Using the DIAL technique, it is possible to retrieve ozone mixing ratios from the backscatter profiles. The technique essentially uses the differential absorption of ozone at two different wavelengths that are relatively close together to minimize aerosol effects but far enough apart to have a sufficiently large difference in their ozone absorption cross sections. Consequently, the ozone calculation uses the two-wavelength solution of the lidar equation given below (Kovalev and McElroy, 1994): Nz=-12αon-αoff[ddzln⁡Pon(z)Poff(z)-ln⁡βon(z)βoff(z)+2Δσz+2∑iΔαini(z)], where αon-αoff is the differential ozone absorption cross section, Pon(z)Poff(z) is the signal ratio, βon(z)βoff(z) is the ratio of the backscatter coefficient at the “on” and “off” wavelengths and 2Δσ(z) is the total two-way extinction coefficient differential, Δαi is the differential cross section of the interfering trace gas and ni(z) is the number density profile of the interfering trace gas. In our system, 287 nm represents the on wavelength, and 299 nm represents the off wavelength. It is possible to express the backscatter contribution overall to the ozone calculation at the on and off wavelengths based solely on the ratio between the aerosol and molecular backscatter,B∗(z), at some reference wavelength. This is represented by ddzln⁡βon(z)βoff(z)=ddzln⁡1+B∗(z,λref)λonλref4-υ1+B∗(z,λref)λoffλref4-υ, where υ is the Ångström exponent representing the wavelength dependence of aerosol Mie backscatter and in our case the reference wavelength is 355 nm. The difficulty arises in determining the Ångström exponent in regions where the aerosol concentrations are lower and with enough precision to provide an accurate correction. This point is illustrated for a forest fire case shown in Sect. 4.3. In addition, the Ångström exponent determined by our system reflects the wavelength dependence of aerosol backscatter between 355 and 532 nm, which may be different than the wavelength dependence between the on- and offline wavelengths of 287 and 299 nm.

During the early stages of the optical detector design for the aerosol lidar, it was determined that with the addition of a few optics (see Fig. 2), which requires very little additional space in the detector design, it would be possible to measure nighttime water vapor using the Raman technique on the 355 nm laser wavelength. This would be particularly valuable when identifying STT events where the dry stratospheric air can be easily identified by the water vapor lidar measurements. Raman scattering is an inelastic quantum-mechanical scattering process, in which the wavelength of the incident radiation is shifted as a result of the interaction of the photons with target molecules. The Raman wavelength shift, related to the exciting laser wavelength (λL), is proportional to the distinct ro-vibrational energy levels and provides a unique fingerprint for each molecule. The Raman scattering process can involve either energy absorption by the molecule, producing Stokes Raman scattered light with less energy (longer wavelength), or energy transferred to the molecule, producing anti-Stokes Raman scattered light with more energy (shorter wavelength). Most atmospheric species are vibrationally active – resulting in a net Raman shift to longer wavelengths (λr>λL), which indicates that atmospheric target molecules gain energy from the radiation field. The most probable Raman shifts for N2 and H2O are at 2330.7 and 3652.0 cm-1, respectively (Whiteman et al., 1992).

The ratio between the water vapor and nitrogen Raman signals yields a mathematical expression for the dependence of Raman signals ratio on water vapor and nitrogen molecular density (NH2O and NN2), namely Pz,λl,λH2OPz,λl,λN2∝RNH2ONN2dσH2O/dΩdσN2/dΩexp⁡-∫0zαλH2O,z′-αλN2,z′dz′, where R is the proportionality constant dependent on the instrument specifications. This equation ignores the temperature-dependent functions required for very narrow bandwidth filters, typically used for daytime operation (see Whiteman, 2003).

