdFeB, magnets (grade N42, dimensions 3” × 1” × 0.5”, surface field 3510 Gauss) are placed on top of the Rb cell heater assembly. The magnetic field Zeeman splits the Rb transition allowing high transmission through the cell (increase of 40 dB) while the laser wavelength remains locked to the Rb 87 D2, F=2 line center. This allows us to peak up the laser signal on the molecular detector where the rubidium filter would otherwise heavily attenuate the narrow line laser and make assessment of fine adjustments difficult.
Each detector channel consists of a fiber coupled single photon counting module (SPCM, Excelitis SPCM-AQRH-13-FC). The SPCM output a digital pulse each time a photon is detected by the avalanche photo-diode front end which is passed to a multi-channel scalar (MCS) to generate a histogram of received pulses with 37.5 m range bins. The MCS then writes out histograms after 2 seconds of integration (14,000 laser shots).
3. Processing and calibrations
DLB-HSRL data is processed according to the same conventions applied to iodine based HSRL [9,26] where two lidar equations are used to derive HSRL data products. The combined channel is described by
where Sc(R) is the measured combined channel signal as a function of range R, K is an arbitrary constant for various power and efficiency terms (e.g. laser power, collection aperture, etc.), ηc(R) is the combined channel geometric overlap function, βm(R) is the molecular backscatter coefficient, βa(R) is the aerosol or cloud backscatter coefficient, α(R) is the extinction coefficient and Bc is the combined channel background counts.Similarly, the molecular channel is described by
where Sm(R) is the measured molecular channel signal, 1/Gm is the difference in efficiency between the combined and molecular channels, ηm(R) is the molecular channel geometric overlap function, Cam is the aerosol to molecular crosstalk representing the amount of narrowband light leakage through optical notch filter (e.g., in this specific case the rubidium cell) and Bm is the molecular channel background counts. The term Cmm(R) is the efficiency of the molecular return through the molecular channel filter. This term is more properly a function of temperature and pressure at the backscatter altitude as these terms dictate the spectral shape of the molecular return. In order to estimate Cmm(R) we use an estimate of atmospheric temperature and pressure based on a surface measurement and −6.5 K/km lapse rate to obtain the shape of the molecular backscattered spectrum. This spectral calculation can be performed quite rapidly and at sufficient accuracy using a linear approximation based on principal component analysis described in Binietoglou et. al [27].With the multimode fiber field stop installed in the system, we assume that the overlap functions are approximately identical such that ηm(R) ≈ ηc(R) and the terms will be replaced with the more general η(R).
In this work, the estimates of parameters will be represented with tildes over them to clarify potential sources of error in the signal processing. For example, G̃m is the estimated value of Gm and is obtained by matching the high altitude signals of the combined and molecular channels. We also estimate background terms B̃c and B̃m by averaging the last 100 range bins (approximately 17.5–21 km) of the profile.
For the purpose of simplifying the processing equations, we will introduce the following definitions for molecular and combined channel observations that include accounting for gain differences, crosstalk in the channels and background counts
and3.1. Backscatter coefficient
The backscatter coefficient of atmospheric aerosols is calculated
where β̃a(R) is the estimated aerosol backscatter coefficient and β̃m(R) is the estimated molecular backscatter coefficient. The molecular backscatter coefficient is estimated from pressure and temperature profiles using [28] where P(R) is the pressure in Pa, T(R) is the temperature in K, λ is the wavelength of the laser in nm and kB is the Boltzmann constant. Our processing estimates atmospheric temperature and pressure profiles using a surface measurement at the instrument location and assuming a standard atmosphere with an average lapse rate of −6.5 K/km for the entire troposphere, or using radiosonde profiles near the operation site when available. Assuming the lapse rate is bounded between −6.5 K/km and −9.8 K/km (moist and dry adiabatic), the error in the lapse rate assumption may produce errors in β̃m(R) (and therefore β̃a(R)) as high as 15% at the tropopause. Errors in estimates of temperature and pressure also contribute uncertainties in estimates of C̃am(R) on similar orders. High quality data processing is probably best achieved by using temperature reanalysis from high resolution weather models to minimize these errors.While we have included the term for completeness, our direct retrievals are generally processed assuming that Cam = 0. This assumption was not valid for a number of early test data sets where Cam was at times as high as 0.03. However applying direct corrections for aerosol crosstalk in the molecular channel were generally problematic due to detector nonlinearity, so we simply accept that for some of these early data sets, high backscatter data is underestimated. In the DLB-HSRL’s latest configuration we estimate that Cam = 0.0007.
