The effect of dust aerosol on satellite data from different color scanners

Толық мәтін

Introduction

Mineral dust is often neglected in the analysis of anthropogenic climate change, as it is considered part of the natural aerosol. Some researchers believe that dust may be an important climate-forming component, especially over certain oceanic areas and regions where its concentrations are high [1, 2]. Although it is impossible to determine the exact impact of mineral dust on the global climate, research on this topic is interdisciplinary and relevant. Complete information on the properties of different types of aerosols (including absorbing aerosols) can be obtained by comprehensive determination of their concentration, microstructure, chemical composition and optical properties [3−5].

This study is a continuation of a series of works dedicated to the examination of optical properties of the dust aerosol over the Black Sea and its influence on Ocean Color products. For the region under study, the results obtained when analyzing satellite data can in many cases have large errors due to incorrect consideration of the optical properties of the aerosol [6−9]. It is worth noting that dust transports from both the African continent as well as from the Middle East and Asia are observed annually over the Black Sea region [10]. Since the MHI RAS scientists have been studying this topic for more than 10 years, there is already a certain method to identify different types of aerosols (background aerosol, smoke and dust) based on the variability of optical properties, such as aerosol optical depth (AOD), Angstrom parameter (α), single scattering albedo (SSA), size distribution and concentration of aerosol particles (fine (PM2.5) and coarse (PM10) particles), asymmetry parameter, etc. The dust aerosol identification method combines a visual analysis of satellite images, which clearly show a dust plume, and an analysis of photometric measurements of the optical properties of the aerosol. To analyze the aerosol over the Black Sea region, data from the AERONET network stations (Galata_Platform, Section_7) located in the western part of the Black Sea are used, as well as unique data from the portable SPM spectrophotometer and the ATMAS dust meter, which were measured daily on the MHI RAS territory [11−13].

It is worth noting that dust aerosol has the greatest impact not only on the variability of the optical properties of the atmosphere, but also on the Ocean Color satellite products. For an objective assessment of the state of the water surface and the procedure for atmospheric correction based on remote sensing data, it is necessary to carry out a comparative analysis of three types of measurement data: satellite, model and in situ. In [8, 14−28], it is shown that in the presence of dust, the sea remote sensing reflectance can have negative values in the shortwave range (400−443 nm). This fact indicates systematic errors in the operation of standard atmospheric correction algorithms, which are based on the principle of extrapolating aerosol properties from the near-IR part of the spectrum to the visible part [27]. In a previous work [18], it was shown analytically that in the presence of dust-absorbing aerosol in the atmosphere above the region, the error in the atmospheric correction is expressed by a fourth-degree power function, i.e. it is close to λ^-4. This is due to the absorption by the aerosol of radiation scattered by the air molecules. The analytical expression describing the dependence of the error value of the standard atmospheric correction on the aerosol stratification, for small values of the optical depth of light absorption by the aerosol a0 (λ), has the following form

$r=\frac{p_m\left(\cos \gamma \right)\cdot {\tau }_m^0\left(\lambda \right)}{4{\mu }_0\mu }{a}_0\left(\lambda \right)\cdot \left(\frac{1}{{\mu }_0}+\frac{1}{\mu }\right)\underset{0}{\overset{1}{\int }}\underset{0}{\overset{p}{\int }}g\left(x\right)dx\cdot dp$,                       (1)

$a_0\left(\lambda \right)=(1-\Lambda ){\tau }_a^0$,

where ${\text{τ}}_a^0$ is the aerosol optical depth; $\Lambda$ is the single scattering albedo; ${\text{μ}}_0$ is the cosine of the solar zenith angle; $\text{μ}$ is the cosine of the observation zenith angle; $\cos \gamma =-{\mu }_1\cdot {\mu }_2+\sqrt{1-{\mu }_1^2}\sqrt{1-{\mu }_2^2}\cos \varphi$ is the cosine of the scattering angle; ${\text{τ}}_m^0$ is the total optical depth of the molecular atmosphere; g(x) is the dust aerosol stratification function, which shows the dependence of the relative concentration of aerosol particles on atmospheric pressure at a given altitude. The first fraction in expression (1) is nothing more than the expression for the sensing reflectance of the molecular atmosphere in the linear Gordon approximation. Consequently, three factors can be identified that influence the magnitude of the atmospheric correction error. The multiplier $\frac{p_m\left(\cos \gamma \right)}{{\mu }_0\mu }\left(\frac{1}{{\mu }_0}+\frac{1}{\mu }\right)$ describes the observation geometry, and the double integral is independent of the wavelength and takes into account stratification of the absorbing aerosol relative to the air molecules. Consequently, the spectral properties of the atmospheric correction error are described by the factors ${\tau }_m^0\left(\lambda \right)$ and a0. It is known that, according to the Rayleigh law, ${\tau }_m^0\approx {\lambda }^{-4}$, spectral properties of aerosol absorption are determined by aerosol microphysics, which for dust aerosol depends on the dust sources and their transformation processes in the atmosphere. Until now, the spectral dependence variability a0 has not been considered. In this paper, it is proposed to estimate for the first time the spectral behavior of the absorption properties of dust aerosol for the case of dust transfer over the Black Sea region.