The water vapor mixing ratio (WVMR, denoted as w(z), in grams of water vapor per kilogram of dry air) as a function of vertical altitude (z) is proportional to the ratio of the number density of water vapor to nitrogen and is given by (Goldsmith et al., 1998) wz=MWH2OMWDryAirNH2O(z)NDryAir(z)≈MWH2OMWDryAirNH2O(z)NN2(z)/0.78≈0.485NH2O(z)NN2(z). The WVMR equation above can be related to experimentally recorded Raman lidar signals, SG, by comparing Eqs. (3) and (4), leading to the following expression: wz=DSGH2OSGN2, where D (in grams per kilogram) is a constant depending on instrumental specifications, ratio between N2 and H2O backscattering cross sections, N2 mixing ratio, and Raman lidar signals extinction due to the aerosols and air molecules (Dionisi et al., 2009). The D constant is commonly evaluated by comparison with independent measurement (radiosonde) of water vapor mixing ratio (w(z)).

The current system described here builds upon the successes of the autonomous aerosol lidars built over the past decade by ECCC (Strawbridge, 2013). AMOLITE uses a synergistic approach which combines a dual-laser (for redundancy), dual-lidar design (tropospheric ozone DIAL and aerosol lidar) housed in the same trailer. In order to accommodate two lidar systems, the trailer needed to have a slightly larger interior footprint of 2.1 m by 4.3 m long. A picture of AMOLITE, operating in fully autonomous mode, deployed on a field experiment is shown in Fig. 1a. The external infrastructure of the trailer was very similar to previous designs utilizing a meteorological tower, precipitation-sensor-enabled hatch cover, modified vertically pointing radar interlock system and the other safety equipment required for operation of a class IV laser. The main differences in the design were the addition of a second radome to provide safety radar redundancy, larger hatch opening to allow the operation of two lidar receivers simultaneously and a greatly improved heating and cooling system. The second radar system allows one to remotely change between radar sources in the event that a system failure occurs. We found that these radomes would typically last between 2 and 4 years. However, when a failure occurs, the lidar system is shut down for safety reasons until a site visit can be arranged and a new radar system installed. The addition of a second radar reduced system downtime and operational costs. The larger hatch not only is necessary for dual-lidar operation, but it was also modified to allow the wiper system to operate while the hatch is either open or closed. It was also designed to accommodate exterior blower fans to prevent the accumulation of insects on the window attracted by the UV laser light. The most significant upgrade was the addition of two Mitsubishi Mr. Slim ducted units capable of delivering between 6000 and 24 000 BTU of cooling as well as heat units mounted in the duct allowing an operational range of -40 to +40 ∘C. The ducting allows for better distribution of cool and warm air, maintaining a much more thermally stable environment throughout all the seasons of operation. The internal infrastructure of the trailer followed the early design of rack-mounted components and a single optical bench. The optical bench layout (see Fig. 1b) was large enough to mount both lidar systems including the two laser sources per lidar. The details of the optical bench layout are discussed in Sect. 3.2 and 3.3. The main improvements of the trailer infrastructure were the inclusion of a battery-operated propane furnace and charger capable of maintaining trailer heat for at least 48 h in the event of a power failure. This is particular important should there be a power failure during the winter season, which can leave the trailer without heat for hours at a time, causing the laser coolant to freeze and in turn resulting in severe damage to the lasers. The other major change was the analog-to-digital computer card with a modular Advantech ADAM I/O system with greater flexibility and robustness. These improvements to the trailer infrastructure provided a more stable, reliable environment for improved data quality and uptime.