The theory presented here has neglected the nonlinear response of the detectors. This typically this is not significant when count rates are below 1–2 MHz. Nonlinear response in detectors has been problematic when attempting to correct for aerosol crosstalk in the molecular channel. The greatest impact of nonlinearity is well correlated with the greatest impact of crosstalk – in regions of high backscatter. Note that without correction, crosstalk defines a maximum resolvable backscatter coefficient dictated by the molecular target at the corresponding range bin and errors in the retrieved backscatter coefficient will become large observations approach the value.
3.2. Molecular denoising
In our processing routines, we apply a denoising routine developed by Marais et al. [29]. Poisson maximum likelihood estimation with a total variance penalty function is used to approximate the signal as a discrete piecewise function multiplied by an assumed profile shape. This approach is used to denoise the molecular channel, where shot noise tends to diminish the precision of backscatter coefficient retrievals in instances of light aerosol loading in the free troposphere and high altitude clouds during daytime. Because changes in molecular signals tend to be sparse, they are well suited to estimation by a piecewise function. The optimizer uses the estimated background, geometric overlap function and molecular backscatter as a baseline for the signal estimate and modifies a state variable x̃ (R) (the discrete piecewise function) to obtain an estimate for the noisy profile. The estimate is given by
The estimated parameters are used to impose an expected structure on the profile, but the denoising algorithm does not require that these parameters be terribly accurate. The process of denoising matches the signal estimate S̃m(R) to the observed noisy signal Sm(R) largely irrespective of accuracy errors in the underlying structure. A reasonable estimate for the molecular baseline and background however, helps improve the sparsity (reduces the number of pieces) of the retrieval. Once we obtain the estimate S̃m(R) from the denoising procedure, Eq. (4) is used to correct for background and gain mismatch between the molecular and combined channels.Figure 3 shows a comparison between the range corrected molecular signal and the denoised molecular signal in a one minute profile on June 7, 2017 (1530 UTC, 0930 local time) in Boulder, CO, USA. The variance of the high altitude molecular signal is significantly reduced by applying this denoising technique. Note that the dynamics of the denoised signal become increasingly sparse at higher altitudes where at lower altitudes it nearly matches the original signal. In effect the denoising algorithm uses lower range resolution in regions where the signal is noisy.
Fig. 3 Vertical profile of background subtracted molecular backscatter data integrated over one minute (blue) with the denoised molecular signal (green). This data is from June 7, 2017 during daytime conditions (1530 UTC).
3.3. Extinction coefficient
For all lidar instruments (i.e, elastic backscatter, and more advanced HSRL and Raman systems), direct calculation of extinction is considerably more challenging than backscatter coefficient [15]. The calculation uses only the molecular channel, so we lose the advantage of signal ratios to cancel out instrument terms. Conventionally we invert the lidar equation to obtain the following formula for extinction coefficient:
where α̃(R) is the estimated atmospheric extinction coefficient. Here the removal of aerosol crosstalk becomes important, because substantial leakage of cloud signals will likely produce negative or underestimated extinction coefficients.There are considerable challenges in obtaining the extinction coefficient [15]. While the derivative conveniently eliminates DC factors such as K, it also amplifies noise, so that low signal variance is typically required to obtain reasonable extinction precision. At the same time, the extinction coefficient has a nonlinear relationship to the recorded signals, so integrating over space or time in a variable extinction field can result in biases in the retrievals.
We need to know the geometric overlap function of the lidar system in order to estimate extinction. The molecular channel is used for this estimation process where we assume the difference between β̃m(R) and the observed molecular signal is largely due to the lidar overlap function. We include an additional correction term to adjust for the aerosol extinction in the estimation process. This extinction estimate is informed by the aerosol backscatter coefficient to define a structure and an assumed lidar ratio. This gives the estimated overlap function
where s̃LR is the assumed lidar ratio for all aerosol particles in the profile. This calibration is generally only performed under relatively clear conditions in an effort to minimize the effects of the lidar ratio assumption. Fortunately, an HSRL is able to deliver a quantitative assessment of aerosol loading using the retrieved backscatter coefficient or backscatter ratio. The value for K is generally not known, but is not important for retrievals. The term falls out of extinction estimates due to the derivative operation.The use of the molecular channel for estimating geometric overlap η(R) is an improvement over what is possible with elastic backscatter systems. With a molecular channel, we are only concerned about bias errors resulting from aerosol extinction where in an elastic backscatter lidar, both aerosol extinction and – even more significantly – aerosol backscatter will generate biases in the calibration. Nevertheless, the approach in Eq. (10) is still not ideal. The assumption of the lidar ratio s̃LR to account for aerosol extinction during calibration effectively limits the accuracy to which we can subsequently estimate extinction and lidar ratio in corrected regions. We should note that wide-field-of-view channels have also been employed for this calibration process as is done in [15] which reduces the number of assumptions but require additional detector channels.
