Retrieval of Aerosol Size Distribution from Microtops II Sunphotometer in Hong Kong

A new method has been developed for retrieving Aerosol Size Distribution (ASD) from spectral aerosol optical thickness (AOT) measured by Microtops II sunphotometer. The rationale of this method is based on a Spectral Extinction Fitting Technique (SEFT) between measured and model-simulated aerosol extinctions. Numerical simulation of aerosol extinctions includes seven aerosol microphysical properties and the resulting total of 64,324,260 datasets were used to compare with Microtops II sunphotometer AOT measurements at five wavelengths (340, 500, 675, 870, and 1020 nm). This SEFT method works directly on AOT measurements and does not require any specific inputs of aerosol properties. The accuracy was examined using AERONET inversion products, and the results show a difference between SEFT and AERONET inversion products of bimodal lognormal volume size concentrations, of ± 10%. The SEFT algorithm was applied for two specific conditions (hazy and clear atmospheres) in Hong Kong, and the corresponding ASD retrievals show reasonable agreements with AERONET inversion products.


INTRODUCTION
Aerosols are defined as particles suspended in the atmosphere in either liquid or solid form.They have different size distributions, shapes and residence times, and are from different sources.The study of aerosols is important because of their influence on the earth radiation budget, climate change, atmospheric conditions and human health.Recent research has focused on studying the aerosol size distribution (ASD) and this is an important factor in climate studies.Particle size has a major influence on the particle scattering, and the estimation of ASD is one of the major inputs for aerosol retrieval from satellite images (Soufflet et al., 1992;Kaufman et al., 1994).Because of spatial and temporal variability of atmospheric aerosols, measurements of ASD traditionally rely on local observations.Although several well-known algorithms to determine aerosol optical thickness (AOT, τ), ångström exponent (α), single-scattering albedo (SSA) from spectral radiation measurements have been developed (Rao et al., 1987;Kaufman and Sendra, 1988;Kaufman et al., 1990;Liousse et al., 1995;Isakov et al., 1996;Liu et al., 1996;Lee et al., 2007), there is still no direct method for retrieving aerosol microphysical properties using remote sensed aerosol extinction measurements.
Over the past decades, determinations of ASD from spectral radiation measurements have been used by complex numerical inversion methods.King et al. (1978) used iterative inversion procedures which provide fairly accurate information about the aerosol size distributions, if the Lagrange multiplier, refractive index, and radius range are carefully selected (King, 1982).Nakajima et al. (1996) and Dubovik and King (2000) also used numerical inversion techniques with additional angular observations to derive size distribution information.Kassianov et al. (2007) developed an alternative approach using combined direct and diffuse irradiance measurements for determining aerosol size distribution.However, this requires massive computational power and is also time-consuming.
In this study, a novel method, the Spectral Extinction Fitting Technique (SEFT) is proposed, which is based on look-up tables (LUTs) for retrieving ASD from spectral aerosol optical thickness measurements and can significantly improve the efficiency of data retrieval.Using the SEFT method, ASD retrieved from Microtops II hand-held sunphotometer measurements can be compared to the size distribution products from the AErosol RObotic NETwork (AERONET) database (http://aeronet.gsfc.nasa.gov)by sunphotometer measurements (Dubovik and King, 2000).
The proposed retrieval method is given in the Methodology section and potential biases in the retrieval arising from input parameters are discussed in the Results section.

STUDY AREA AND DATA USED
The skies over Hong Kong are obscured for 20% of the time in a year due to frequent episodes of reduced visibility (visibility less than 8 km) (Chang and Koo, 1986;Lai and Sequeira, 2001).These low visibilities are thought to be due to the interactions between aerosols including anthropogenic, sea salts and long-distance dust, and water vapor (Bell et al., 1970).During the dry seasons, northerly and northeasterly winds generally bring continental pollution from neighboring industries in the Pearl River Delta (PRD) region of the Chinese mainland to Hong Kong.The PRD region is deemed to be the main source of anthropogenic emissions in south China, with high average PM 10 concentrations above 200 µg m -3 in winter and PM 2.5 concentrations of around 100 µg m -3 in autumn (Wei et al., 1999;Cao et al., 2004).These values are comparable with the World Health Organization PM 2.5 24-hour standards of 35 µg m -3 .
This study used sunphotometers at two Hong Kong AERONET sites (Hong Kong PolyU -operating from 2005 onwards, and Hong Kong Hok Tsui -operating since 2007), as well as a Microtops II handheld sunphotometer at the Hong Kong International Airport -operating from 2007 onwards (Fig. 1).AERONET is a federated network of ground sunphotometers, comprising CE-318 sunphotometers which measure aerosol extinction every fifteen minutes over multiple wavelengths (Holben et al., 1998).The Hong Kong PolyU AERONET station is situated at the center of the urbanized Kowloon Peninsula (Fig. 1).The Hong Kong Hok Tsui AERONET is deployed on a remote peninsula near the coast for monitoring rural (background) aerosols.

