**5. Aerosol vertical structures monitored by satellite lidar**

#### **5.1. Introduction**

The CALIPSO combines an active lidar instrument with passive infrared and visible imagers to probe the vertical structure and properties of thin clouds and aerosols over the globe [35], to study the climate impact of clouds and aerosols in the atmosphere [36]. The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) is the primary instrument on the CALIPSO satellite, which is the first polarization lidar in orbit and the first satellite lidar [37]. It provides vertically resolved measurements of aerosol distribution with aerosol extinction coefficients [38] and observes aerosols over bright surfaces and beneath thin clouds as well as in clear sky conditions [39].

The CALIOP lidar consists of a laser and a receiver with the stability of the transmitter-toreceiver alignment [40]. The lasers produce simultaneous pulses at 532–1064 nm at a pulse repetition rate of 20.16 Hz, with a pulse length of about 20 ns. Polarization out-coupling provides a highly polarized beam with a beam diameter of 70 m at the Earth's surface [36]. A polarization beam splitter separates the 532 nm parallel and perpendicular returns. The receiver consists of the 1-m telescope and the photomultiplier tubes (PMTs).

The CALIOP acquires vertical profiles of elastic backscatter at two wavelengths from a near nadir-viewing geometry during both day and night phases of the orbit, to derive the accurate aerosol and cloud heights and extinction coefficient profiles. It measures the profiles of linear depolarization at 532 nm, to discriminate ice clouds from water clouds and to identify nonspherical aerosol particles [41]. The aerosol particle sizes can be obtained from the ratios of the signals obtained at the two wavelengths [42].

#### **5.2. The algorithms to retrieve the aerosol vertical structures**

For FMA category (**Figure 18b**), six clusters were observed. Clusters 2 (13%) and 3 (8%) represent the northwest flows, while Cluster 1 represents the flow of the southwest sector (17%). Clusters 4 (23%), 5 (19%) and 6 (20%) have westerly, easterly and north easterly origins, respectively. The most notable difference of trajectories between CMA and FMA is the distance of transport flows. The long northwest trajectories (70%) are dominant for CMA, while the short transport flows (62%) are the most for FMA. This result is attributed to different aerosol types under different air masses. Therefore, anthropogenic aerosols mainly lead to fine mode,

**Figure 18.** Mean HYSPLIT 72-h backward trajectories of 1500 m altitude at 02:30 GMT by cluster analysis during 2001–

The CALIPSO combines an active lidar instrument with passive infrared and visible imagers to probe the vertical structure and properties of thin clouds and aerosols over the globe [35], to study the climate impact of clouds and aerosols in the atmosphere [36]. The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) is the primary instrument on the CALIPSO satellite, which is the first polarization lidar in orbit and the first satellite lidar [37]. It provides vertically resolved measurements of aerosol distribution with aerosol extinction coefficients [38] and observes aerosols over bright surfaces and beneath thin clouds as well as in clear sky

The CALIOP lidar consists of a laser and a receiver with the stability of the transmitter-toreceiver alignment [40]. The lasers produce simultaneous pulses at 532–1064 nm at a pulse repetition rate of 20.16 Hz, with a pulse length of about 20 ns. Polarization out-coupling provides a highly polarized beam with a beam diameter of 70 m at the Earth's surface [36]. A polarization beam splitter separates the 532 nm parallel and perpendicular returns. The re-

2009 at Fuyang station. (a) CMA, Angstrom exponent <1 and (b) FMA, Angstrom exponent > 1.

**5. Aerosol vertical structures monitored by satellite lidar**

ceiver consists of the 1-m telescope and the photomultiplier tubes (PMTs).

**5.1. Introduction**

130 Aerosols - Science and Case Studies

conditions [39].

and natural aerosols dominate the long northwest trajectories.

The fundamental algorithms of the CALIPSO are profile processes, to locate all layer boundaries and to identify each layer as being either cloud or aerosol. There are three types of algorithms related with the layer detection, the feature classification and the optical properties analysis. The data products include layer heights (e.g. feature top and base altitudes), layer identification (i.e. clouds versus aerosols, ice clouds versus water clouds), layer structures of clouds, aerosol backscatter and extinction coefficients.