Evaluating Eq. (9) with Eq. (10) and assuming all other estimated terms are equal to the actual value, we find error in the assumed lidar ratio generates a linear bias in extinction retrievals. In effect, the retrieval is biased by the error in assumed extinction during calibration
where the superscript (cal) denotes that the term is from the time of the geometric overlap calibration and is the true range dependent lidar ratio during the calibrated observation. Thus, there are two ways to minimize the error contribution: perform the calibration when the aerosol loading is small or accurately estimate the aerosol lidar ratio. Neither of these present great options. Clearly we cannot control the amount of aerosol loading but we can at least assess it using backscatter coefficient retrievals. By contrast, we can neither control the aerosol lidar ratio, nor can we easily assess the accuracy of its assumed value.Furthermore, the retrieved aerosol lidar ratio is also impacted
If we assume a possible spread in lidar ratios of ±15 sr, then the aerosol backscatter needs to be five times larger than that during calibration to obtain a lidar ratio accuracy of ±3 sr. Characterizing the extinction properties of low extinction atmospheric aerosols layers presents a considerable challenge, even with a quantitative lidar system. We should note that these error calculations were derived using the direct calculation of extinction, but other retrieval methods will suffer from the same errors because they fundamentally rely on the same geometric overlap calibration to perform their inversions.This analysis effectively lays out the theory for determining fundamental limits in retrieving extinction with the current architecture of this HSRL. At this time direct retrieval of extinction using Eq. (9) does not appear feasible on DLB-HSRL data due to shot noise limitations. We are investigating, and have had some success, using an optimal estimation approach developed by Marais et al. [29] for obtaining extinction and lidar ratio estimates. This allows the processor to use backscatter coefficient retrievals to isolate the location of extinction components (extinction typically only occurs where backscatter also occurs) to better constrain the retrievals. This approach has significantly improved signal precision but still requires the same calibration terms in Eq. (9) to obtain accurate results. Preliminary analysis suggests processing DLB-HSRL data using this technique will be able to resolve extinction in clouds, but it is not yet clear if it will be able to resolve extinction from moderate aerosol loading at short (e.g. 1 minute) time scales or in low altitude regions requiring significant overlap correction.
4. Observations
First light of the DLB-HSRL occurred on Dec. 8, 2016 in Boulder, CO, USA. The lidar was operated for the next seven months with down time for development and modifications. In April 2017, the fiber field stop was added to the system to reduce the need for a differential overlap correction. In May 2017 the system was integrated with a WV-DIAL system and in late May 2017, the wide band etalon was added to the system to improve solar background rejection.
During initial operation, the aerosol crosstalk term was on the order of 10−3. However after system realignment in March and April we noticed the aerosol crosstalk increased to approximately 0.03 which significantly limited our ability to accurately measure the backscatter coefficient of clouds. We developed a maximum likelihood estimator that accounted for this crosstalk, but better hardware performance was clearly desirable. After installation of the second wide band etalon in May, the crosstalk was significantly diminished. Further investigation is warranted, but we speculate that side modes in the DBR (specified at −30dB) are passing, unattenuated, through the rubidium cell. The better performance may have been due to the receiver etalon blocking the laser side modes. We have found that by adjusting the narrowband etalon, we can effectively eliminate cloud returns in our molecular channel and cloud observations suggest aerosol crosstalk is less than 10−3 (currently estimated at 0.0007).
