A time series AOD reconstruction method based on multi-sensor remote sensing

Abstract

The present invention relates to a time series AOD reconstruction method based on multi-sensor remote sensing, including: data collection and preprocessing; generating effective AOD data sets and missing AOD data sets; combining the effective AOD data sets with a Bayesian automatic optimization parameter adjustment algorithm to find the optimal hyperparameters of an integrated learning CatBoost model; constructing a CatBoost model using the effective AOD data sets and the optimal hyperparameters; inputting the missing AOD data sets into the CatBoost model to reconstruct the missing AOD data; combining the ground-based site AERONET AOD data to verify the reconstructed AOD data; merging the reconstructed AOD data and the effective AOD data to form an AOD product with complete space-time coverage. The method is conducive to providing an AOD product with high spatial resolution, wide spatial coverage and strong temporal continuity, and improving the accuracy of satellite remote sensing monitoring of aerosol optical thickness.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing monitoring, and in particular relates to a time series AOD reconstruction method based on multi-sensor remote sensing. Background Art

[0002] Atmospheric aerosol is an important factor in studying the global climate and atmospheric environment, and is one of the main pollutants that affect the quality of my country's atmospheric environment. AOD, as one of the most important parameters of atmospheric aerosol, is an important physical quantity that characterizes atmospheric turbidity or the total content of aerosols in the atmosphere. At present, the methods for obtaining AOD mainly include ground-based observations and satellite remote sensing observations. Ground-based observations have the advantages of high accuracy and strong temporal continuity, but insufficient spatial coverage; satellite remote sensing observations have the advantages of being dynamic, macroscopic, and fast, and provide a powerful monitoring method for obtaining information on a wide range of spatiotemporal coverage, changing trends, and aerosol transport sources. The spatial coverage is good, but the time is not continuous. Due to factors such as cloud cover, snow or water, there is a phenomenon of data loss, resulting in data loss in time and space. Summary of the invention

[0003] The purpose of the present invention is to provide a time-series AOD reconstruction method based on multi-sensor remote sensing, which is conducive to providing AOD products with high spatial resolution, wide spatial coverage and strong temporal continuity, and improving the accuracy of satellite remote sensing monitoring of aerosol optical thickness.

[0004] To achieve the above object, the technical solution adopted by the present invention is: a time series AOD reconstruction method based on multi-sensor remote sensing, comprising the following steps:

[0005] Step S1, data collection and preprocessing: collect ground-based site AERONETAOD data, different satellite AOD products, terrain elevation data, and meteorological data products, perform format conversion, resampling, reprojection, clipping, and numerical conversion operations on different satellite AOD products to achieve consistency of AOD product data monitored by different satellite sensors, and clip and resample terrain elevation data and meteorological data;

[0006] Step S2, fusing the pre-processed AOD products of different satellites to generate fused AOD products, wherein the fused AOD products with values ​​are called valid AOD data;

[0007] Step S3: for the fused AOD product, according to whether the AOD has a value, the effective AOD data set and the missing AOD data set are generated respectively by combining the annual accumulation day, terrain elevation data, and meteorological data;

[0008] Step S4: Combine the effective AOD data set with the Bayesian automatic optimization parameter adjustment algorithm to find the optimal hyperparameters of the ensemble learning CatBoost model;

[0009] Step S5, constructing a CatBoost model using the optimal hyperparameters obtained in step S4 with the effective AOD data set;

[0010] Step S6, inputting the missing AOD data set into the CatBoost model constructed in step S5 to reconstruct the missing AOD data and generate reconstructed AOD data;

[0011] Step S7, verifying the reconstructed AOD data by combining the AERONETAOD data of the ground-based site;

[0012] Step S8: merge the reconstructed AOD data and the valid AOD data to form an AOD product with complete temporal and spatial coverage.