Since the aerosol lidar design described in Strawbridge (2013) was the backbone of this new system, only the changes will be discussed. The main differences are adding a laser for redundancy and adding an additional transmitted wavelength (355 nm), which in turn add the ability to acquire more particle information, and a water vapor channel arising from the Raman-shifted 355 nm output (407 nm) to provide nighttime water vapor profiles. The second identical laser, a Continuum Inlite III Nd:YAG operating at 20 Hz (see Fig. 1b), shares the same steering mirror (see Fig. 2) as the primary laser and can therefore be engaged remotely by a computer-controlled interface. The folding mirrors and steering mirror, manufactured by Blue Ridge Optics, are triple-coated (anti-reflection coating at 355, 532 and 1064 nm) and 50 by 6 mm flat, have a high damage threshold, and are mounted in a Thorlabs mount with encoded Thorlabs actuators to permit remote alignment if necessary. A schematic of the aerosol lidar in Fig. 2 shows the transmitter beam path and receiver design. The receiver was designed to image the aperture on the photomultiplier tube rather than the field stop. This is necessary to avoid signal modulations due to the inhomogeneous sensitivity of the cathode. The Continuum laser has an output energy of at least 65 mJ at 355 nm, 65 mJ at 532 nm and 100 mJ at 1064 nm. The seven-channel receiver (see Fig. 2) measures the backscatter at each of the emitted wavelengths as well as the depolarization at 355 nm, the nitrogen Raman channels at 387 and 607 nm, and the water vapor Raman channel at 407 nm. All of the channels except the 1064 nm channel use Licel photomultiplier tubes coupled into a Licel analog–photon-counting transient recorder to increase the dynamic range. The 1064 nm channel is focused onto a Perkin Elmer C30956E avalanche photodiode (APD). The APD incorporates a logarithmic amplifier (25 mV rms noise), made by Optech Inc., to increase dynamic range. The amplifier was calibrated prior to the experiment via a transfer function, to convert the signal to a linear scale, in addition to second-order corrections provided by Optech Inc. The signal is directed into a 14 bit Gage CompuScope computer card. The 1064 nm channel is generally used for qualitative information only because of issues such as APD sag and higher noise background. Both the Licel transient recorder and Gage computer card were externally triggered by the same Stanford Research Systems delay generator. The collected data are averaged to produce aerosol profiles from 100 m to 15 km above ground level (a.g.l.) every minute and water vapor profiles from 100 m to 10 km a.g.l. every 5 min.

The ozone DIAL system optical bench layout and detector design is shown in Fig. 3. A dual-laser design is also used for redundancy and can be engaged remotely by a user-controlled translation stage that moves the folding mirror in and out of the optical axis of the transmitter. The folding mirrors have an anti-reflection coating at 266 nm. The lasers are Continuum Inlite III Nd:YAG operating at 20 Hz with an output energy specification of 45 mJ at 266 nm. The laser pumps a 1 m long CO2-filled Raman cell (Nakazato et al., 2007) manufactured by Light Age. The two 45∘ mirrors together provide enough adjustment to align the laser beam to the optical axis of the Raman cell. The multi-wavelength output from the Raman cell is directed to the zenith by a steering mirror that is broadband coated from 266 to 320 nm. This 50 mm optic mounted in a Thorlabs mount with encoded Thorlabs actuators has a user-controlled interface to permit remote alignment if necessary. The differential pair chosen for the DIAL is the second and third Stokes lines from the Raman conversion, namely 287 and 299 nm. The two wavelengths are separated out via the detector block where the signals from the Licel photomultiplier tubes are directed into a Licel analog–photon-counting transient recorder. Again, the optical design imaged the aperture onto the photomultiplier tube for the same reason discussed in Sect. 3.2. A slight delay is imposed on the DIAL using a Stanford Research Systems delay generator to minimize cross-talk between the two lidar systems. The single-telescope design is capable of measuring ozone as low as 400 m a.g.l. to altitudes reaching 15 km during the night every 5 min. It operates 24 h a day, 7 days a week, except during precipitation events. The system is operated remotely, and the data are updated hourly to a website, providing near-real-time capability.

