Comparisons of aerosol backscatter using satellite and ground lidars: implications for calibrating and validating spaceborne lidar

The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument on the polar orbiter Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) is an elastic backscatter lidar that produces a global uniformly-calibrated aerosol data set. Several Calibration/Validation (Cal/Val) studies for CALIOP conducted with ground-based lidars and CALIOP data showed large aerosol profile disagreements, both random and systematic. In an attempt to better understand these problems, we undertook a series of ground-based lidar measurements in Atlanta, Georgia, which did not provide better agreement with CALIOP data than the earlier efforts, but rather prompted us to investigate the statistical limitations of such comparisons. Meaningful Cal/Val requires intercomparison data sets with small enough uncertainties to provide a check on the maximum expected calibration error. For CALIOP total attenuated backscatter, reducing the noise to the required level requires averaging profiles along the ground track for distances of at least 1,500 km. Representative comparison profiles often cannot be acquired with ground-based lidars because spatial aerosol inhomogeneities introduce systematic error into the averages. These conclusions have implications for future satellite lidar Cal/Val efforts, because planned satellite lidars measuring aerosol backscatter, wind vector, and CO2 concentration profiles may all produce data requiring considerable along-track averaging for meaningful Cal/Val.

Data Analysis CALIOP Level 1.5 data is reported as total attenuated backscatter in units of km -1 sr -1 , viewing from the top of the atmosphere downward. Each profile contains an average of 60 individual attenuated backscatter profiles from the Level 1 data product merged with cloudand-aerosol-layer detection from the Level 2 data product. As such, Level 1.5 data has its own unique spatial averaging and resolutions. The averaged profile, which covers a 20-km horizontal distance along the ground track, is reported at 60-m altitude intervals from -500 to 20,000 m mean sea level (MSL) 2 .
ASC lidar data is exported in terms of range-corrected, background-subtracted digitizer counts for each 80,000-pulse average in two receiver channels: short-range (SR) and long-range (LR). The SR channel has better signal-to-noise ratio (SNR) up to 2-2.5 km while the LR channel is better above 2-2.5 km. Both channels contain vertical profiles to 30 km in 15-m altitude increments. To obtain the best SNR over the entire vertical profile, we merged the SR and LR data at an altitude where both channels had achieved full crossover and had high SNR.
In elastic backscatter lidar measurements, the signal from any altitude is attenuated by the atmosphere between that altitude and the lidar, due to molecular scattering as well as aerosol scattering and absorption. For this reason, nadir-viewing and zenith-viewing lidars observing the same atmosphere are not directly comparable, so the zenith-viewing ground lidar data must be converted into the downward view of CALIOP. In some circumstances, this conversion process distorts the profile 3 . To convert our profiles, we used the equations described by Grigas et al. 4 to calculate the profiles for backscatter β(z) and extinction α(z) of both the molecular (mol) and aerosol (aer) data.
Although we used similar methods, our procedure for obtaining the profiles of βmol(z), βaer(z), αmol(z), and αaer(z) varied slightly from Grigas et al. 4 Elastic backscatter lidars, such as the ASC lidar, require a Klett algorithm, constrained by AOD values, to separate βaer(z) 5 . The Klett algorithm requires αmol(z) and βmol(z) as inputs, which we computed using molecular number density from local radiosonde data and cross-sections for extinction and scattering at 523.5 nm. The molecular profiles were interpolated into the same altitude intervals as the ASC data and were converted to 532 nm.
The Klett Algorithm is a recurrence relation that starts at a highest altitude and works downward to the lowest altitude at which the lidar has achieved complete crossover (500 m in our case). At the highest altitude, the value of βaer must be known, so it is convenient to pick an aerosol-free altitude where the value of βaer is zero. To find the altitudes where the atmosphere was aerosol-free, the averaged range-corrected ASC signal was plotted along with βmol scaled to match the lidar signal at the highest altitudes, as shown in Supplementary  Fig. S1. The altitude at which the two curves merged was taken as the lowest aerosol-free altitude. Figure S1. An example fit of the molecular backscatter (blue) to the ASC lidar signal (red). The Klett algorithm also requires Sa to be known. For polluted urban aerosol, CALIOP uses Sa = 70 sr, so this value was used in all of our retrievals except when the aerosol level was very low, in which case we assumed clean continental aerosol (Sa = 35 sr) 6 . After retrieving the βaer(z) profile for each episode, we found αaer(z) according to Eq. 6 from Grigas et al. 4 To validate the assumed Sa, we extrapolated our αaer(z) profile down to the surface and then integrated it from the surface to the top. The resulting lidar AOD was then compared to the value measured at the Georgia Tech AERONET station. The AERONET values are displaced in time from the nighttime lidar measurements, so they may not be very representative for those data sets. The comparisons are shown in Supplementary Table S2. On all occasions but two (14 and 16 January 2014), the Sa value we used matched the reported CALIOP aerosol type. 16 January 2014 was the only nighttime data set with very high boundary layer signals, so we used a polluted air value. 14 January 2014 had an active boundary layer that CALIOP did not see, so we used the polluted value even though CALIOP saw clear air. All of the other Sa values match CALIOP's reported aerosol class.
Once βaer(z) and αaer(z) were determined for the ASC lidar's altitude range, the coefficients in the aerosol-free region were set to zero while the coefficients at other altitudes were translated from 523.5nm to 532 nm using the AAE measured at GTRI's AERONET station. Total backscatter was calculated using βmol(z) and βaer(z). Total extinction was calculated using αmol(z) and αaer(z), as well as ozone extinction. The ozone number density is included in the Level 1.5 data products, but only up to 20 km, which includes approximately one-half of the total ozone. For this reason, ozone data in Dobson units was retrieved for each episode from the World Ozone and Ultraviolet Radiation Data Centre. Dobson units are column-integrated, and one unit corresponds to 2.69 x 10 20 molecules/m 2 . The ozone cross section at 532 nm is 2.7 x 10 -25 m 2 /molecule 7 , yielding an optical depth per Dobson unit of 7.3 x 10 -5 .
The two-way transmittance was calculated using the total extinction values and then multiplied by the total backscatter to find the total attenuated backscatter for a nadir-viewing lidar. We smoothed our values to 60-m resolution using a moving-average filter. Both the ASC and CALIOP total attenuated backscatter data were expressed in units of Mm -1 sr -1 . The CALIOP Level 1.5 data were averaged spatially using the five closest 20-km overpass data sets in order to improve their low SNR. The profiles were then plotted on the same graph for each episode. Results Supplementary Fig. S2 shows our comparisons of the nadir-viewing attenuated backscatter profiles converted from our ASC lidar data with the corresponding CALIOP profiles. Figure S2. Comparisons of lidar profiles. Mean 532-nm attenuated backscatter profiles are shown for CALIOP (red) and the ASC lidar (blue).