[0013] Furthermore, the step S1 is specifically implemented as follows:

[0014] Step S11, collecting AERONETAOD data observed by base stations in the study area;

[0015] Step S12: Collect different satellite AOD products, including: MAIAC AOD, an AOD product provided by NASA using Terra and Aqua MODIS data based on a multi-angle atmospheric correction algorithm, and L3 hourly AOD monitored by the Himawari-8 satellite AHI released by the Satellite Program Division of the Japan Meteorological Agency. merged Product AHIAOD;

[0016] Step S13, collecting terrain elevation data;

[0017] Step S14, collecting meteorological data products, mainly from the fifth generation reanalysis product ERA5 released by the European Centre for Medium-Range Weather Forecasts, including: pressure, precipitation, 10mU type wind, 10mV type wind, temperature, relative humidity, total column water and boundary layer height;

[0018] Step S15, calculating the daily average product of the AHIAOD hourly product;

[0019] Step S16, resampling, reprojecting, clipping and numerical conversion operations are performed on different satellite AOD products to standardize them into the same data format; resampling and reprojecting the daily average value of AHIAOD using the MAIACAOD product as a template to make it have the same spatial resolution, clipping it using the study area as a template to achieve spatial consistency, dividing the MAIACAOD value by 1000 to make the MAIACAOD value between 0 and 2; processing the AHIAOD product at 500nm into the AOD at 550nm consistent with MAIACAOD according to the Angstrom empirical equation to achieve numerical consistency;

[0020] The Angstrom empirical equation is shown in formula (1):

[0021] τ λ =βλ -a (1)

[0022] In the formula, τ λ is the aerosol optical depth corresponding to the wavelength λ; β is the turbidity coefficient, obtained according to the equation relationship between the wavelength 500nm, AOD value and the corresponding wavelength index a; λ is the wavelength; a is the Angstrom wavelength index, obtained from the Angstrom variable provided by the L3 AHI AOD product;

[0023] Step S17: The terrain elevation data and meteorological data are clipped using the study area as a template, and resampled using the MAIAC AOD data as a template to achieve spatial consistency.

[0024] Furthermore, the step S2 is specifically implemented as follows:

[0025] Step S21, converting the AOD product pre-processed in step S16 into a numerical matrix;

[0026] Step S22, fusing the converted numerical matrix according to formula (2);

[0027]

[0028] Where P (i,j) is the AOD value of the image at (i, j) after AOD fusion; M (ij) is the AOD value of MAIACAOD at (i, j); H (i,j) is the AOD value of AHIAOD at (i, j); T means AOD has a value, and F means AOD has no value.

[0029] Furthermore, the step S3 is specifically implemented as follows:

[0030] Step S31, forming a spatially corresponding data set according to grid coordinates by combining the annual accumulated data, terrain elevation data, meteorological data, and the AOD product fused in step S2;

[0031] Step S32: Generate a valid AOD data set and a missing AOD data set according to whether the AOD value in the data set is valid.

[0032] Furthermore, the step S4 is specifically implemented as follows:

[0033] Step S41, dividing the valid AOD data set generated in step S3 into a verification data set and a training data set at a ratio of 3:7;

[0034] Step S42, standardize the verification data and the training data so that the data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1;

[0035] Step S43, using the Bayesian automatic optimization parameter adjustment algorithm to adjust the main parameters of the CatBoost model: depth, iterations, where the value range of depth is 2 to 16, the value range of iterations is 2 to 500, the root mean square error is set as the loss function, and the optimal hyperparameters of the CatBoost model are obtained by training the training data set;

[0036] Step S44: Use the verification data set to verify whether the CatBoost model with the optimal hyperparameters can be used for the reconstruction of AOD. If the correlation coefficient is greater than the set value, it is considered that the CatBoost model with the optimal hyperparameters can be used for the reconstruction of AOD.

[0037] Furthermore, the step S5 is specifically implemented as follows:

[0038] The CatBoost model is constructed using the optimal hyperparameters obtained in step S4 using the effective AOD dataset, as shown in formula (3):

[0039] AOD valid =f(Day,Elev,Mete d ) (3)

[0040] Where AOD valid is the effective AOD data, f is the CatBoost model, Day is the annual accumulation day, Elev is the terrain elevation data, Mete d It is a matrix composed of 8 meteorological variables, including pressure, precipitation, 10mU wind, 10mV wind, temperature, relative humidity, total column water and boundary layer height.

[0041] Furthermore, the step S6 is specifically implemented as follows:

[0042] For the missing AOD data set, the reconstruction of the missing AOD data is achieved through the constructed CatBoost model, as shown in formula (4):

[0043] AOD predict =f(Day,Elev,Mete d ) (4)

[0044] Where AOD predict It is the reconstructed AOD.