The raw data for AMOLITE are acquired every minute with a vertical resolution of 3.75 m. The data are then averaged (10 min for color-coded plots and sometimes longer for individual profiles) and processed using a boxcar filter to produce a simple smoothing of the raw data, followed by a second-order Savitzky–Golay convolution to compute the derivative with respect to altitude of the signal ratio and βon(z)βoff(z). Although the Savitsky–Golay approach may cause issues at the top of the stratospheric ozone profiles (Godin et al., 1999), it does not have as much of a negative impact for tropospheric ozone due to the vertical structure of ozone typically increasing at the top of the profile. This is primarily due to the signal-to-noise ratio being large enough at most altitudes. Alternate, more sophisticated filters are being considered (Leblanc et al., 2016a) and may be implemented in future data versions, but for now all TOLNet lidars are using the same approach. The boxcar smoothing used on AMOLITE data is a simple first-pass noise removal technique where a centered smoothing window is moved along the lidar signal profile and the average value across the window is calculated for each altitude. The averaged values then become the resulting smoothed profile. The size of the smoothing window starts at 10 bins and increases slightly with altitude (e.g., window is 150 at 12 km) to compensate for the lower signal-to-noise ratio encountered at increased range. The ozone data are also dead-time-corrected using a value of 4 ns. The background correction was determined by using the average background value calculated over a 10 km range starting at 35 km. For the Rayleigh extinction term we used the formulation described by Eq. (2.25) from Kovalev and Eichinger (2004). Also, the temperature-dependent ozone cross sections, at the AMOLITE wavelengths, were introduced using the Brion–Daumont–Malicet (BDM) values found in Weber et al. (2016). The BDM values were interpolated onto a 0.01 nm × 0.1 K grid.

The ozone data was not corrected for SO2 interference. There are only a few times during all the time periods presented in this paper where the ground level concentration of SO2 was above 5 ppbv. These events generally only lasted an hour or two and were thought to be associated with the industrial plumes. Unfortunately, we did not have vertical profile information for SO2, which can be highly variable in the lower troposphere, and so these data were screened out of the ozone plots. The clouds and regions near sharp aerosol gradients were also screened out of the ozone plots. Discussions are underway within TOLNet to reach a consensus on how to correct the ozone DIAL profiles when aerosols are present.

When undertaking a system validation, it is important to compare the final ozone profiles between DIAL systems and determine whether the differences are instrumental and/or algorithm dependent. As a result, AMOLITE's ozone algorithm was tested against a standardized algorithm developed for the SCOOP validation campaign. The first step required a data importer to be written that could read the simulated data into the AMOLITE algorithm. The simulated data included both the simulated lidar data and simulated sonde profiles. Next a boxcar smoothing that is applied to the AMOLITE data was turned off as there is no equivalent in the standardized algorithm. The algorithm testing began by turning off the dead-time correction (saturation), background correction, Savitzky–Golay smoothing, Rayleigh extinction correction and temperature-dependent ozone absorption cross sections (constant values were used for both wavelengths), leaving only the bare-bones ozone calculation. The concept was to use the simulated input in both the AMOLITE and standardized algorithms, comparing the results to the original simulated ozone profile with each algorithm. With all of the above terms turned off, the results matched perfectly after it was ensured that all unit conversions were done correctly and verified that both algorithms were using the same resolution functions. The next test involved using a different simulated ozone profile with saturation turned on. Comparing this to both algorithms with dead-time correction set to 4 ns gave confidence that the algorithms were both handling the saturation effects correctly. The next test involved turning off all the terms except the Rayleigh extinction correction and testing this new simulated ozone product against both the algorithms. Once it was established that both algorithms were calculating the Rayleigh profile from the simulated sonde input, the output matched with less than a 0.05 % bias, acceptable and not unexpected from math rounding errors. Proceeding to the next test, all terms were turned off except for the temperature-dependent ozone absorption cross sections. Here is was important to make sure the wavelengths of the system were taken to sufficient accuracy to minimize errors in the values picked form the standardized look-up table. In our case the wavelength values were set to the AMOLITE DIAL wavelengths of 287.20 and 299.14 nm. Once again with a successful outcome the final test was to turn on random (Poisson) noise and added sky background to the simulated ozone profile. For this final test all the terms were turned off except the background correction, and a second-order Savitzky–Golay convolution was applied, yielding a final result within 0.2 %. The end result of this testing gave us confidence that the AMOLITE ozone algorithm was performing flawlessly. Details of the results and comparisons to the other TOLNet lidar systems will be presented in the SCOOP validation paper (manuscript in preparation).