Eight days of observations of backscatter coefficient are shown in Fig. 4. This data is processed with one minute time bins and 37.5 m range bins (though the transmitted pulse is 300 m in length therefore limiting range resolution to 150 m). The DLB-HSRL is able to resolve aerosol structure throughout the troposphere in both day and night, clear and cloudy conditions. This particular time period was selected because there is substantial aerosol structure and the system was in its final configuration. The vertical white bar near the end of day 4 is a period of approximately 1.5 hours on June 5 (Day 4) where we have no data. This was the result computer malfunction.
Fig. 4 Backscatter coefficient over 8 days in Boulder, CO, USA starting on June 1, 2017. The data is processed with 1 minute time bins and 37.5 m range bins.
In addition to its ability to operate unattended at long durations, the DLB-HSRL can deliver backscatter coefficient data at short time scales. Figure 5 shows backscatter coefficient estimates processed with 10 second time bins, enabling better resolution of turbulent processes. Observational data from DLB-HSRL is close to being useful for resolving turbulence using methods described in McNicholas and Turner [30]. Based on the signal quality we see at 10 seconds, we are optimistic that higher temporal resolution is feasible in the boundary layer.
Fig. 5 Backscatter coefficient in Boulder, CO, USA starting on June 6, 2017. The data is processed with 10 second time bins and 37.5 m range bins.
The GV-HSRL (developed by University of Wisconsin and operated by NCAR) is a high performance HSRL operating at 532 nm, with a short transmit pulse (15 m pulse length) [31,32]. The instrument transmits 400 mW or 100 μJ per pulse at 4 kHz – considerably more transmit power and energy than that of the DLB-HSRL. With a shorter wavelength, the GV-HSRL also obtains approximately 4.6 times higher Rayleigh scattering efficiency than the DLB-HSRL. The GV-HSRL uses an iodine cell, providing an optical depth greater than 4 on line center, to block aerosol returns in the molecular channel. Unlike the DLB-HSRL, the GV-HSRL has no fiber in the receiver and requires periodic correction for the difference in geometric overlap function between the molecular and combined channels (ηm(R) ≉ ηc(R)). The GV-HSRL generally provides a higher standard of measurement than that expected for the DLB-HSRL. For the purpose of evaluating DLB-HSRL accuracy, it is useful to compare the retrievals of the two instruments.
The DLB-HSRL and the GV-HSRL were collocated and briefly operated simultaneously on April 11, 2017. During this time, the DLB-HSRL had relatively high aerosol to molecular channel crosstalk (Cam was estimated at 0.028, which is more than an order of magnitude higher than the instrument’s current crosstalk) so it is difficult to compare high backscatter retrievals without additional corrective processing. Data was processed on both instruments at a range bin size of 37.5 m and at time bins of 60 seconds. The DLB-HSRL backscatter coefficient data is corrected using a maximum likelihood estimator that accounts for both aerosol to molecular crosstalk and detector dead time. This approach is similar to that reported in Marais et al. [29], but it allows the optimizer to adjust calibration coefficients. This allowance for tuning calibration parameters likely eliminates the guarantee that the error minimum obtained in optimization corresponds to a global minimum, but because we have reasonable initial estimates of the state variables, the nearby minimum very likely corresponds to the global minimum of the error function.
Figure 6 shows the time height profiles of the retrieved backscatter coefficients for the GV-HSRL (top) and DLB-HSRL (bottom). The color bar of the GV-HSRL is scaled by the ratio of wavelengths (532/780) for easier comparison of the aerosol backscatter in the time height profiles. This is based on the approximate expected difference in backscatter cross sections for aerosol particles at different wavelengths [33,34], however the clouds are not expected to have wavelength dependent backscatter as discussed below and in Fig. 7. We can see some distinctions between the higher performance GV-HSRL and the DLB-HSRL. Thin cloud features are better resolved in the GV-HSRL which has a 15 m laser pulse length so that data processed with 37.5 m range bins truly has this resolution. Due to its long pulse length, the DLB-HSRL has an effective range resolution of 150 m. Even though data is processed with smaller range bins, cloud features are smoothed by the laser pulse. This effect is perhaps most visible in the thin region of high backscatter at approximately 8 km just before 4 UTC. We should also point out that no denoising is applied to the GV-HSRL data, where denoising the molecular channel is part of our standard processing routine in the DLB-HSRL.
Fig. 6 Time height profiles of backscatter coefficient retrievals from the GV-HSRL (top) and DLB-HSRL (bottom) on April 11, 2017 in Boulder, CO, USA. The color bar of the GV-HSRL profile is scaled relative to the DLB-HSRL color bar to obtain similar color scales.
