Method for Inverting Aerosol Optical Depth Based on Multi-Band Satellite Data

Abstract

The present invention discloses a method for retrieving aerosol optical depth based on multi-band satellite data. The method comprehensively considers the influence of surface anisotropy and the variation of aerosol types, constructs a database of bidirectional reflectance distribution function (BRDF) of the surface, a candidate aerosol model for each region, and a cost function based on the retrieval results of AOD (550nm) for each band, simplifies the iterative calculation of AOD retrieval, realizes the simultaneous determination of AOD and aerosol types, and improves the retrieval accuracy of AOD.

Description

Technical Field

[0001] The present invention belongs to the technical field of aerosol satellite remote sensing, and particularly relates to a method for retrieving aerosol optical depth based on multi-band satellite data. Background Technique

[0002] Aerosols affect the global radiation balance and climate change through direct and indirect radiative forcing. At the same time, aerosols can increase the incidence and mortality of respiratory and cardiovascular diseases, cause haze weather, and reduce environmental quality. To better understand the impact of aerosols on climate change, air pollution, and human health, it is necessary to accurately obtain the spatio-temporal distribution of aerosols.

[0003] Aerosol Optical Depth (AOD) is an important optical property of aerosols. It is defined as the integral of the extinction coefficient of aerosols in the vertical direction of the atmosphere and can be used to represent the concentration of aerosols. Currently, AOD is the most important parameter for aerosol retrieval. Internationally, a variety of satellite AOD products have been released and widely applied. The Medium Resolution Spectral Imager-II (MERSI-II) sensor carried by the domestic FY-3D satellite can be used to retrieve AOD, but there is still a lack of a stable algorithm. For satellite AOD retrieval over land, a variety of retrieval algorithms have been developed and evolved. Among them, the most representative ones are the Dark Target (DT) method and the Deep Blue (DB) algorithm. The DT algorithm utilizes the stable linear relationship of the surface reflectance of dark targets such as dense vegetation in the blue, red, and short-wave infrared bands. By introducing this linear relationship, the unknowns during aerosol retrieval are reduced to achieve the retrieval of AOD. However, this method is limited to dark target areas and fails in bright surface areas such as deserts and cities, and there is an obvious overestimation in this algorithm. The DB algorithm mainly uses long-term apparent reflectance data and constructs a priori surface reflectance databases using the minimum reflectance method to support the retrieval of AOD. This algorithm has achieved the retrieval of AOD in bright surface areas such as deserts and has been widely applied. However, the construction of the surface reflectance database relies on a large amount of satellite data, and the application of the minimum reflectance method is also limited for satellite payloads lacking the deep blue band. In addition, a variety of AOD retrieval algorithms have emerged in recent years. The basic idea of these methods is to construct a reflectance database or a database of reflectance relationships between bands, which has promoted the development of satellite AOD retrieval to a certain extent. More and more studies have also begun to consider the influence of the Bidirectional Reflectance Distribution Function (BRDF) of the surface, improving the retrieval accuracy of AOD. The determination of aerosol type is a key factor for accurate satellite AOD retrieval. In terms of the determination of aerosol type, multi-angle - polarization satellite observations can achieve multi-parameter retrieval of aerosols using ground-based aerosol retrieval methods, while single-angle satellite observations usually lack sufficient information and need to make assumptions about aerosol types first. In terms of the assumption of aerosol types, there are mainly two methods. One is to obtain multiple candidate aerosol types through cluster analysis, construct a cost function based on apparent reflectance, and select the optimal type by minimizing the cost function. The other is to make regional and seasonal assumptions about aerosol types based on ground-based observation data and select aerosol types according to time and region. The method of minimizing the cost function can capture the changes in aerosol types but requires complex calculations. Usually, interpolation calculations of the lookup table are performed multiple times to achieve iterative solutions. The method of determining aerosol types according to time and region is difficult to apply to sudden situations of aerosols, such as sand and dust, and haze weather. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides a method for retrieving aerosol optical depth based on multi-band satellite data. This method comprehensively considers the influence of surface anisotropy and the variation of aerosol types, simplifies the iterative calculation of AOD retrieval by constructing a new cost function, realizes the simultaneous determination of AOD and aerosol types, and improves the retrieval accuracy of AOD.