The performance of the ozone lidar was evaluated through an intercomparison study with four other tropospheric ozone lidars, all of which are part of TOLNet. The campaign named SCOOP took place at the JPL Table Mountain Facility in Wrightwood, California. This provided an opportunity to compare AMOLITE ozone profiles between other lidar instruments and 14 ozone sondes launched during the study. The vertical resolution of the ozone lidar was chosen to be range dependent to provide sufficient detail in the lower troposphere as well as providing ozone profile information to altitudes reaching the tropopause, where the return signal is significantly weaker. Figure 4 shows the effective range-dependent resolution obtained using the algorithm developed by Leblanc et al. (2016a). The left y axis shows the effective resolution during SCOOP in meters above sea level that was applied to the AMOLITE ozone data in Fig. 5. Figure 5a represents a 30 min average of the lidar data starting from the time of the sonde launch at 04:01 UTC on 10 August 2016, and Fig. 5b is also a 30 min average at 21:03 UTC on 16 August 2016. These two profiles were shown to represent the typical results contrasting the range of the ozone DIAL during nighttime and daytime operation. Typically, the DIAL measurements at night will reach a range of over 10 km a.g.l. and dip to 7 km a.g.l. around midday, when the solar background is high. The agreement between AMOLITE and the ozone sonde on both days is very good, with the lidar generally staying within approximately 10 %–20 % of the ozone sonde values and no obvious bias throughout the profile. There are a few regions, notably around layer transitions, where the difference reaches 50 %. This is often due to the difference in vertical resolution of the two instruments. Note the sonde data are plotted at the highest vertical resolution available. It is also important to note that the geophysical separation of the sonde at altitudes of 12 km above sea level is 20–30 km for these cases, which can easily account for the larger differences between the sonde and lidar as the altitude increases. On some days during the study the lidar–sonde agreement varied significantly, particularly at the higher altitudes, due to the large geophysical separation of the two measurements. This is shown in Fig. 6, which represents the average of all 14 lidar–sonde comparisons. The middle panel clearly shows that up to 8 km the lidar agrees to within 5 % of the sonde, with larger differences aloft where there are fewer number of coincidences and the geophysical separation with the sonde increases.

The entire SCOOP campaign is captured in the false-color ozone DIAL plot shown in Fig. 7. AMOLITE was the only fully autonomous lidar operating during SCOOP. The advantages of a fully autonomous lidar system are easily recognized in its ability to capture a continuous dataset throughout the complete diurnal cycle while capturing the dynamics and mixing of long-term events. The ozone DIAL reaches the lower stratosphere, enabling observations of STT events. The signal-to-noise ratio was affected during 11–14 August, when there was an air conditioner failure. The outside temperature reached over 30 ∘C, and the single remaining air conditioner was unable to keep up with the cooling demand of two lidars operating simultaneously. A decision was made to turn off the aerosol/water vapor lidar for the remainder of the study to focus on the ozone intercomparison.

The two color-coded plots in Fig. 8 show the advantage of coincident measurements of ozone and water vapor: in this case, a stratospheric intrusion which starts just after 12:00 UTC on 10 August, descends to approximately 4 km above sea level and persists for over 3 days. The water vapor plot (see Fig. 8b), even though it represents nighttime measurements only, clearly shows the very dry air coincident with the high ozone concentrations of the stratospheric intrusion. The water vapor measurements below 4 km on 10 August show very dry air (and high ozone values), which may also represent a prior stratospheric intrusion, followed by a more defined boundary layer with an increase in water vapor, more typical of boundary layer air. The water vapor channel was calibrated as described by Al Basheer and Strawbridge (2015) using the SCOOP radiosonde data.