A histogram comparing the backscatter coefficient retrievals is shown in Fig. 7 where the ratio of the two wavelengths is denoted by the gold dashed line and the 1:1 line is the red dashed line. At lower backscatter coefficients, the histogram is centered on the gold line (ratio of wavelengths). Near 10−6m−1sr−1, the histogram appears to shift from the gold line to the red line (1:1). It is expected that the aerosol backscatter coefficients should roughly scale with wavelength, but large cloud particles that we typically associate with higher backscatter will have no wavelength dependence [33,34]. The histogram seems to reflect this expectation remarkably well. We should note that the data processing on this comparison was blind to the other observational dataset. The two systems were processed according to their standard calibrations and configurations. Other than processing at the same resolution, no other attempt was made to match the profiles.
Fig. 7 Two dimensional histogram for comparison of backscatter coefficient retrievals of the GV-HSRL (horizontal axis) and DLB-HSRL (vertical axis) on April 11, 2017 in Boulder, CO, USA. The red dashed line indicates the 1:1 line and the gold dashed line is the wavelength ratio between the two systems. Note the histogram color bar is logarithmic.
5. Conclusion
We have demonstrated a quantitative lidar system that can deliver high precision observations throughout the troposphere. This same diode-laser-based architecture has demonstrated success for delivering an affordable, reliable, low maintenance instrument [21]. The new instrument operates at a wavelength of 780 nm and uses a commercially available DBR seed laser and TSOA within the transmitter. In addition, we utilize a heated cell of isotopic rubidium 87 to act as the aerosol blocking filter in the molecular channel. This is a relatively low cost approach to obtaining a narrow band notch filter with minimal étendue limitations on performance and complexity. With the relatively small diameter absorption cell, high optical depths and a spectrally narrow notch were obtained.
The DLB-HSRL has operated in various states of completeness since December 2016. In its current configuration, backscatter coefficient retrievals have been shown to have a useful resolution down to 10 second time bins with a range resolution of 150 m (range bins of 37.5 m). In general we process the data at 1 minute resolution while using Poisson denoising [29] to reduce noise on the molecular channel. This allows us to obtain measurements of aerosol backscatter coefficient throughout the troposphere in both day and night, clear and cloudy conditions. We are currently investigating the instrument’s ability to provide extinction retrievals using the technique described in Marais et al. [29]. Based on preliminary work, we anticipate that resolving cloud extinction is possible but aerosol extinction under light loading conditions maybe outside the capabilities of the current instrument.
Funding
The National Center for Atmospheric Research is sponsored by the National Science Foundation.
Acknowledgments
We’d like to thank Ed Eloranta for his valuable insights into HSRL, early discussions and encouragement to pursue this instrument. We’d also like to thank Josh Carnes and Robert Stillwell for providing valuable insights through internal reviews and Kevin Repasky for loaning us subcomponents that were used for this demonstration.
References and links
1. M.R. Vuolo, M. Schulz, Y. Balkanski, and T. Takemura, “A new method for evaluating the impact of vertical distribution on aerosol radiative forcing in general circulation models,” Atmos. Chem. Phys. 14, 877–897 (2014). [CrossRef]
2. B. de Foy, S.P. Burton, R.A. Ferrare, C. A. Hostetler, J.W. Hair, C. Wiedinmyer, and L.T. Molina, “Aerosol plume transport and transformation in high spectral resolution lidar measurements and WRF-Flexpart simulations during the MILAGRO Field Campaign,” Atmos. Chem. Phys. 11, 3543–3563 (2011). [CrossRef]