[0045] Furthermore, the step S7 is specifically implemented as follows:

[0046] Step S71, extracting the corresponding reconstructed AOD value according to the location of the AERONET ground-based site;

[0047] Step S72: Take the AOD value monitored by the AERONET ground-based site as the true value to verify the AOD value extracted and reconstructed in step S71, calculate the correlation coefficient R and the root mean square error RMSE, and evaluate the reconstructed result.

[0048] Compared with the prior art, the present invention has the following beneficial effects: the method can effectively give play to the advantages of multi-platform complementary observations, use machine learning CatBoost to predict and reconstruct the missing AOD, make up for the phenomenon of missing data, and provide AOD products with high spatial resolution, wide spatial coverage and strong temporal continuity. It can also improve the accuracy of satellite remote sensing monitoring of AOD and provide support for atmospheric environmental quality analysis, assessment and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The present invention is a flowchart of a method implementation according to an embodiment of the present invention.

[0050] Figure 2 This is a distribution map generated by the MAIACAOD product produced by the MODIS satellite in the Beijing-Tianjin-Hebei region for 5 consecutive days from November 30 to December 4, 2017 in the embodiment of the present invention, and the spatial distribution is missing.

[0051] Figure 3 This is a distribution map of the daily product generated by the AHI L3 hourly AOD product produced by the Himawari-8 satellite in the Beijing-Tianjin-Hebei region for 5 consecutive days from November 30 to December 4, 2017 in the embodiment of the present invention, and the spatial distribution is missing.

[0052] Figure 4 This is a distribution map generated by fusing the MAIACAOD product and the AHI AOD daily product for five consecutive days in the Beijing-Tianjin-Hebei region according to the method in the embodiment of the present invention.

[0053] Figure 5 This is an AOD product distribution map with complete AOD spatial coverage in the Beijing-Tianjin-Hebei region for 5 consecutive days generated by this method in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0055] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0056] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0057] like Figure 1 As shown, this embodiment provides a time series AOD reconstruction method based on multi-sensor remote sensing, comprising the following steps:

[0058] Step S1, data collection and preprocessing: collect AERONETAOD data from ground-based sites, different satellite AOD products, terrain elevation data, and meteorological data products, perform format conversion, resampling, reprojection, clipping, and numerical conversion operations on different satellite AOD products to achieve consistency in AOD product data monitored by different satellite sensors, and clip and resample terrain elevation data and meteorological data.

[0059] Step S2: Fuse the pre-processed AOD products of different satellites to generate fused AOD products, where the fused AOD products with values ​​are called valid AOD data.

[0060] Step S3: for the fused AOD product, according to whether AOD has a value, the effective AOD dataset and the missing AOD dataset are generated by combining the annual accumulation day, terrain elevation data, and meteorological data.

[0061] Step S4: Combine the effective AOD data set with the Bayesian automatic optimization parameter adjustment algorithm to find the optimal hyperparameters of the ensemble learning CatBoost model.

[0062] Step S5: construct a CatBoost model using the optimal hyperparameters obtained in step S4 with the effective AOD dataset.

[0063] Step S6: Input the missing AOD data set into the CatBoost model constructed in step S5 to reconstruct the missing AOD and generate reconstructed AOS data.

[0064] Step S7: Verify the reconstructed AOD data by combining the AERONETAOD data of the ground-based site.

[0065] Step S8: merge the reconstructed AOD data and the valid AOD data to form an AOD product with complete temporal and spatial coverage.

[0066] In this embodiment, step S1 is specifically implemented as follows:

[0067] Step S11, collect AERONETAOD data observed by base stations in the Beijing-Tianjin-Hebei region, mainly including: Beijing station data, Beijing-CAMS station data.

[0068] Step S12: Collect different satellite AOD products, including: AOD products provided by NASA (National Aeronautics and Space Administration) using Terra and AquaMODIS data and based on the Multi-Angle Atmospheric Correction (MAIAC) algorithm (hereinafter referred to as MAIAC AOD), and L3 hourly AOD monitored by the Himawari-8 satellite AHI released by the Satellite Program Division of the Japan Meteorological Agency. merged Product (hereinafter referred to as AHIAOD).