An uncertainty in the ozone concentration from AMOLITE can be calculated mathematically for several components. For consistency with other DIAL systems within TOLNet, the uncertainty calculation was based on the paper by Leblanc et al. (2016b). For a detailed description of the mathematical formulations please refer to that paper. In brief, the total uncertainty determined for AMOLITE (e.g., see Figs. 5 and 6) was based on six different components: uncertainty due to detector noise, uncertainty due to saturation, uncertainty due to the Rayleigh cross section, uncertainty due to the background calculation, uncertainty due to the ozone cross section and uncertainty due to the air number density. To calculate these uncertainties, one must also make estimates of dead-time error (estimate: 10 %), the Rayleigh error (estimate: 1 %), the sonde pressure uncertainty (estimate: 20 Pa) and the temperature uncertainty (estimate: 0.3 K). The AMOLITE uncertainty calculations, for each individual uncertainty, were successfully compared to the standardized algorithm uncertainty for a test profile. The altitude at which the AMOLITE ozone profiles get truncated is based on a total uncertainty threshold value, chosen to be 15 % based on AMOLITE–sonde comparisons. There is no threshold value set for the ozone false-color plots. This can sometimes provide additional context for the existence of layers, particularly at higher altitudes.

Stratospheric intrusions were frequently observed throughout the year, with sometimes three or four occurrences per week. In recent years, there has been more understanding about the mechanism that enables these STT events (Langford et al., 2018). However, there is still very little data on the frequency and magnitude of these events and their impact on the tropospheric ozone budget. For example, Fig. 9 shows three false-color plots of ozone, water vapor and aerosol backscatter ratio for the bottom 10 km of the atmospheric from 6 to 13 November 2016. During this week-long period two stratospheric intrusions were observed (and evidence that a third started on 13 November). The white areas on the ozone plot, represent cloudy regions where the DIAL system is unable to retrieve ozone values. These white areas correlate very well with the cloud regions displayed in the aerosol backscatter ratio plot. The water vapor plot shows dry air (less than 0.2 g kg-1) coincident with the higher ozone concentrations of the stratospheric air reaching down into the moist regions more typical of the lower atmosphere. During most of the stratospheric intrusions over the Oski-ôtin site, it was noted that, although the free-tropospheric ozone levels were increased significantly, the ozone intrusion does not always penetrate the boundary layer and increase surface values.

A series of ozone vertical profiles during the stratospheric intrusion between 9 and 11 November is plotted in Fig. 10. This plot shows the ozone concentration before the intrusion (red line) where the typical background value of approximately 30 ppbv is present in the lowest 4 km. As time progresses, one can clearly see the high ozone concentration, reaching 120 ppbv on 10 November at 00:00 UTC, from the stratospheric transport descending down to lower and lower altitudes. The impact increased the tropospheric budget by almost a factor of 2. Figure 11 shows only the lowest 4 km of the ozone plot compared to the ground level observations of ozone and NOx. There is reasonably good agreement between the ground level measurements and the DIAL measurements around 600 m (the lowest few lidar bins can be unreliable as they are strongly dependent on the alignment and temperature fluctuations inside the trailer). It is also important to consider the height of the boundary layer (see Fig. 12a), which during the wintertime can be significantly lower. The ozone–NOx relationship in Fig. 11b is not the typical diurnal relationship that can be observed, in part due to the stratospheric intrusion event, but also due to industrial plume sources impacting the site. For several hours on 7, 8 and 9 November the ozone values approach 0. There is an increase in ozone not only during the daytime hours (solar day is approximately 14:00 to 00:00 UTC during this period) but also during the nighttime on 10, 11 and 12 November, when the stratospheric intrusion occurred. Figure 12 shows the aerosol lidar plot for the lowest 4 km along with various chemical and particulate tracers from CAM1. The aerosol lidar plot gets down to approximately 100 m above ground level, which during the winter months is necessary to observe the boundary layer and plume dynamics. There is a good correspondence between the increase in aerosol shown by the lidar and the PM2.5 trace over the entire period. The increase in particle concentration is linked to the presence of the plume impacting the site. For example, on 7–9 November and the night of 10 November the aerosol lidar observations show an increase in concentration in the lowest 750 m (see Fig. 12a) typical of industrial plume sources. As the plume impacts the ground directly, there is a substantial bump in the PM2.5 concentration. Figure 12c and d also indicate that the air is from an industrial source where the CO2, CO, CH4 and sulfur compound concentrations are high. This is the first example where the vertical context given by the lidar aids in the understanding of the ground-based measurements.