METHODOLOGY
The Microtops II sunphotometer measures spectral solar radiation and converts to AOT values (Morys et al., 2001).By its relatively accurate AOT observation, it also has been used to compare with AOT observed from other instruments (Srivastava et al., 2008;More et al., 2013;Zawadzka et al., 2014).The spectral AOTs carry information about the aerosol extinction property with the size distribution.Using these spectral AOT (τ(λ)) values, the size distribution function of aerosols has been determined by traditional numerical inversion of the integral equation: where Q ext is the extinction efficiency as a function of complex refractive index (m), radius (r) and wavelength of the incident radiation (λ); n(r) is the number density of aerosols which is a function of r. σ ext is the extinction coefficient.In Eq. ( 1), complex refractive index (m) and vertical distribution of aerosol are crucial for aerosol inversion.
The Mie scattering code (Bohren and Huffman, 1983) for spherical aerosol types and the T-Matrix code (Mishchenko and Travis, 1998) for non-spherical aerosol type are used to calculate the aerosol optical properties.In Mie theory, the refractive index and ASD are the key factors to determine extinction coefficient.If the values of ASD and refractive indices are known, the aerosol extinction as well as AOT can be calculated.In order to illustrate the effectiveness of the proposed method, Fig. 2 compares the SEFT retrieved AOT by the known values of ASD and refractive indices from AERONET inversion products with the direct observed AOT.High correlation coefficients (r 2 ≥ 0.86) imply that accurate spectral extinctions derived from ASD and refractive index are critical and essential for the inversion scheme.In other words, AOT observed from the wavelength dependent signal in the transmitted radiance can be used to retrieve the aerosol optical properties.
In this study, a set of hypothetical refractive indices and size distribution values were input to the Mie theory calculation for deriving hypothetical extinction coefficients in LUTs.Fig. 3 illustrates the method.
In the algorithm, LUTs consist of extinction coefficients at five visible wavelengths (340, 500, 675, 870 and 1020 nm) for a variety of ASD and refractive indices.ASD is assumed to be bimodal lognormal with mode radii and standard deviations, Eq. (3).
where i is aerosol mode (fine and coarse mode), N is total column particle number in # cm -3 , and r m and σ are the median radius in µm and standard deviation for each mode.This normal distribution can also be converted to volume size distribution as in Eq. ( 4).For the comparison with AOTs observed from Microtops II and derived from LUTs, a Root-Mean-Square Difference (RMSD) test was used which could select the appropriate size distribution data with the smallest systematic difference (minimum RMSD) (Eq.( 5)).
where τ meas is the measured AOT, τ calc is calculated AOT for wavelength λ i .

RESULTS
AERONET level 2.0 AOT and aerosol inversion data (Dubovik and King, 2000) were collected for AERONET stations at Hong Kong Hok Tsui (HKHT) (from November 2007 to July 2010) and Hong Kong PolyU (HKPU) (from November 2005 to October 2011).The AERONET spectral AOT data were input to SEFT to compute ASD.The retrieved ASD values were then compared with the AERONET inversion products for validation purpose.Fig. 4 compares the ASD derived from SEFT and that from the AERONET inversion products.It is observed that differences in fine mode median radii of ASD are 0.018 µm and 0.026 µm, and differences in coarse mode median radii are 0.186 µm and 0.231 µm in HKHT and HKPU respectively.ASD results appear to be similar between the SEFT and AERONET inversion products although discrepancies are shown in the volume concentrations in the peaks of coarse mode radius (Fig. 4(a)) and fine mode radius (Fig. 4(b)).The averaged difference of all particle sizes between the SEFT and AERONET inversion products is calculated using Eq. ( 6).
where V SEFT,r , V AERONET,r are the retrieved volume concentration at a given radius r. n is the total number used in ASD.The discrepancies in Fig. 4 show ≤ 10.4%.Thus, these results suggest that the SEFT using the LUT-based inversion scheme performs well, and around the predicted error of 10%.
The SEFT method has also been applied to Microtops II sunphotometer measurements at the Hong Kong International Airport (HKIA) (data available from January 2007 to April 2011).Similar to HKHT AERONET site, HKIA is located near the coast where it is affected by marine aerosols.In contrast, the HKPU AERONET station is deployed in the urban core areas which suffers from anthropogenic aerosols.In order to evaluate the robustness of the SEFT algorithm for Microtops II data, the SEFT derived ASD values were compared with the AERONET inversion products from HKHT and HKPU, collected over the same measurement periods.The data points were chosen within ± 30 minutes from the Microtops II in HKIA and AERONETs at HKHT and HKPU.The ASD in bimodal lognormal distribution is shown in Fig. 5. Approximately 4.6% (HKHT) and 27.8% (HKPU) of averaged differences in the volume concentrations of all particle sizes are identified by the SEFT algorithm when compared with AERONET data (calculated by Eq. ( 6)).Clearly, comparison of the ASD from HKIA Microtops II with HKHT AERONET products appears logical and reasonable since both are in coastal areas which are greatly affected by marine aerosols.
In general, both the magnitude and shape of bimodal lognormal distribution in ASD from the SEFT algorithm show promising agreement with those from the AERONETs (Fig. 5).However, a relatively large discrepancy in size distribution may appear in the coarse mode, which is larger than the discrepancies observed in Fig. 4.This may be caused by the errors from (i) different locations (AOT discrepancies caused by locational difference between three locations: HKIA and two AERONET sites), (ii) systematic errors in AOT measurements from two types of instruments (Microtops II and AERONET sunphotometers), and (iii) larger sized marine aerosols observed at HKIA than those observed at HKHT and HKPU AERONET sites.However, since both SEFT and AERONET inversion products are based on an aerosol optical inversion technique, no absolute accuracy can be concluded although promising agreement is shown between two independent techniques.