[0069] Step S13: Collect terrain elevation data, which mainly comes from the 2015 national 1km SRTM terrain elevation data provided by the Resource and Environmental Science and Data Center of the Chinese Academy of Sciences, which is generated by resampling the latest SRTM V4.1 data.

[0070] Step S14, collect meteorological data products, mainly from the fifth generation reanalysis product ERA5 released by the European Centre for Medium-Range Weather Forecasts, including: pressure, precipitation, 10mU type wind, 10mV type wind, temperature, relative humidity, total column water and boundary layer height.

[0071] Step S15, calculating the daily average product of the AHIAOD hourly product.

[0072] Step S16, resampling, reprojecting, clipping and numerical conversion operations are performed on different satellite AOD products to standardize them into the same data format; resampling and reprojecting the daily average value of AHIAOD using the MAIACAOD product as a template to make it have the same spatial resolution, clipping it using the study area as a template to achieve spatial consistency, dividing the MAIACAOD value by 1000 to make the overall MAIACAOD value roughly between 0 and 2; processing the AHIAOD product at 500nm into the AOD at 550nm consistent with MAIACAOD according to the Angstrom empirical equation to achieve numerical consistency;

[0073] The Angstrom empirical equation is shown in formula (1):

[0074] τ λ =βλ -a (1)

[0075] In the formula, τλ is the aerosol optical depth corresponding to the wavelength λ; β is the turbidity coefficient, which is obtained according to the equation between the wavelength of 500nm, AOD value and the corresponding wavelength index a; λ is the wavelength; a is the Angstrom wavelength index, which is obtained from the Angstrom variable provided by the L3 AHI AOD product.

[0076] Step S17: The terrain elevation data and meteorological data are clipped using the study area as a template, and resampled using the MAIAC AOD data as a template to achieve spatial consistency.

[0077] In this embodiment, step S2 is specifically implemented as follows:

[0078] Step S21, converting the AOD product preprocessed in step S16 into a numerical matrix.

[0079] Step S22, fusing the converted numerical matrix according to formula (2);

[0080]

[0081] Where P (i,j) is the AOD value of the image at (i, j) after AOD fusion; M (ij) is the AOD value of MAIACAOD at (i, j); H (i,j) is the AOD value of AHIAOD at (i, j); T means AOD has a value, and F means AOD has no value.

[0082] In this embodiment, step S3 is specifically implemented as follows:

[0083] Step S31, the annual accumulated data, terrain elevation data, meteorological data, and the AOD product fused in step S2 are formed into a spatially corresponding data set according to grid coordinates.

[0084] Step S32: Generate a valid AOD data set (including: annual accumulation day, terrain elevation data, meteorological data, valid AOD) and a missing AOD data set (including: annual accumulation day, terrain elevation data, meteorological data) according to whether the AOD value in the data set is valid.

[0085] In this embodiment, step S4 is specifically implemented as follows:

[0086] Step S41, dividing the effective AOD data set (annual accumulation day, terrain elevation data, meteorological data, effective AOD) generated in step S3 into a verification data set and a training data set at a ratio of 3:7;

[0087] Step S42, standardize the verification data and the training data so that the data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1;

[0088] Step S43, using the Bayesian automatic optimization parameter adjustment algorithm to adjust the main parameters of the CatBoost model: depth, iterations, where the value range of depth is 2 to 16, the value range of iterations is 2 to 500, and the root mean square error (RMSE) is set as the loss function, and the optimal hyperparameters of the CatBoost model are obtained by training the training data set;

[0089] Step S44: Use the verification data set to verify whether the CatBoost model with the optimal hyperparameters can be used for the reconstruction of AOD. If the correlation coefficient is greater than the set value, it is considered that the CatBoost model with the optimal hyperparameters can be used for the reconstruction of AOD.

[0090] It has been verified that when the correlation coefficient R2 is greater than 0.6, the CatBoost model with the optimal hyperparameters can be used for the reconstruction of AOD;

[0091] Step S45, the optimal hyperparameters of the CatBoost model obtained by the Bayesian automatic parameter tuning algorithm are as follows: depth: 15, iterations: 500.