[0005] To achieve the above object, the technical solution provided by the present invention is a method for retrieving aerosol optical depth based on multi-band satellite data, including the following steps:

[0006] Step 1, obtain satellite data and ground-based observation data;

[0007] Step 2, preprocess the satellite data obtained in Step 1;

[0008] Step 3, construct a BRDF prior database;

[0009] Step 4, construct candidate aerosol types for each study area;

[0010] Step 5, construct an AOD retrieval look-up table;

[0011] Step 6, construct a cost function, and take the AOD(550nm) of the blue light band corresponding to the minimum value of the cost function as the preliminary retrieval result;

[0012] Step 7, eliminate high-value AOD pixels within the radius L km of the cloud pixel to obtain the final retrieval result.

[0013] Moreover, the satellite data obtained in Step 1 includes the L1-level radiometric correction data of FY-3D MERSI-II, MODIS MCD19A3, and MODIS MYD04L2 data. The MERSI-II data contains a total of 25 bands in the range of 0.412 - 12.0 μm, where the spatial resolution of Band 1 - Band 4 and Band 24 - Band 25 is 250 m, and the spatial resolution of other bands is 1 km. The MCD19A3 data contains three kernel parameters of the RTLS model of MODIS Band 1 - Band 8, with a spatial resolution of 1 km, a temporal resolution of 8 days, and the operational algorithm uses the multi-angle atmospheric correction algorithm (MAIAC). The MYD04L2 data is the MODIS standard aerosol product, with a spatial resolution of 10 km, and includes AOD retrieval products of the DT and DB algorithms. The ground-based data comes from the products of AERONET, and the AERONET product parameters used are AOD, extinction Angstrom exponent, absorption Angstrom exponent, SSA, and asymmetry factor.

[0014] Moreover, the Step 2 includes the following steps:

[0015] Step 2.1: After performing radiometric correction and geometric correction on the MERSI-II L1 data, convert it to the WGS84 coordinate system.

[0016] Step 2.2: Perform pixel selection on the MERSI-II L1 data processed in Step 2.1, eliminate the pixels unsuitable for aerosol inversion, and establish a mask file for identification.

[0017] Step 2.2.1: Comprehensively utilize the three characteristics of clouds, namely high brightness in the blue band, lower spatial smoothness at the cloud edge compared to the surface, and lower brightness temperature of clouds, to identify and eliminate cloud pixels. If any of the following conditions is met, it is identified as a cloud pixel: (1) The apparent reflectance of the blue band (Band 1) of the FY-3D medium-resolution spectral imager is greater than 0.4; (2) The standard deviation of the 3×3 pixels in the blue band is greater than 0.0075, and the average standard deviation is greater than 0.0025; (3) The apparent reflectance of the blue band is greater than 0.35, and the standard deviation of the 3×3 pixels in the blue band is greater than 0.01; (4) The brightness temperature BT11 is less than 270K.

[0018] Step 2.2.2: Use the Normalized Difference Snow Index (NDSI) to identify and eliminate ice, snow, and water pixels. The identification condition is that NDSI is greater than 0.35.

[0019] Step 2.2.3: Establish a pixel identification mask and use different numbers to mark different ground objects.

[0020] Moreover, Step 3 includes the following steps:

[0021] Step 3.1: Extract three kernel parameters of the MODIS MCD19A3 data, namely the independent scattering kernel, the volume scattering kernel, and the surface geometric kernel, and perform reprojection and image mosaicking on the three parameters.

[0022] Step 3.2: Construct a monthly BRDF database based on the principle of the minimum relative standard deviation.