Experimental regularities of the atmospheric correction error were analyzed in [18]. It was shown that the largest difference between satellite and in situ remotely sensed ocean reflectance data is recorded in the presence of dust aerosol in the atmosphere. For the selected 49 data obtained on dust transfer days, the principal component method with the first vector contribution estimation was used. The result showed that 86% of the variance of the MODIS-Aqua validation error is explained by the first eigenvector, which is well approximated by a power law ${\lambda }^{-3.57\pm 0.32}$. This confirmed the reliability of the analytical estimates [18].

The objective of this study is to estimate the magnitude of the atmospheric correction error of MODIS-Aqua, MODIS-Terra, VIIRS-SNPP, VIIRS-JPSS, NASA HawkEye (SeaHawk) and OLCI-Sentinel-3A satellite data for 28–29 July 2021, when dust transport over the Black Sea region was recorded.

The present study focuses on an analysis of the variability of the optical properties of the atmosphere on 28–29 July 2021 over the Black Sea region and the evidence (based on satellite and model data) that it is dust aerosol that is recorded over the region during the period under consideration.

The second stage is dedicated to the estimation of the impact of the absorbing dust aerosol on the size of the atmospheric correction error in the calculation of the sea remote sensing reflectance for 28 and 29 July 2021. For these dates, the largest number of different satellite measurements, synchronized with in-situ remote sensing sea reflectance measurements according to the AERONET – Ocean Color (AERONET-OC) network, were obtained. In this study, the validation error was calculated for MODIS-Aqua/Terra, VIIRS-JPSS, Sentinel-3A and HawkEye (SeaHawk).

Instruments and materials

The photometric data from the international AERONET (Aerosol ROboties NETwork) were used as a source for the in-situ AOD measurements. The data from the AERONET-OC extension, which allow the measurement of radiation from underwater, were used to analyze the sea remote sensing reflectance data [29]. Currently, two stations, Black Sea Section_7 (29.45°E, 44.45°N) and Galata_Platform (28.19°E, 43.05°N), provide information on seawater color. In this paper, an array of daily mean data on normalized LWN level 2 (higher quality) water radiation has been analyzed. The level 1.5 data array is selected taking into account cloudiness through a series of quality tests, and the level 2 data array consists of fully cleaned data obtained after calibration and software verification. During the studies, LWN(λ) values were converted to Rrs(λ) values by dividing by the solar constant Fo(λ) [30].

To compare satellite and in situ measurements for 28–29 July 2021 and to correct for inaccuracies caused by variability and anomalies in atmospheric parameters, the data from the international AERONET photometer network, freely available at http://aeronet.gsfc.nasa.gov and the MODIS-Aqua/Terra, VIIRS-SNPP/JPSS, HawkEye and OCLI data, freely available at https://Ocean Color.gsfc.nasa.gov, were selected. The MODIS optical properties data are the result of a combination of Terra and Aqua satellite measurements, providing near real-time information. The resolution of the MODIS sensor is 0.5°, the resolution of the images is 2 km, and the temporal resolution is diurnal.

A complicating factor in the study is that the wavelengths at which the AERONET-OC station measurements are provided do not fully match the channels measured by the satellites, especially in the visible range. Thus, the MODIS-Aqua/Terra measurement channels have wavelengths of 412, 443, 469, 488, 531, 547, 555, 667 and 678 nm; VIIRS-JPSS − wavelengths of 411, 445, 489, 556 and 667 nm. For HawkEye, Rrs(λ) are measured at wavelengths of 412, 488, 510, 556 and 670 nm. The problem of interpolating the sea remote sensing reflectance values obtained in the CIMEL-318 photometer channels to the satellite channels is due to the complex shape of the seawater absorption spectrum. Scattering also affects the shape of Rrs(λ). However, the corresponding spectral dependencies are monotonous and smoother, allowing the use of a second-degree polynomial in the interpolation. In the absorption spectrum, special attention is paid to the absorption of pure seawater, since it greatly affects the variability of the Rrs(λ) values in the long wavelength region of the spectrum. The absorption spectrum of pure seawater introduces the largest interpolation errors.