DISCUSSION AND CONCLUSION
This study investigated the size distribution retrieval method from observed spectral AOT measurements from Microtops II sunphotometer.With only AOT data observed from Microtops II instrument, the derivation of ASD has long been a controversial issue, and is difficult to establish.However, the integration of LUTs, and the scattering scheme, as well as appropriate estimations of refractive indices can resolve this challenge.
A new method, named SEFT algorithm, has been developed to retrieve the ASD information from Microtops II.A number of datasets for LUTs were generated in this study.The measured AOTs from the sunphotometer were then compared with the extinction coefficients in the LUTs to select the appropriate ASD.For this comparison, a RMSD test was used which can select the appropriate ASD with the smallest systematic difference.The proposed method in this study provides promising results and acceptable accuracies.The averaged differences are 10.4% and 9.8% when compared with AERONET inversion products at rural and urban sites, respectively.The ASD from Microtops II at HKIA was compared with AERONET inversion products and reasonable accuracy was found.These comparisons appear logical and reasonable since all sites are located within 50 km but with different degrees of urbanization.In addition, two case studies, for clear and hazy days were tested, and the results show that the SEFT for retrieving aerosol size distribution can operate well under those atmospheric conditions.
Although the SEFT shows good agreement with AERONET inversion products, possible error sources include the following: 1) Systematic differences from two different instruments (Microtops II and Cimel sunphotometer used in AERONET).The Cimel sunphotometers are annually calibrated by NASA from instruments at Mauna Loa, Hawaii, under substantially different climatic conditions and the uncertainty of AOT is 0.01-0.02AOT (wavelength dependent) due to calibration uncertainty of the field instruments.The Microtops II sunphotometer was also calibrated and the uncertainty is 3-5%.Although only small systematic errors are observed from the two instruments, these small errors may accumulate and be magnified during the inversion procedure.
2) Error caused from the manual measurements of Microtops II.At the time of measurement using Microtops II, the user has to hold the sunphotometer firmly, pointing it towards the sun by keeping the image of the sun centered in the bull's-eye of the sun target.A sharp sun target during the observations is needed to maintain data quality, and a small offset during observation can cause extreme AOT deviations in AOT (Massen, 2005).
3) Inadequate representation or step size of ASD parameters for LUT calculations.Although this study generated and adopted a large number of LUTs for the retrieval scheme, the insufficient intervals of the hypothetical refractive indices and other ASD parameters may be other possible error source.
All these diverse sources of potential errors need to be taken into account in quality control procedures.The accuracy reported in this study is relative and no absolute accuracy can be concluded.More evaluation using particle profiler and optical particle counter is suggested for further study.

Fig. 4 .
Fig. 4. Average aerosol size distribution retrieved from SEFT and AERONET inversion products at (a) Hong Kong Hok Tsui and (b) Hong Kong PolyU.

Fig. 5 .
Fig. 5. Average aerosol size distribution retrieved from SEFT at Hong Kong International Airport and AERONET inversion products at Hong Kong Hok Tsui and Hong Kong PolyU.