[0023] Principle of the minimum relative standard deviation: For a specific month, there are multiple images at different times for each parameter. Arrange the pixel values corresponding to the same position on different images in groups of 4 for permutation and combination to obtain m combinations. Calculate the relative standard deviation of each group, and take the mean of the 4 valid pixels corresponding to the group with the minimum relative standard deviation as the final value of the pixel.

[0024] Moreover, Step 4 includes the following steps:

[0025] Step 4.1: Classify the aerosol property data of AERONET sites in the Chinese region obtained in Step 1 according to three natural regions: the East Asian Monsoon Region (EM), the Arid Region of Northwest China (ANC), and the Qinghai-Tibet Plateau Region (QTP).

[0026] Step 4.2: Conduct cluster analysis of aerosol types within each region using the K-means method.

[0027] The parameters used in the cluster analysis include the extinction Angstrom exponent, the absorption Angstrom exponent, the single scattering albedo (SSA) at 440 nm, the difference in SSA between 870 nm and 440 nm, and the asymmetry factor at 440 nm.

[0028] Step 4.3: Determine the final N candidate aerosol types within each region according to the clustering results of Step 4.2.

[0029] Moreover, in Step 5, an AOD retrieval lookup table for each band and each type of aerosol is constructed using the 6SV model. The parameters included in the lookup table are the solar zenith angle, the observation zenith angle, the relative azimuth angle, the atmospheric model, the AOD, the surface elevation, the BRDF-independent scattering kernel, the BRDF-volume scattering kernel, the BRDF-surface geometry kernel, and the simulated apparent reflectance.

[0030] Moreover, in Step 6, it is assumed that the aerosol types within an n×n km area centered on the target pixel are the same. Based on the equality of the AOD (550 nm) retrieved for each band under the condition of accurate aerosol types, a new cost function, the cost function J M is calculated as follows:

[0031]

[0032] In the formula, N, k, and M represent the number of pixels, the number of bands, and the aerosol types respectively; AOD 550(k) (M) represents the retrieval result of the AOD (550 nm) of the k-th band of the M-type aerosol, which is obtained by single-band retrieval through the retrieval lookup table constructed in Step 5 based on the data of Steps 2 and 3. is the average value of the retrieval results of the AOD (550 nm) of the k-th band of the M-type aerosol.

[0033] Moreover, in Step 7, quality control is performed on the retrieval results obtained in Step 6 according to the pixel identification mask established during the pixel screening process in Step 2.2.3, that is, the high-value AOD pixels within the range of the cloud pixel radius of L km are removed to obtain the final result.

[0034] Compared with the prior art, the present invention has the following advantages:

[0035] It makes full use of the anisotropic information of the surface reflection, takes into account the characteristics and variations of aerosol types, simplifies the complexity of multi-band AOD inversion, and can achieve accurate AOD inversion and accurate determination of aerosol types. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flowchart of an embodiment of the present invention.

[0037] Figure 2 They are the aerosol model parameters of an embodiment of the present invention, where Figure 2 (a)(e)(i) are the aerosol normalized extinction coefficients, Figure 2 (b)(f)(j) are the scattering phase functions, Figure 2 (c)(g)(k) are the single-scattering albedos, Figure 2 (d)(h)(l) are the particle size distributions.

[0038] Figure 3 They are the verification scatter plots of the satellite AOD inversion results of the present invention and the data of ground-based observation stations, as well as the verification scatter plots of the MODIS AOD products and the data of ground-based observation stations. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Aiming at the key problem that the existing AOD inversion algorithms cannot reduce the computational complexity while considering the surface anisotropy and aerosol type variations, and at the same time cannot ensure stable AOD inversion results, the present invention uses the AOD(550nm) results independently inverted under various aerosol type assumptions from multi-band satellite data to construct a new cost function to achieve the inversion of AOD and aerosol types. At the same time, the observational data of the 5×5 pixels around the target pixel are used to reduce the accidental error of the inversion. The prior BRDF database is used in the present invention to reduce the influence of surface anisotropy, and the Aerosol Robotic Network (AERONET) provides data support for the construction of aerosol types and the verification of inversion results.