3. A. Hodzic, H. Chepfer, R. Vautard, P. Chazette, M. Beekmann, B. Bessagnet, B. Chatenet, J. Cuesta, P. Drobinski, P. Goloub, M. Haeffelin, and Y. Morille, “Comparison of aerosol chemistry transport model simulations with lidar and Sun photometer observations at a site near Paris,” J. Geophys. Res. 109(D23), D23201 (2004). [CrossRef]
4. J. D. Klett, “Stable analytical inversion solution for processing lidar returns,” Appl. Opt. 20(2), 211–220 (1981). [CrossRef] [PubMed]
5. F. G. Fernald, “Analysis of atmospheric lidar observation: some comments,” Appl. Opt. 23(5), 652–653 (1984). [CrossRef] [PubMed]
6. S. Groß, V. Freudenthaler, K. Schepanski, C. Toledano, A. Schafler, A. Ansmann, and B. Weinzierl, “Optical properties of long-range transported Saharan dust over Barbados as measured by dual-wavelength depolarization Raman lidar measurements,” Atmos. Chem. Phys. 15, 11067–11080 (2015). [CrossRef]
7. A. J. Illingworth, H.W. Barker, A. Beljaars, M. Ceccaldi, H. Chepfer, N. Clerbaux, J. Cole, J. Delanoë, C. Domenech, D.P. Donovan, S. Fukuda, M. Hirakata, R.J. Hogan, A. Huenerbein, P. Kollias, T. Kubota, T. Nakajima, T.Y. Nakajima, T. Nishizawa, Y. Ohno, H. Okamoto, R. Oki, K. Sato, M. Satoh, M.W. Shephard, A. Velázquez-Blázquez, U. Wandinger, T. Wehr, and G.-J. van Zadelhoff, “The EarthCARE Satellite: The next step forward in global measurements of clouds, aerosols, precipitation and radiation,” Bull. Amer. Meteor. Soc. 96, 1311–1332 (2015). [CrossRef]
8. D. N. Whiteman, “Examination of the traditional Raman lidar technique. I. evaluating the temperature-dependent lidar equation,” Appl. Opt. , 42(15), 2571–2592 (2003). [CrossRef] [PubMed]
9. E. W. Eloranta, “Chapter 5: High Spectral Resolution Lidar,” in Lidar: Range-Resolved Optical Remote Sensing of the Atmosphere, C. Weitkamp, ed. (Springer, 2005). [CrossRef]
10. G. Fiocco, G. Benedetti-Michelangeli, K. Maischberger, and E. Madonna, “Measurement of Temperature and Aerosol to Molecule Ratio in the Troposphere by Optical Radar,” Nat. Phys. Sci. 229, 78–79 (1971). [CrossRef]
11. S. T. Shipley, D. H. Tracy, E. W. Eloranta, J. T. Trauger, J. T. Sroga, F. L. Roesler, and J. A. Weinman, “High spectral resolution lidar to measure optical scattering properties of atmospheric aerosols. 1: Theory and instrumentation,” Appl. Opt. 22(23), 3716–3724 (1983). [CrossRef] [PubMed]
12. C. Y. She, R. J. Alvarez, L. M. Caldwell, and D. A. Krueger, “High-spectral-resolution Rayleigh-Mie lidar measurement of aerosol and atmospheric profiles,” Opt. Lett. 17(7), 541–543 (1992). [CrossRef] [PubMed]
13. P. Piironen and E. W. Eloranta, “Demonstration of a high-spectral-resolution lidar based on an iodine absorption filter,” Opt. Lett. 19, 234–236 (1994). [CrossRef] [PubMed]
14. G. Pappalardo, A. Amodeo, A. Apituley, A. Comeron, V. Freudenthaler, H. Linné, A. Ansmann, J. Bösenberg, G. D’Amico, I. Mattis, L. Mona, U. Wandinger, V. Amiridis, L. Alados-Arboledas, D. Nicolae, and M. Wiegner, “EARLINET: towards an advanced sustainable European aerosol lidar network,” Atmos. Meas. Tech. 7, 2389–2409 (2014). [CrossRef]
15. E. W. Eloranta, “Extinction measurements with low-power HSRL systems–Error limits,” in Proceedings of 28th International Laser Radar Conference (2017), paper 205.
16. M. Wiegner, F. Madonna, I. Binietoglou, R. Forkel, J. Gasteiger, A. Geiß, G. Pappalardo, K. Schäfer, and W. Thomas, “What is the benefit of ceilometers for aerosol remote sensing?” Atmos. Meas. Tech. 7, 1979–1997 (2014). [CrossRef]
17. F. Madonna, F. Amato, J. Vande Hey, and G. Pappalardo, “Ceilometer aerosol profiling versus Raman lidar in the frame of the INTERACT campaign of ACTRIS,” Atmos. Meas. Tech. 8, 2207–2223 (2015). [CrossRef]
18. American National Standard Institute, American National Standard for Safe Use of Lasers Outdoors Z136.6–2015 (Laser Institute of America, 2015), p. 33.
