An automatic monitoring method for atmospheric dust sources based on the Environmental Satellite No. 2

Abstract

This invention discloses an automatic monitoring method for atmospheric dust sources based on the Environmental Satellite No. 2. By using machine learning to calculate surface reflectance, invert aerosol optical depth, and identify and classify dust sources, a rapid remote sensing monitoring method for dust sources is constructed to comprehensively and accurately reflect changes in dust source pollution. This invention not only rapidly identifies dust sources at high frequency but also meticulously classifies them, effectively guiding the precise control of dust sources. This overcomes the limitations of traditional prediction methods, such as insufficient monitoring coverage, high investment costs, and inaccurate control support, which can lead to an inability to accurately control dust sources.

Description

Technical Field

[0001] The present invention relates to the technical field of atmospheric environmental pollution source supervision, and in particular to an automatic monitoring method for atmospheric dust sources based on the Environmental Satellite No. 2. Background Art

[0002] Dust emissions are one of the primary contributors to atmospheric particulate matter in northern Chinese cities. Surface wind erosion in urban areas and surrounding suburbs, in particular, has a significant impact on urban air quality. It is a major source of atmospheric particulate matter in many northern Chinese cities and is receiving increasing public attention. Accurately understanding the location, control, and emission characteristics of dust sources is crucial for dust regulation. Dust sources are unorganized and open, characterized by uncertain source intensity, random emissions, and difficulty in quantification. Traditional fixed monitoring instruments and manual field surveys, while highly accurate, are time-consuming and labor-intensive, have poor spatial scalability, and are difficult to adapt to the widespread and rapidly changing nature of dust sources.

[0003] As an emerging technology, satellite remote sensing boasts macroscopic, dynamic, objective, and accurate characteristics. Compared to traditional ground-based monitoring methods, it offers unique advantages in spatial coverage. High-resolution multispectral monitoring continuously captures the distribution of dust sources and the spatiotemporal variations in land use distribution over large areas. This effectively supports precise monitoring of regional dust sources and promotes sustained improvement in regional air quality. On September 27, 2020, the Environmental Disaster Reduction-2A and -2B satellites (referred to as Environmental-2 satellites) were successfully launched from the Taiyuan Satellite Launch Center. Each satellite is equipped with four optical payloads: a 16m camera, a hyperspectral imager, an infrared camera, and an atmospheric correction instrument. The 16m camera payload consists of four 16m-resolution visible-light CCD cameras, which provide multispectral imagery with a field of view of 800 km through stitching. The two satellites, operating in the same orbit, can rapidly acquire ground imagery. They provide high-resolution scanning of visible and near-infrared multispectral data, covering the entire country every two days, meeting the needs for refined and frequent monitoring and regulation of dust pollution sources.

[0004] Currently, dust source pollution monitoring and analysis are mainly based on manual field inspections and real-time measurements using high-resolution satellite visual interpretation. However, this is time-consuming, labor-intensive, and lacks accuracy. On the one hand, manual field inspections require high human, material, and financial resources. At the same time, urban construction and land changes are relatively rapid, making it difficult for general cities to conduct high-frequency inspections of high-pollution dust sources. Therefore, it is difficult to monitor and investigate dust sources over a large area based solely on manual inspections, and it is impossible to accurately locate specific areas. On the other hand, to compensate for the lack of coverage of manual inspections, some studies have introduced high-resolution satellite remote sensing visual interpretation to identify bare ground patches and provide target reference areas for dust source supervision. In practice, visual interpreters can only identify bare ground and cannot determine the pollution impact of dust sources. In addition, there are certain differences in the visual interpretation and identification results of different personnel. The workload of investigating each one is also relatively large, requiring a lot of manpower and time costs. At the same time, visual interpretation often requires sub-meter high-resolution satellite data, which also limits the frequency of satellite monitoring. Therefore, the traditional dust source monitoring methods have insufficient monitoring coverage, high investment costs, and inaccurate control support, which will lead to the inability to quickly, accurately, and automatically detect and supervise dust source pollution.

[0005] Therefore, in order to solve the above problems, this application proposes an automatic monitoring method for atmospheric dust sources based on the Environmental Satellite No. 2. Through machine learning to calculate surface reflectivity, aerosol optical thickness inversion and dust source identification and classification, a remote sensing rapid monitoring technology method for dust sources is constructed to comprehensively and accurately reflect the changes in dust source pollution. It can not only quickly identify dust sources at a high frequency, but also finely classify dust sources, and efficiently guide the precise control of dust sources, thereby making up for the limitations of traditional prediction methods such as insufficient monitoring coverage, high investment costs, and inaccurate control support, which lead to the inability to accurately control dust sources. Summary of the Invention

[0006] The purpose of the present invention is to fill the gaps in the prior art and provide an automatic monitoring method for atmospheric dust sources based on the Environmental Satellite No. 2. Through machine learning to calculate surface reflectivity, aerosol optical thickness inversion and dust source identification and classification, a remote sensing rapid monitoring technical method for dust sources is constructed to comprehensively and accurately reflect the changes in dust source pollution. It can not only quickly identify dust sources at a high frequency, but also finely classify dust sources, and efficiently guide the precise control of dust sources, thereby making up for the limitations of traditional prediction methods such as insufficient monitoring coverage, high investment costs, and inaccurate control support, which lead to the inability to accurately control dust sources.