The first step is to identify the layer types of clouds and aerosols. Thus the layer detection algorithm is used to identify regions of enhanced scattering, and to record simple characteristics of these atmospheric features, and to detect the vertical location of several different classes of geophysical objects. Specifically, the lidar returns are used to obtain information on the base and top altitudes of clouds and aerosol layers [43]. The layer finding algorithm is based primarily on separating the genuine features from the pseudo-features to find layers repeatedly on several passes through the data. The distribution of the roughly separated aerosols with clouds is shown in **Figure 19**. The regions with white are recognized as clouds and the regions with yellow are taken as aerosols. The intensified backscattering values of clouds are centred at 1.5 km and those of aerosol layers are centred at 0.5 km.

**Figure 19.** Aerosols and clouds can be roughly separated from the satellite data on Aug. 18, 2015.

#### **5.3. Scene classification algorithm**

The second processing step is to classify the layers and identify the types. The scene classifi‐ cation algorithm (SCA) is used to determine the types of the features (tropospheric or strato‐ spheric) by checking the base altitude of the feature. The tropopause altitude is derived from ancillary data obtained from the global modelling and assimilation office and used to classify the tropospheric feature depending on whether the feature base is lower than this altitude, while the stratospheric feature is classified by that higher than this altitude. The further classification algorithms are conducted to sub‐type the feature for the tropospheric feature.

The SCA is used to determine the aerosol layer or cloud layer, primarily based on the scattering strength and the spectral dependence of the lidar backscattering, which are primarily obtained from the mean value of the attenuated backscatter coefficient and the ratio of the mean attenuated backscatter coefficients (the attenuated colour ratio) at 1064–532 nm (**Figure 20**). If the layer is classified as cloud, the SCA will then determine whether it is an ice cloud or water cloud using the backscatter intensity and the depolarization ratio profiles together with ancillary information of the layer height and temperature. The SCA will also use a combination of observed parameters and a priori information to select an appropriate extinction‐to‐ backscatter ratio and multiple scattering function required for subsequent process of the extinction and optical depth retrieve.

**Figure 20.** The attenuated colour ratio at 1064–532 nm for determining the layer type.

#### **5.4. Retrieved optical properties analysis**

The last processing step is to retrieve the extinction coefficient and optical depth from the calibrated, range‐corrected lidar signal. Some widely used algorithms have been developed, based on the Fernald method [44], the Klett method [45] and the so‐called linear iterative method [46]. The Fernald and Klett methods offer analytic solutions with closed form, while the linear iterative technique is a simple numerical solution. Both Fernald and Klett methods are developed from the single scattering analyses to the multiple scattering analyses using a correction factor to the range-resolved extinction coefficients. The CALIPSO algorithms account for multiple scattering by applying a correction factor derived from the phase functions of aerosol models [47]. These methods are used to retrieve the extinction and backscatter profiles of cloud and aerosol layers.

**5.3. Scene classification algorithm**

132 Aerosols - Science and Case Studies

extinction and optical depth retrieve.

**Figure 20.** The attenuated colour ratio at 1064–532 nm for determining the layer type.

The last processing step is to retrieve the extinction coefficient and optical depth from the calibrated, range‐corrected lidar signal. Some widely used algorithms have been developed, based on the Fernald method [44], the Klett method [45] and the so‐called linear iterative method [46]. The Fernald and Klett methods offer analytic solutions with closed form, while

**5.4. Retrieved optical properties analysis**

The second processing step is to classify the layers and identify the types. The scene classifi‐ cation algorithm (SCA) is used to determine the types of the features (tropospheric or strato‐ spheric) by checking the base altitude of the feature. The tropopause altitude is derived from ancillary data obtained from the global modelling and assimilation office and used to classify the tropospheric feature depending on whether the feature base is lower than this altitude, while the stratospheric feature is classified by that higher than this altitude. The further classification algorithms are conducted to sub‐type the feature for the tropospheric feature.