[0040] Next, taking the FY-3D MERSI-II multi-band satellite data as an example, the technical solution of the present invention will be further described in conjunction with the accompanying drawings.

[0041] As Figure 1 shown, the process of the embodiment of the present invention includes the following steps:

[0042] Step 1, obtain satellite data and ground-based observation data.

[0043] Satellite data include the L1-level radiometric correction data of FY-3D MERSI-II, MODIS MCD19A3, and MODIS MYD04L2 data. The MERSI-II data contains a total of 25 bands in the range of 0.412 - 12.0 μm. The spatial resolution of Bands 1 - Band 4 and Bands 24 - Band 25 is 250 m, and the spatial resolution of the other bands is 1 km. Bands 1 - Band 3 (blue, green, and red bands) are used for the retrieval of AOD, and Bands 1, 2, 7, and 24 are used for pixel selection. The MCD19A3 data contains three kernel parameters of the RTLS model for MODIS Bands 1 - Band 8, with a spatial resolution of 1 km and a temporal resolution of 8 days. The operational algorithm uses the multi-angle atmospheric correction algorithm (MAIAC). The MCD19A3 data is used to construct a BRDF prior knowledge base for determining reflectance / albedo. The MYD04L2 data is the MODIS standard aerosol product with a spatial resolution of 10 km, containing AOD retrieval products of the DT and DB algorithms, which is the most widely used product at present. This data is used to compare with the retrieval results of the method proposed in the present invention. The ground-based data comes from the products of AERONET. The AOD error of this product is about 0.02 and is often used for the verification of aerosol satellite retrieval algorithms. The AERONET product parameters used are AOD, extinction Angstrom exponent, absorption Angstrom exponent, SSA, and asymmetry factor. These parameters are used on the one hand to construct candidate aerosol optical models and on the other hand to verify the satellite retrieval results.

[0044] Step 2: Preprocess the MERSI-II L1 satellite data obtained in Step 1.

[0045] Step 2.1: After radiometric correction and geometric correction of the obtained MERSI-II L1 data in 2021, convert it to the WGS84 coordinate system.

[0046] Step 2.2: Select pixels from the MERSI-II L1 data processed in Step 2.1, eliminate pixels not suitable for aerosol retrieval, and establish a mask file for identification.

[0047] Step 2.2.1: Identify and eliminate cloud pixels by comprehensively utilizing the characteristics of clouds in the blue band being highly reflective, the spatial smoothness at the cloud edges being lower than that of the land surface, and the cloud's brightness temperature being lower. A pixel is identified as a cloud pixel if it meets one of the following conditions: (1) The apparent reflectance of the blue band (Band 1) of the FY-3D medium-resolution spectral imager is greater than 0.4; (2) The standard deviation of the 3×3 pixels in the blue band is greater than 0.0075, and the average standard deviation is greater than 0.0025; (3) The apparent reflectance of the blue band is greater than 0.35, and the standard deviation of the 3×3 pixels in the blue band is greater than 0.01; (4) The brightness temperature (BT11) is less than 270K.

[0048] Step 2.2.2: Identify and eliminate ice, snow, and water pixels using the Normalized Difference Snow Index (NDSI), with the identification condition being that NDSI is greater than 0.35.

[0049] Step 2.2.3: Establish a pixel identification mask and use different numbers to label different land features.

[0050] Step 3: Construction of the BRDF prior database.

[0051] Step 3.1: Collect the MCD19A3 BRDF products covering the Chinese land area in the two years from 2018 to 2019.

[0052] MCD19A3 is the latest BRDF product of MODIS, obtained by using a time series method for multi-angle atmospheric correction, with a temporal resolution of 8 days and a spatial resolution of 1 km.

[0053] Step 3.2: Extract the three kernel parameters of the MCD19A3 data, namely the independent scattering kernel, the volume scattering kernel, and the surface geometric kernel, and perform reprojection and image mosaicking on the three parameters.