Another occurrence that can change the ozone budget is forest fires (Jaffe and Wigder, 2012). During the period of 29–31 August, smoke from a forest fire was advected into the region as shown in the MODIS (Moderate Resolution Imaging Spectroradiometer) true-color image acquired form the Terra satellite on 31 August 2017 (see Fig. 13). The ozone plot shown in Fig. 14a presents a significant amount of ozone in the free troposphere. The enhanced ozone signature on 29 August is from a stratospheric intrusion, whereas the enhanced ozone on 30 and 31 August is a result of forest fire smoke. The extent of the forest fire smoke is shown by the large aerosol burden in Fig. 14b coincident with the ozone as well as the depolarization ratio plot in Fig. 14c, showing a value of about 5 %, consistent with other smoke plume measurements (Aggarwal et al., 2018).

The diurnal cycle of ozone over 3 days is shown in Fig. 15b with increased ozone due to the smoke impacting the surface around 00:00 UTC 31 August, from what we hypothesize to be enhancement from the forest fire. In Fig. 15a a series of ozone traces at different altitudes from the DIAL measurements are plotted against the ground ozone values. In this plot the ozone aloft tracks the ground level ozone quite well until the ozone-enhanced air from the forest fire smoke begins to descend over the site. The noisy ozone values around the 1000 m level are a result of an error in ozone when the aerosol concentrations were very high (see Fig. 14a and b around 15:00 to 17:00 UTC on 31 August). There is also evidence that the smoke impacted the surface from 00:00 to 18:00 UTC on 31 August as shown in Fig. 15c–e, where an increase in H2S, PM2.5 and CO also occurs. An alternative way to plot aerosol lidar data is to plot extinction coefficients instead of backscatter coefficients. Since we are measuring the nitrogen Raman channel during the nighttime, we can calculate the backscatter coefficient; extinction coefficient; and extinction-to-backscatter ratio, also known as the S ratio. The S ratio is a useful quantity for determining the aerosol type (see Strawbridge, 2013). The three-panel plot in Fig. 16 shows the 355 nm backscatter coefficient, extinction coefficient and S ratio for 03:00 to 12:00 UTC (nighttime) on 31 August 2017 using 10 min average data. The near-field overlap is corrected, and the data are plotted in kilometers above mean sea level (m.s.l.), primarily because the atmospheric density obtained from sonde data is also relative to m.s.l. The white noisy regions aloft on the extreme left and right are artifacts due to the increase in sky background. The backscatter coefficient plot reveals the dynamic nature of the smoke plume between 1 and 5 km and a cirrus cloud layer between 8.5 and 11 km. The extinction coefficient plot is useful because one can directly relate it to aerosol optical depth by integrating along the altitude range. The S ratio plotted as a 10 min average shows extraordinary detail within the smoke plume, with values ranging approximately from 40 to 65 sr. These values are consistent with the value of 45 to 65 sr reported by Barbosa et al. (2014) and are consistent with several other observations provided in Table 3 of Ortiz-Amezcua et al. (2017). Figure 16 also shows the boundary layer aerosols with an S ratio of 20 to 35 sr, indicative of larger particles in the moist boundary layer air (see water vapor plot in Fig. 14d), and 10 to 15 sr in the cirrus cloud. Figure 17 shows a 1 h average taken between 08:00 and 09:00 UTC. For highly variable conditions such as a forest fire plume, the 1 h average may result in underestimating the maximum S value. It is also very difficult to measure the S value in the free troposphere when there is very little aerosol present, such as in this case. Those values will be very noisy and have been discriminated out of the dataset shown here.