[0092] In this embodiment, step S5 is specifically implemented as follows:

[0093] The CatBoost model is constructed using the optimal hyperparameters obtained in step S4 using the effective AOD dataset, as shown in formula (3):

[0094] AOD valid =f(Day,Elev,Mete d ) (3)

[0095] Where AOD valid is the effective AOD data, f is the CatBoost model, Day is the annual accumulation day, Elev is the terrain elevation data, Mete d It is a matrix composed of 8 meteorological variables, including pressure, precipitation, 10mU wind, 10mV wind, temperature, relative humidity, total column water and boundary layer height.

[0096] In this embodiment, step S6 is specifically implemented as follows:

[0097] For the missing AOD data set, the reconstruction of the missing AOD data is achieved through the constructed CatBoost model, as shown in formula (4):

[0098] AOD predict =f(Day,Elev,Meted ) (4)

[0099] Where AOD predict It is the reconstructed AOD.

[0100] In this embodiment, step S7 is specifically implemented as follows:

[0101] Step S71, extracting the corresponding reconstructed AOD value according to the location of the AERONET ground-based site;

[0102] Step S72: Take the AOD value monitored by the AERONET ground-based site as the true value to verify the AOD value extracted and reconstructed in step S71, calculate the correlation coefficient R and the root mean square error RMSE, and evaluate the reconstructed result.

[0103] In this embodiment, the MAIACAOD product and AHIAOD product for five consecutive days in the Beijing-Tianjin-Hebei region in 2017 are fused and reconstructed. Figure 2 This is the distribution map generated by MAIACAOD products for the Beijing-Tianjin-Hebei region for 5 consecutive days. Figure 3 The distribution map of daily products generated from AHI L3 hourly AOD products for the Beijing-Tianjin-Hebei region for 5 consecutive days shows that there are missing spatial distributions. Figure 4 This is the distribution map generated by fusing the MAIACAOD product and AHI AOD daily product for five consecutive days in the Beijing-Tianjin-Hebei region using the method of the present invention. Compared with a single sensor, the spatial coverage is relatively complete. Figure 5 The distribution map generated by the method of the present invention for the prediction of the AOD fusion products of different sensors in the Beijing-Tianjin-Hebei region for 5 consecutive days shows that the method of the present invention successfully integrates the AOD fusion and prediction of multi-satellite sensor remote sensing images and achieves complete spatial coverage.