[0054] Step 3.3: Construct a monthly BRDF database based on the principle of the minimum relative standard deviation.

[0055] Principle of the minimum relative standard deviation: For a specific month, there are multiple images at different times for each parameter. The pixel values corresponding to the same position on different images are arranged in groups of 4 for permutation and combination, obtaining m combinations. Calculate the relative standard deviation of each group, and take the mean of the 4 valid pixels corresponding to the group with the minimum relative standard deviation as the final value of the pixel.

[0056] Step 4: Construction of candidate aerosol types in each study area.

[0057] Step 4.1: Obtain the aerosol property data of AERONET sites in the Chinese region from 2010 to 2021, and classify the sites according to the three natural regions of China (Eastern Monsoon Region (EM), Northwest Arid Region (ANC), and Qinghai-Tibet Plateau Region (QTP)).

[0058] Step 4.2, in each region, the K - means method is used for cluster analysis of aerosol types.

[0059] The parameters used in the cluster analysis include the extinction Angstrom exponent, the absorption Angstrom exponent, the single - scattering albedo (SSA) at 440 nm, the difference between the SSA at 870 nm and 440 nm, and the asymmetry factor at 440 nm. These parameters can reflect the micro - physical characteristics of aerosols.

[0060] Step 4.3, according to the clustering results of Step 4.2, 4, 2, and 2 aerosol models are constructed in the East Asian Monsoon Region, the Northwest Arid Region, and the Qinghai - Tibet Plateau Region respectively, and the relevant parameters are as Figure 2 shown.

[0061] Step 5, use the 6SV radiative transfer model to construct an AOD inversion look - up table.

[0062] By constructing an AOD inversion look - up table, the time cost of AOD inversion is reduced. The radiative transfer model used is the 6SV model, which is the radiative transfer code basis for calculating the look - up table in the MODIS atmospheric correction algorithm and can accurately simulate satellite observations. The 6SV model takes into account the ground elevation, atmospheric model, aerosol type, etc., and is one of the most widely used radiative transfer models at present. Use the 6SV model to construct AOD inversion look - up tables for each band and each type respectively. The parameters included in the look - up table are solar zenith angle, observation zenith angle, relative azimuth angle, atmospheric mode, AOD, surface elevation, BRDF - independent scattering kernel, BRDF - volume scattering kernel, BRDF - surface geometric kernel, and simulated apparent reflectance. By modifying the source code, specific aerosol types are defined, and the parts that need to be modified are the extinction coefficient, phase function, and SSA at each wavelength.

[0063] Using the 6SV radiative transfer model, aerosol look - up tables for three bands of MERSI - II Band1 - Band3 are constructed. Each band contains 8 look - up tables for different aerosol models. The settings of the look - up table are shown in Table 1.

[0064] Table 1. Settings of MERIS - II AOD Inversion Look - up Table

[0065]

[0066]

[0067] Step 6, construct a cost function, and take the AOD (550 nm) of the blue - light band corresponding to the minimum value of the cost function as the preliminary inversion result.

[0068] Assume that the aerosol types are the same within an n×n km area centered on the target pixel (n < 50). Based on the equality of the retrieved AOD(550nm) in each band under the condition of accurate aerosol types, a new cost function, the cost function J M is calculated as follows:

[0069]

[0070] where N, k, and M represent the number of pixels, the number of bands, and the aerosol type, respectively; AOD 550(k) (M) represents the retrieved result of AOD(550nm) in the k-th band of the M-th type of aerosol, which is obtained by single-band retrieval through the retrieval look-up table constructed in step 5 based on the data in steps 2 and 3; is the average value of the retrieved results of AOD(550nm) in the k-th band of the M-th type of aerosol.

[0071] According to the retrieval look-up table established in step 5, retrieve the AOD(550nm) values corresponding to different candidate aerosol types based on the data of Band1, Band2, and Band3 respectively. Let the number of bands k be 3 and the number of pixels N be 25 (5×5 pixels), calculate the value of the cost function, and take the AOD(550nm) retrieved from the blue light band corresponding to the minimum value as the retrieval result.