In this study, a method was used which consists of multiplying the sea remote sensing reflectance obtained from field measurements by the seawater model absorption value:

$a_w\left({\lambda }_i\right)=a_{\text{pw}}\left({\lambda }_i\right)+0.1\cdot C_y\cdot \exp \left[0.015\cdot \left(400-{\lambda }_i\right)\right],$             (2)

where $C_y$ is estimated by statistical relationship with color index:

 $C_y=2.3\cdot CI{(555/510)}^{2.18}.$               (3)

Expression (3) was obtained based on the regression dependencies given in [31]. After multiplying the natural sea remote sensing reflectance by the model absorption, a second-degree polynomial interpolation was applied to the satellite channels. The resulting values were then divided by the model absorption value at the satellite wavelengths.

The data from the CALIPSO satellite were analyzed to determine the predominant aerosol type during the study period. This is an American-French research satellite launched as part of NASA EOS (Earth Observing System) program to study the Earth’s cloud cover and the vertical structure of atmospheric aerosol. Its main instrument is a three-channel imaging radiometer (8.65, 10.6 and 12.05 mkm). Aerosol types are determined by the value of the integrated backscattering coefficient and the particle depolarization coefficient. The aerosol types determined by the calculations of the CALIPSO algorithms are: smoke (from forest fires), dust, contaminated dust (mixtures of dust and smoke), contaminated continental and clean continental aerosol [32]. Each aerosol type is characterized by a set of lidar ratios at the 532 and 1064 nm wavelengths [33].

To obtain information on the source of the smoke aerosols, the results of the calculation of reverse trajectories of air mass transfer, obtained using the HYSPLIT modeling software package, were used. Reverse trajectory analysis makes it possible to follow the movement of air flows at different altitudes and to identify the location of likely sources of pollution entering the atmosphere [34].

Results and discussion

On 28–29 July 2021, satellite data showed intense dust transport from the Arabian Peninsula and the Sahara across the Black Sea region. According to the VIIRS false-color satellite images, dust transport is recorded on both sides of the illuminated area, which means that the size of the dust event is more than a thousand square kilometers. All presented satellite images also clearly show the area of intense fires on the Mediterranean coast (territory of Turkey). Intense absorption due to the presence of smoke aerosol west of the island of Crete is confirmed by high AOD values over the eastern part of the Mediterranean Sea on 29 July 2021 (Fig. 1). The next step is confirming or refuting of the dust transfer event over the Black Sea region during the study period was the analysis of the reverse trajectories of the air flow movement using the HYSPLIT model [34] (Fig. 1, b, d). As shown in Fig. 1, dust transfer from the Sahara was recorded at the 3 km level for all days.

Table 1. Optical characteristics of atmospheric aerosol over the AERONET-OC stations in the Black Sea

In this paper, a comparative analysis of the aerosol optical properties at AERONET stations (Galata_Platform and Section_7_Platform) was performed for cases of different aerosol activity, namely: for 28 July 2021 (the day of the intense dust transfer), 27 July 2021 (the day before the dust transfer in the presence of background aerosol) and 29 July 2021 (the day after the start of the intense dust transfer). It is worth noting that in July 2021 cloudiness was often observed over the AERONET stations and therefore the initial monthly mean AOD values were overestimated, which is typical for summer months [35] (Table 1).

Fig. 1. Satellite images of VIIRS-SNPP/JPSS from 28 July 2021 (a) and 29 July 2021 (c) (source: https://oceancolor.gsfc.nasa.gov), and corresponding HYSPLIT back trajectories (b, d) (source: https://www.ready.noaa.gov/HYSPLIT_traj.php)

At both western AERONET stations in the Black Sea region, a large amount (compared to background values) of coarse aerosol fraction (more than 2.5 μm) and low (SSA < 1) values of the single scattering albedo (SSA) were observed on 28 July 2021 (Fig. 2). In general, a similar situation was observed on 29 July 2021, except for higher values of the Angstrom parameter.