The SCA is used to determine the aerosol layer or cloud layer, primarily based on the scattering strength and the spectral dependence of the lidar backscattering, which are primarily obtained from the mean value of the attenuated backscatter coefficient and the ratio of the mean attenuated backscatter coefficients (the attenuated colour ratio) at 1064–532 nm (**Figure 20**). If the layer is classified as cloud, the SCA will then determine whether it is an ice cloud or water cloud using the backscatter intensity and the depolarization ratio profiles together with ancillary information of the layer height and temperature. The SCA will also use a combination of observed parameters and a priori information to select an appropriate extinction‐to‐ backscatter ratio and multiple scattering function required for subsequent process of the After the layer profiles identified by layer detection algorithm and classified by the SCA, the hybrid extinction retrieval algorithm (HERA) performs extinction retrievals on regions of profile data [48]. HERA incorporates a sophisticated retrieval engine that extracts profiles of particulate backscatter and extinction coefficients from the profiles of attenuated backscatter coefficients identified as layers. One example of particulate backscatter and extinction at 532– 1064 nm is shown in **Figure 21**. Following the retrieval of extinction profiles, other parameters can be computed by HERA, including optical depth, particle colour ratio and particle depolarization ratio [49].

**Figure 21.** The left panel means profile of mean extinction at 532 nm; the right panel means profile of attenuation-corrected backscatter coefficient at 532 nm.

Column atmospheric optical depth can be estimated from the measured two-way transmittance at the ocean surface using ocean surface backscatter and collocated wind speed. Once the collocated wind speed is determined, theoretical ocean surface lidar backscatter can be derived from the wave slope variance—wind speed relation. The column optical depth is then derived from the ratio of the CALIPSO attenuated backscatter measurement and theoretical backscatter [50]. The cloud and aerosol optical depth derived from this approach is a direct measurement, without assuming aerosol and cloud physical properties. An example of aerosol optical depth distribution retrieved from CALIPSO data can be seen in **Figure 22**.

**Figure 22.** An example of AOD distribution retrieved from CALIPSO data with the AOD at 532 nm (the left panel) and 1064 nm (the right panel).

#### **6. Conclusions**

The ground reflectance is significantly different over land and ocean, leading to the need for different approaches of the atmospheric correction in order to optimize the results. However, this difference can be expressed by the spectra of the LUT of in situ measurements in the UAC model. Based on the LUT of the ground reflectance, the aerosol reflectance can be obtained by the best nonlinear least square fit function based on the Angstrom law. The reflectance is then used to determine the epsilon spectra that are used to select the two closest aerosol models instead of one epsilon value from the NIR band. The results show that this approach is more robust to overcome the problem caused by small variations in the aerosol reflectance in the two NIR bands. Meanwhile, this approach can also reduce some abnormal epsilon errors caused by factors including the data quality of satellite reflectance and the mismatch between the actual ground reflectance and the LUT. The use of the LUT provides a unified approach for estimating the aerosol reflectance and the ground reflectance over land and ocean. The performance of the UAC model is evaluated using a SeaWiFS image. The results show that the model can completely separate the aerosol scattering reflectance from the radiance at TOA. The relative error is 22.1% when it is validated by the in situ measured AOT data using the MICROTOP instruments, and the error is 31.4% using the four sites of AERONET measured AOT data over the ECS in 2006 and 2007. The UAC model can provide the aerosol products from satellite remote sensing data over land and ocean.

MODIS AOTs at 12 stations around China using the AERONET observations have been evaluated with the result that both agree generally well over ocean and land. The seasonal changes and spatial distributions of aerosol optical properties have also been observed with a notable temporal variation over the China Sea. The AOT reaches maxima in spring and winter and minima in summer and fall, while the FMF reaches maxima in summer and fall and minima in spring and winter. Both AOT and FMF have an obvious spatial distribution over the China Sea. At latitude direction, the AOT appears maxima between 30 and 40°N and the FMF increases from south to north. At longitude direction, both AOT and FMF decrease with longitudinal increasing. Over the land, the lowest AOT values are observed during the autumn and the highest during the spring. The highest AOT, appearing in June, is possibly related to burning of agricultural residues. The Angstrom exponent appears maxima in summer and minima in winter. The higher AOT and lower Angstrom exponent are observed in a large plain region at the north and middle, while smaller AOT and higher Angstrom exponent can be found over the mountainous region, which is obviously influenced by topography. Meteorological conditions obviously affect the aerosol optical properties. Continent aerosol is carried to the China Sea by wind. The influences of different air masses are great disparity for aerosol optical properties over the land.