[0072] Step 7, perform quality control on the preliminary retrieval result obtained in step 6 according to the pixel identification mask established in the pixel screening process in step 2.2.3, that is, eliminate the high-value AOD pixels within a radius of 2 km from the cloud pixel to obtain the final result.

[0073] Experimental verification:

[0074] Use the method proposed in the present invention to retrieve the MERSI-II AOD covering the land area of China in 2021, and use the data of 14 ground-based observation stations located in the land area of China to verify the retrieval results. The verification indicators used are the correlation coefficient (R), root mean square error (RMSE), bias (Bias), relative mean bias (RMB), and expected error (EE: ±0.05 ± 20%). The larger the R and the percentage meeting the EE, the higher the accuracy. The lower the RMSE, the closer the RMB is to 1, and the closer the Bias is to 0, the higher the accuracy. The verification results are as Figure 3 shown:

[0075] From Figure 3It can be seen that the method proposed in the present invention has higher accuracy compared with the DT and DB products of MODIS MYD04L2 data, with the highest correlation coefficient (R = 0.91) with ground data, the lowest RMSE (0.13), the RMB closer to 1 (0.93), and about 70.91% of the data falling within the expected error line. In addition, the inversion results of the method proposed in the present invention have more valid data (N = 1124) compared with the DT and DB products of MODIS MYD04L2 data. The DT product shows obvious overestimation, with about 43.63% of the data higher than the expected error line and the RMB being 1.32, an overestimation of about 32%. The DB product also shows obvious overestimation when the aerosol concentration is high.

[0076] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Those skilled in the art to which the present invention pertains can make various modifications or supplements to the described specific embodiments or use similar means for substitution, but will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A method for retrieving aerosol optical depth based on multi - band satellite data, characterized in that, It includes the following steps: Step 1: Obtain satellite data and ground-based observation data; Step 2: Preprocess the satellite data obtained in Step 1; Step 3: Construct a BRDF prior database; Step 4: Construct candidate aerosol types for each study area; Step 5: Construct an AOD inversion lookup table; Step 6: Construct a cost function, and take the AOD 550nm at the blue light band corresponding to the minimum value of the cost function as the preliminary inversion result; Assume that the aerosol types within n×n km centered on the target pixel are the same. Based on the equality of the AOD 550nm retrieved in each band under the condition of accurate aerosol types, a new cost function, cost function J M is calculated as follows: where N, k, and M represent the number of pixels, the number of bands, and the aerosol type, respectively; AOD 550(k) (M) represents the inversion result of AOD at 550 nm for the k-th band of the M-th type of aerosol, which is obtained by single-band inversion through the inversion lookup table constructed in step 5 based on the data in steps 2 and 3; is the average value of the inversion results of AOD at 550 nm for the k-th band of the M-th type of aerosol; Step 7: Eliminate high-value AOD pixels within the range of the cloud pixel radius L km to obtain the final inversion result.

2. The aerosol optical depth inversion method based on multi-band satellite data according to claim 1, characterized in that: The satellite data obtained in Step 1 includes the L1-level radiometric correction data of FY-3D MERSI-II, MODIS MCD19A3, and MODIS MYD04L2 data; the MERSI-II data contains a total of 25 bands in the range of 0.412 - 12.0μm, among which the spatial resolution of Band 1 - Band 4 and Band 24 - Band 25 is 250m, and the spatial resolution of other bands is 1km; the MCD19A3 data contains three kernel parameters of the RTLS model of MODIS Band 1 - Band 8, with a spatial resolution of 1km, a temporal resolution of 8 days, and the operational algorithm uses a multi-angle atmospheric correction algorithm; the MYD04L2 data is the MODIS standard aerosol product, with a spatial resolution of 10km, and includes AOD inversion products of DT and DB algorithms; the ground-based data comes from the products of AERONET, and the AERONET product parameters used are AOD, extinction Angstrom exponent, absorption Angstrom exponent, SSA, and asymmetry factor.