Fig. 2. At the AERONET network stations on 28 July 2021: contribution of fine (less than 2.5 mkm) and coarse particles (2.5 mkm and more) to the overall distribution of AOD at Galata_Platform (a) and Section_7 (с), single scattering albedo at Galata_Platform (b) and Section_7 (d), particle size distribution at two stations (e)

For the first time, an estimate of the optical depth of aerosol absorption is given for the general assessment of the absorption properties of dust aerosol:

$a_0\left(\text{λ}\right)=(1-\Lambda (\text{λ}\left)\right){\text{τ}}_a^0\left(\text{λ}\right).$                    (4)

Consequently, the average daily variation of the power function of the optical absorption depth was analyzed for synchronous pairs of AOD and SSA measurements for AERONET stations (Fig. 3).

Fig. 3. Mode of the power function of aerosol absorption optical depth at Galata_Platform (а) and Section_7 (b) for 28 July 2021, and at Galata_Platform (c) and Section_7 (d) for 29 July 2021

From Fig. 3 it is evident that the power function is close to λ⁻¹. Consequently, the magnitude of the atmospheric correction error depends not only on the λ⁻⁴ factor, but also on the light absorption by the aerosol. As a result, the error of the standard atmospheric correction will increase at shorter wavelengths. It should be noted that the standard atmospheric correction procedure is not capable of qualitatively estimating the change in spectral properties of aerosol scattering under the influence of light absorption in the near IR region, due to the smallness of this effect in the long wavelength part of the spectrum. For this reason, it is necessary to use additional information about the optical properties of the underlying surface in the shortwave region. Thus, the application of standard algorithms for atmospheric correction of satellite data in the presence of absorbing dust aerosol requires an additional regional correction. The product a0(λ)∙λ⁻⁴ should be used as the interpolation function, and the proportionality coefficient is found from the conditions imposed on the sea remote sensing reflectance in the shortwave region of the spectrum.

The next stage of the study is to calculate the atmospheric correction error for the MODIS-Aqua/Terra, VIIRS-SNPP/JPSS, HawkEye and Sentinel-3A satellites for the dates considered. The validation procedure for the satellite data was similar to that used for the SeaBASS database: synchronous pairs of measurements with the smallest time difference within a radius of 5 km around the western Black Sea AERONET-OC stations were selected. Using the SeaDAS software package, all pixels with the following error flags were similarly excluded: LAND, STRAYLIGHT, HIGLINT, HILT, MODGLINTATMWAR and NAVFAILE [36]. Unfortunately, the VIIRSS-SNPP satellite data for 28 July 2021 were excluded from further analysis because all pixels within a 5 km radius of the AERONET-OC stations were within the satellite illumination zone. The synchronous in situ measurements of Rrs(λ) at the AERONET-OC stations in the western Black Sea during the day changed little, namely: the standard deviation (SD) was less than 10% of the value, allowing the use of daily averages. The interpolation results constructed from the results of sea remote sensing reflectance measurements by MODIS-Aqua, VIIRS-JPSS, Hawkeye and OCLI Sentinel-3A satellites for 28 July 2021 are shown in Fig. 4.

Fig. 4. Errors of atmospheric correction and their approximation by power dependence for 28 July 2021

The atmospheric correction error was similarly calculated for 29 July 2021, when the dust aerosol AOD was higher, but the Ångström parameter was lower (Fig. 5). Unfortunately, the Modis-Terra and VIIRS-SNPP data had strong outliers and after filtering the error flags, there were no data left.

Fig. 5. Errors of atmospheric correction based on the results of measurements of sea remote sensing reflectance by satellites MODIS-Aqua, VIIRS-JPSS, Hawkeye and OCLI Sentinel 3A for the Black Sea AERONET-OC stations for 29 July 2021