19. H. Shimizu, S. A. Lee, and C. Y. She, “High spectral resolution lidar system with atomic blocking filters for measuring atmospheric parameters,” Appl. Opt. 22(9), 1373–1381 (1983). [CrossRef] [PubMed]
20. A. R. Nehrir, K. S. Repasky, J. L. Carlsten, M. D. Obland, and J. A. Shaw, “Water vapor profiling using a widely tunable, amplified diode-laser-based differential absorption lidar (DIAL),” J. Atmos. Ocean. Technol. 26, 733–745 (2009). [CrossRef]
21. S. M. Spuler, K. S. Repasky, B. Morley, D. Moen, M. Hayman, and A. R. Nehrir, “Field-deployable diode-laser-based differential absorption lidar (DIAL) for profiling water vapor,” Atmos. Meas. Tech. 8, 1073–1087 (2015). [CrossRef]
22. T. M. Weckwerth, K. J. Weber, D. D. Turner, and S. M. Spuler, “Validation of a Water Vapor Micropulse Differential Absorption Lidar (DIAL),” J. Atmos. Ocean. Technol. 33, 2353–2372 (2016). [CrossRef]
23. S. Spuler, K. Repasky, M. Hayman, and A. Nehrir, “Micro-Pulse, Differential Absorption Lidar (DIAL) Network for Measuring the Spatial and Temporal Distribution of Water Vapor in the Lower Atmosphere,” in Proceedings of 28th International Laser Radar Conference (2017), paper 262.
24. K. Takase, J. K. Stockton, and M. A. Kasevich, “High-power pulsed-current-mode operation of an overdriven tapered amplifier,” Opt. Lett. 32(17), 2617–2619 (2007). [CrossRef] [PubMed]
25. A. R. Nehrir, K. S. Repasky, and J. L. Carlsten, “Eye-Safe Diode-Laser-Based Micropulse Differential Absorption Lidar (DIAL) for Water Vapor Profiling in the Lower Troposphere,” J. Atmos. Ocean. Technol. 28(2), 131–147 (2011). [CrossRef]
26. M. Esselborn, M. Wirth, A. Fix, M. Tesche, and G. Ehret, “Airborne high spectral resolution lidar for measuring aerosol extinction and backscatter coefficients,” Appl. Opt. 47(3), 346–358 (2008). [CrossRef] [PubMed]
27. I. Binietoglou, P. Giampouras, and L. Belegante, “Linear approximation of Rayleigh-Brillouin scattering spectra,” Appl. Opt. 55(27), 7707–7711 (2016). [CrossRef] [PubMed]
28. R. M. Measures, Laser Remote Sensing (Wiley-Interscience, 1984).
29. W. J. Marais, R. E. Holz, Y. H. Hu, R. E. Kuehn, Edwin E. Eloranta, and R. M. Willett, “Approach to simultaneously denoise and invert backscatter and extinction from photon-limited atmospheric lidar observations,” Appl. Opt. 55(29), 8316–8334 (2016). [CrossRef] [PubMed]
30. C. McNicholas and D. D. Turner, “Characterizing the convective boundary layer turbulence with a High Spectral Resolution Lidar,” J. Geophys. Res. Atmos. 119(22), 12910–12927 (2014). [CrossRef]
31. UCAR/NCAR - Earth Observing Laboratory and University of Wisconsin, High Spectral Resolution Lidar for the Gulfstream-V; G5-HSRL, doi: [CrossRef] (2010).
32. I. A. Razenkov, E. W. Eloranta, J. P. Hedrick, and J. P. Garcia, “The design of a new airborne of a high spectral resolution lidar,” in Proceedings of 24th International Laser Radar Conference (2010), pp. 56–59.
33. M. L. Wright, E. K. Proctor, L. S. Gasiorek, and E. M. Liston, “A preliminary Study of Air-Pollution Measurement by Active Remote Sensing Techniques,” NASA CR-132724, 178 (1975).
34. V. Srivastava, J. Rothermel, A. D. Clarke, J. D. Spinhirne, R. T. Menzies, D. R. Cutten, M. A. Jarzembski, D. A. Bowdle, and E. W. McCaul, “Wavelength dependence of backscatter by use of aerosol microphysics and lidar data sets: application to 2.1-μm wavelength for space-based and airborne lidars,” Appl. Opt. 40(27), 4759–4769 (2001). [CrossRef]