3. The aerosol optical depth inversion method based on multi-band satellite data according to claim 2, wherein: Step 2 includes the following steps: Step 2.1: After performing radiometric correction and geometric correction on the MERSI-II L1 data, convert it to the WGS84 coordinate system; Step 2.2: Select pixels from the MERSI-II L1 data processed in Step 2.1, eliminate pixels not suitable for aerosol inversion, and establish a mask file for identification; Step 2.2.1: Comprehensively utilize the three characteristics of clouds being highly bright in the blue light band, having a lower spatial smoothness at the cloud edge compared to the ground surface, and having a lower brightness temperature of clouds to identify and eliminate cloud pixels. If any of the following conditions is met, it is identified as a cloud pixel: (1) The apparent reflectance of Band 1 in the blue light band of the FY-3D medium-resolution spectral imager is greater than 0.4; (2) The standard deviation of the 3×3 pixels in the blue light band is greater than 0.0075, and the average standard deviation is greater than 0.0025; (3) The apparent reflectance in the blue light band is greater than 0.35, and the standard deviation of the 3×3 pixels in the blue light band is greater than 0.01; (4) The brightness temperature BT11 is less than 270K; Step 2.2.2: Use the normalized difference snow index NDSI to identify and eliminate ice, snow, and water body pixels, and the identification condition is that NDSI is greater than 0.35; Step 2.2.3: Establish a pixel identification mask and use different numbers to mark different ground objects.

4. A method for retrieving aerosol optical depth based on multi-band satellite data according to claim 2, characterized in that: Step 3 includes the following steps: Step 3.1: Extract three kernel parameters of MODIS MCD19A3 data, namely the independent scattering kernel, the volume scattering kernel, and the surface geometric kernel, and re-project and splice the three parameters into an image. Step 3.2: Construct a monthly BRDF database according to the principle of the minimum relative standard deviation. Principle of the minimum relative standard deviation: For a specific month, there are multiple images at different times for each parameter. The pixel values corresponding to the same position on different images are arranged in groups of 4 for permutation and combination to obtain m combinations. Calculate the relative standard deviation of each group, and take the mean of the 4 valid pixels corresponding to the group with the minimum relative standard deviation as the final value of the pixel.

5. The aerosol optical depth inversion method based on multi-band satellite data according to claim 2, wherein: Step 4 includes the following steps: Step 4.1: Classify the aerosol property data of AERONET sites in the Chinese region obtained in Step 1 into sites according to three natural regions: the Eastern Monsoon Region (EM), the Northwest Arid Region (ANC), and the Qinghai-Tibet Plateau Region (QTP). Step 4.2: Use the K-means method to perform cluster analysis of aerosol types within each region. The parameters used in the cluster analysis include the extinction Angstrom exponent, the absorption Angstrom exponent, the single-scattering albedo (SSA) at 440 nm, the difference in SSA between 870 nm and 440 nm, and the asymmetry factor at 440 nm. Step 4.3: Determine the final N candidate aerosol types within each region according to the clustering results of Step 4.

2.

6. The aerosol optical depth inversion method based on multi-band satellite data according to claim 5, wherein: Step 5: Use the 6SV model to construct an AOD inversion lookup table for each band and each type of aerosol. The parameters included in the lookup table are the solar zenith angle, the observation zenith angle, the relative azimuth angle, the atmospheric model, AOD, surface elevation, BRDF - independent scattering kernel, BRDF - volume scattering kernel, BRDF - surface geometric kernel, and the simulated apparent reflectance.

7. The aerosol optical depth inversion method based on multi-band satellite data according to claim 3, characterized in that: Step 7: Perform quality control on the inversion results obtained in Step 6 according to the pixel identification mask established during the pixel screening process in Step 2.2.3, that is, eliminate the high-value AOD pixels within the range of the cloud pixel radius L km to obtain the final result.

Images