[0104] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A time series AOD reconstruction method based on multi-sensor remote sensing, characterized in that: The following steps are involved: Step S1, data collection and preprocessing: collect ground-based site AERONETAOD data, different satellite AOD products, terrain elevation data, and meteorological data products, perform format conversion, resampling, reprojection, clipping, and numerical conversion operations on different satellite AOD products to achieve consistency of AOD product data monitored by different satellite sensors, and clip and resample terrain elevation data and meteorological data; Step S2, fusing the pre-processed AOD products of different satellites to generate fused AOD products, wherein the fused AOD products with values ​​are called valid AOD data; Step S3: for the fused AOD product, according to whether the AOD has a value, the effective AOD data set and the missing AOD data set are generated respectively by combining the annual accumulation day, terrain elevation data, and meteorological data; Step S4: Combine the effective AOD data set with the Bayesian automatic optimization parameter adjustment algorithm to find the optimal hyperparameters of the ensemble learning CatBoost model; Step S5, constructing a CatBoost model using the optimal hyperparameters obtained in step S4 with the effective AOD data set; Step S6, inputting the missing AOD data set into the CatBoost model constructed in step S5 to reconstruct the missing AOD data to obtain reconstructed AOD data; Step S7, verifying the reconstructed AOD data by combining the AERONETAOD data of the ground-based site; Step S8, merging the reconstructed AOD data and the valid AOD data to form an AOD product with complete temporal and spatial coverage; The step S1 is specifically implemented as follows: Step S11, collecting AERONETAOD data observed by base stations in the study area; Step S12: Collect different satellite AOD products, including: MAIACAOD, an AOD product provided by NASA using Terra and AquaMODIS data based on a multi-angle atmospheric correction algorithm, and L3 hourly AOD monitored by the Himawari-8 satellite AHI released by the Satellite Program Division of the Japan Meteorological Agency. merged Product AHIAOD; Step S13, collecting terrain elevation data; Step S14, collecting meteorological data products, mainly from the fifth generation reanalysis product ERA5 released by the European Centre for Medium-Range Weather Forecasts, including: pressure, precipitation, 10mU type wind, 10mV type wind, temperature, relative humidity, total column water and boundary layer height; Step S15, calculating the daily average product of the AHIAOD hourly product; Step S16, resampling, reprojecting, clipping and numerical conversion operations are performed on different satellite AOD products to standardize them into the same data format; resampling and reprojecting the daily average value of AHI AOD using the MAIAC AOD product as a template to make it have the same spatial resolution, clipping it using the study area as a template to achieve spatial consistency, dividing the MAIAC AOD value by 1000 to make the MAIAC AOD value between 0 and 2; processing the AHI AOD product at 500nm into the AOD at 550nm consistent with the MAIAC AOD according to the Angstrom empirical equation to achieve numerical consistency; The Angstrom empirical equation is shown in formula (1): t λ =see -a (1) In the formula, τ λ is the aerosol optical depth corresponding to the wavelength λ; β is the turbidity coefficient, obtained according to the equation relationship between the wavelength 500nm, AOD value and the corresponding wavelength index a; λ is the wavelength; a is the Angstrom wavelength index, obtained from the Angstrom variable provided by the L3 AHI AOD product; Step S17, the terrain elevation data and meteorological data are clipped using the study area as a template, and resampled using the MAIACAOD data as a template to achieve spatial consistency; The step S2 is specifically implemented as follows: Step S21, converting the AOD product pre-processed in step S16 into a numerical matrix; Step S22, fusing the converted numerical matrix according to formula (2); Where P (i,j) is the AOD value of the image at (i, j) after AOD fusion; M (ij) is the AOD value of MAIACAOD at (i, j); H (i,j) is the AOD value of AHIAOD at (i, j); T means AOD has a value, and F means AOD has no value; The step S3 is specifically implemented as follows: Step S31, forming a spatially corresponding data set according to grid coordinates by combining the annual accumulated data, terrain elevation data, meteorological data, and the AOD product fused in step S2; Step S32, generating a valid AOD data set and a missing AOD data set according to whether the AOD value in the data set is valid; The step S4 is specifically implemented as follows: Step S41, dividing the valid AOD data set generated in step S3 into a verification data set and a training data set at a ratio of 3:7; Step S42, standardize the verification data and the training data so that the data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1; Step S43, using the Bayesian automatic optimization parameter adjustment algorithm to adjust the main parameters of the CatBoost model: depth, iterations, where the value range of depth is 2 to 16, the value range of iterations is 2 to 500, the root mean square error is set as the loss function, and the optimal hyperparameters of the CatBoost model are obtained by training the training data set; Step S44: using the verification data set to verify whether the CatBoost model configured with the optimal hyperparameters can be used for the reconstruction of AOD, and if the correlation coefficient is greater than the set value, it is considered that the CatBoost model configured with the optimal hyperparameters can be used for the reconstruction of AOD; The step S5 is specifically implemented as follows: The CatBoost model is constructed using the optimal hyperparameters obtained in step S4 using the effective AOD dataset, as shown in formula (3): AOD valid =f(Day,Elev,Mete d ) (3) Where AOD valid is the effective AOD data, f is the CatBoost model, Day is the annual accumulation day, Elev is the terrain elevation data, Mete d It is a matrix composed of 8 meteorological variables, including: pressure, precipitation, 10mU wind, 10mV wind, temperature, relative humidity, total column water and boundary layer height; The step S6 is specifically implemented as follows: For the missing AOD data set, the reconstruction of the missing AOD data is achieved through the constructed CatBoost model, as shown in formula (4): AOD predict =f(Day,Elev,Mete d ) (4) Where AOD predict It is the reconstructed AOD.

2. The method for reconstructing time series AOD based on multi-sensor remote sensing according to claim 1, characterized in that: The step S7 is specifically implemented as follows: Step S71, extracting the corresponding reconstructed AOD value according to the location of the AERONET ground-based site; Step S72: Take the AOD value monitored by the AERONET ground-based site as the true value to verify the AOD value extracted and reconstructed in step S71, calculate the correlation coefficient R and the root mean square error RMSE, and evaluate the reconstructed result.

Images