The ozone plot for 4–9 September 2017 (see Fig. 18a) shows several processes occurring throughout the entire altitude range. There is a stratospheric intrusion on 4 September that extends in 5 September (see dry air in Fig. 18d). The increased ozone in the free troposphere from 6 to 9 September is due to the forest fire activity being advected back into the region. The forest fire smoke is clearly visible in the aerosol backscatter plot (see Fig. 18b) and the depolarization ratio plot (see Fig. 18c). There is also a fairly dominant feature between 800 and 2200 m where the ozone values reach very close to 0. There are also time periods where these near-0 ozone features appear to reach closer to ground level (12:00 UTC on each day during the 4–7 September period). This is also shown in Fig. 19a, where the ozone values from the DIAL at 500, 700 and 900 m are plotted against the ground level ozone. The very low surface ozone around 12:00 UTC on 7 September remains low well up into the lower troposphere. The ozone data around 12:00 UTC on 8 September are an artifact due to the very high aerosol loading and have been removed. The surface ozone levels (see Fig. 19b) on 4 September ranged from a low of 10 ppbv around 12:00 UTC to 20 ppbv. Figure 20 shows the winds were primarily coming from the north, where there are fewer industrial sources to impact the ground site. However, on 5 September the winds were coming from the south, where the industrial sources impact the site as shown by the increase in NOx (see Fig. 19b), sulfates (see Fig. 19c), PM2.5 (see Fig. 19d) and CO2 (see Fig. 19e).

The ground level ozone increased to 70 ppbv around 18:00 UTC on 7 September and dropped to 50 ppbv around 03:00 UTC on 8 September, which was mostly due to the southerly wind (see Fig. 20) bringing the industrial plumes to the ground site. The DIAL ozone shows ozone levels reaching 80 ppbv within 500 m of the surface. There is also an increase in SO4 and CH4 during this time period. The diurnal ozone cycle is very well established throughout this entire study period, except when the elevated ozone from the forest fire smoke is mixed down to the surface starting around 06:00 UTC on 8 September (note the increase in NO2 but not NO). The increase in ground level ozone throughout the nighttime reaches values of up to 35 ppbv. There is also a steep increase in PM2.5 levels (from 25 to 50 µg m-3) and CO around 15:00 UTC on 8 September coincident with the lidar backscatter ratio plot shown in Fig. 18b indicative of an increased concentration of the biomass burning plume impacting the ground site. The wind has also shifted from a southerly flow to eventually a northerly flow. The resultant ozone at the ground is a mixture of local chemistry and ozone-rich air transported into the region.

A plot of the backscatter coefficient, extinction coefficient and S ratio from 7 to 9 September (see Fig. 21) shows the contrast between the smoke plume on 8 September and the boundary layer aerosols and industrial plume (around 2 km on 7 September). The smoke plume S ratios are slightly smaller (35 to 55 sr) compared to 31 August, likely indicative of more aged smoke (see the 1 h average plot between 10:00 and 11:00 UTC shown in Fig. 22c). The lidar ratio, S, can also be calculated for 532 nm. However, there is a significantly lower signal-to-noise ratio, so the 10 min average false-color plots were not produced. A comparison was made for a 1 h average when the smoke plumes were present on 31 August and 8 September (see Fig. 23a and b). The S ratio for 532 nm on 31 August ranges from 40 to >100 sr, while on 8 September it ranges from 40 to 70 sr. These values are consistent with the higher 532 nm S ratio values reported in Table 3 of Ortiz-Amezcu et al. (2017). The Ångström exponent (see Fig. 23c and d) is inversely related to the average size of the particles. On 31 August the extinction Ångström exponent was 1.56±0.2 between 2 and 4.2 km, in contrast to 1.35±0.2 between 2 and 4 km on 8 September. These values are consistent with what others have reported for biomass burning (see Table 3 by Ortiz-Amezcu et al., 2017). During this 6-day period it would be very difficult to understand the ground measurements without the vertical context of the lidars.