As a result of approximation of the atmospheric correction errors for 28 July 2021, power-law dependencies close to λ⁻⁵ were obtained. This is explained by the total contribution of: 1) the molecular component (λ⁻⁴) and 2) the aerosol absorption (λ⁻¹). Fig. 4 shows that the atmospheric correction error of the sea remote sensing reflectance for the Galata_Platform station, obtained from VIIRS-JPSS and HawkEye measurements, is close to the power-law dependence λ⁻⁴ − λ⁻⁵, and for the Section_7 station it has a more pronounced power-law dependence, namely λ⁻⁷. A pronounced shape of the power-law function is observed on 29 July 2021, when the dust aerosol concentration increases and its aerosol absorption is already close to λ⁻². It is worth noting that the largest atmospheric correction errors were found for 29 July 2021 according to MODIS-Aqua data, whose interpolation function is close to the form λ⁻⁸. We believe that this is due to underestimated Rrs measurements in the longwave region for this day and consequently large errors in the standard power law approximation – the logarithmic method followed by linear optimization. Using a nonlinear approximation, the power function was obtained in the form λ⁻⁴ − λ⁻⁵, which also indicates large errors in the shortwave region of the spectrum. The maximum errors in the blue region are observed in Fig. 4, a, c; 5, a, c.

It is worth noting that the new HawkEye and Sentinel-3A satellites, despite their short lifetime and a low number of reprocessings and calibrations, show more accurate results. This may be due to the better spatial resolution of the new satellite instruments.

Analysis of CALIPSO satellite data on the stratification of different aerosol types for 28 and 29 July 2021 confirmed the presence of dust particles in the surface atmospheric column up to 5 km over the Black Sea. In addition to dust aerosol, contaminated dust and smoke aerosol were recorded on 29 July 2021, which also confirms the spatial distribution of smoke towards The Black Sea region shown in Fig. 1, c.

Conclusion

As a result of the approximation of the atmospheric correction errors of the satellite data for 28 July 2021, power-law dependencies close to λ⁻⁵ were obtained. This is explained by the total contribution of the molecular component (λ⁻⁴) and the aerosol absorption (λ⁻¹). For 29 July 2021, a pronounced power-law behavior is observed as the dust aerosol concentration increases and the contribution of aerosol absorption becomes close to the λ⁻² power-law dependence. In addition, for 29 July 2021, the presence of both dust and smoke aerosols was shown over the study region according to CALIPSO satellite data. According to the HYSPLIT air flow backtracking modeling data, the aerosol masses on this day moved from the southwest (Crete) towards the Black Sea, which is further confirmed by the high AOD values over the eastern Mediterranean on 29 July 2021. It is assumed that the combination of two absorbing aerosol types caused even greater inaccuracies in the determination of the spectral sea remote sensing reflectance for the period under study.

Авторлар туралы

Anna Papkova

Email: hanna.papkova@gmail.com
Scopus Author ID: 57203015832

Junior Researcher, CSc. (Phys.-Math.)

Evgeniy Shybanov

Хат алмасуға жауапты Автор.
Email: e-shybanov@mail.ru
Scopus Author ID: 6507075380
ResearcherId: ABB-9097-2021

Leading Researcher, DSc. (Phys.-Math.)

Darya Kalinskaya

Email: kalinskaya@mhi-ras.ru
Scopus Author ID: 56380591500

Junior Researcher

Әдебиет тізімі

Қосымша файлдар

Қосымша файлдар

Әрекет

1. JATS XML

Жүктеу

2. Fig. 1. Satellite images of VIIRS-SNPP/JPSS from 28 July 2021 (a) and 29 July 2021 (c) (source: https://oceancolor.gsfc.nasa.gov), and corresponding HYSPLIT back trajectories (b, d) (source: https://www.ready.noaa.gov/HYSPLIT_traj.php)

Жүктеу (387KB)

Метадеректер

3. Fig. 2. At the AERONET network stations on 28 July 2021: contribution of fine (less than 2.5 mkm) and coarse particles (2.5 mkm and more) to the overall distribution of AOD at Galata_Platform (a) and Section_7 (с), single scattering albedo at Galata_Platform (b) and Section_7 (d), particle size distribution at two stations (e)

Жүктеу (419KB)

Метадеректер

4. Fig. 3. Mode of the power function of aerosol absorption optical depth at Galata_Platform (а) and Section_7 (b) for 28 July 2021, and at Galata_Platform (c) and Section_7 (d) for 29 July 2021

Жүктеу (144KB)

Метадеректер

5. Fig. 4. Errors of atmospheric correction and their approximation by power dependence for 28 July 2021

Жүктеу (248KB)

Метадеректер

6. Fig. 5. Errors of atmospheric correction based on the results of measurements of sea remote sensing reflectance by satellites MODIS-Aqua, VIIRS-JPSS, Hawkeye and OCLI Sentinel 3A for the Black Sea AERONET-OC stations for 29 July 2021

Жүктеу (276KB)

Метадеректер