[0007] In order to achieve the above object, the present invention provides a method for automatically monitoring atmospheric dust sources based on the Environmental Satellite No. 2, the method comprising the steps of:

[0008] S1. Based on the CCD image data of the Environmental Satellite-2 and combined with the aerosol products of MODIS, a random forest method was used to establish a remote sensing estimation model for the surface reflectance of the Environmental Satellite-2.

[0009] S2. Based on the CCD images of the Huanhuan-2 satellite, the atmospheric radiation transfer model is used to invert and obtain high-resolution aerosol optical depth;

[0010] S3. Extract the aerosol spatial maximum grid cells based on the high-resolution aerosol optical depth and calculate the reliability of the high-value grid cells;

[0011] S4. Use high-resolution images to screen out dust sources from the underlying surface of high-value aerosol optical depth areas and interpret and classify high-value aerosol optical depth areas.

[0012] S1 includes:

[0013] S1.1 Collect CCD image data of the Environmental Protection Satellite No. 2 in each quarter of the urban built-up area, extract the 550nm aerosol optical depth product data from the MODIS aerosol product (MCD19A2) of the corresponding date, and reproject it according to the projection method of the Environmental Protection Satellite No. 2 data.

[0014] S1.2 Extract calibration coefficients and satellite observation angles from the auxiliary XML file of the Huanhuan-2 CCD image. Calculate the apparent reflectance of the built-up area in the blue band (around 0.47 μm), green band (around 0.55 μm), red band (around 0.66 μm), near-infrared band (around 0.76 μm), and red-edge band of the CCD image. Simultaneously, extract the aerosol optical depth parameter at each pixel of the CCD image from the MODIS aerosol optical depth product.

[0015] S1.3 Use the 6SV atmospheric radiation transfer model to perform atmospheric correction on the red band (around 0.66 μm) and blue band (around 0.47 μm) of the CCD image based on the observation geometry, observation date, band spectral response function apparent reflectance, and aerosol optical depth to obtain the surface reflectance of the two bands of the CCD image;

[0016] S1.4 uses a stratified sampling method to randomly obtain the solar zenith angle, satellite observation zenith angle, relative azimuth, and apparent reflectance of the blue band (near 0.47μm), green band (near 0.55μm), red band (near 0.66μm), near-infrared band (near 0.83μm), and red-edge band (near 0.71μm) for 1000 pixels in each season in urban built-up areas. At the same time, the surface reflectance of the blue band (near 0.47μm) and red band (near 0.66μm) of the corresponding pixels is extracted to form a training sample dataset;

[0017] S1.5 is based on the training samples established in S1.4. The solar zenith angle, satellite observation zenith angle, relative azimuth, blue band (near 0.47μm), green band (near 0.55μm), red band (near 0.66μm), near-infrared band (near 0.83μm) and red edge band (near 0.71μm) apparent reflectance are used as independent variables, and the surface reflectance is used as the dependent variable. The random forest method is used for fitting training to establish the surface reflectance inversion model of the red and blue bands of the CCD camera of the Environmental-2 satellite.

[0018] S2 includes:

[0019] S2.1: Based on the real-time environmental No. 2 CCD camera image, extract the calibration coefficients from the auxiliary file and calculate the apparent reflectance of the blue band (near 0.47 μm), green band (near 0.55 μm), red band (near 0.66 μm), near-infrared band (near 0.83 μm), and red-edge band (near 0.71 μm) according to the calibration coefficients. Areas with an apparent reflectance of the red band greater than 0.2 are considered cloud-covered areas.

[0020] S2.2 Read the solar zenith angle, satellite zenith angle, and relative azimuth from the auxiliary file of the Real-Time Environment II CCD camera image, match the apparent reflectance data in step S2.1 to form an input parameter dataset, and obtain the red and blue band surface reflectances of the cloud-free land pixels of the Real-Time Environment II CCD camera based on the surface reflectance inversion model of the Real-Time Environment II CCD camera established in step S1.5;

[0021] S2.3, based on the 6SV atmospheric radiation transfer model and the red and blue band surface reflectances in S2.2, set the AODs to 0, 0.3, 0.6, 0.9, 1.2, 1.5, 1.8, 2.1, 2.4, 2.7, and 3.0, respectively. The aerosol model is set to dust aerosol. The red and blue band apparent reflectances at different AODs under specific observation conditions such as solar zenith angle, satellite zenith angle, and relative azimuth angle are simulated by reading the auxiliary file of the real-time environmental No. 2 CCD camera image.

[0022] S2.4 Calculate the simulated apparent reflectance errors for the red and blue bands based on the simulated apparent reflectance at different AODs in S2.3. The calculation method is as follows:

[0023]

[0024] in, represents the apparent reflectance simulation error, is the apparent reflectance of the blue band observed by satellite, To simulate and calculate the apparent reflectance of the blue band, is the apparent reflectance of the red band observed by satellite, Calculate the apparent reflectance of the red band for simulation;

[0025] S2.5 uses the least squares method to fit the relationship between the apparent reflectance error and AOD for each pixel using a quadratic polynomial based on the apparent reflectance error at different AODs in S2.4. The AOD at which the simulated apparent reflectance error is minimized is calculated as the inverted AOD for that pixel, thereby obtaining the AOD of the cloud-free land pixels in the entire image.

[0026] S3 includes:

[0027] S3.1: For the built-up area of ​​the target region, a buffer zone with a radius of 3 km is established. The high-resolution cloud-free land pixel AOD obtained by inversion in S2.5 is clipped using the buffer zone to obtain the cloud-free AOD of the built-up area and surrounding areas.

[0028] S3.2 Based on the high-resolution cloud-free AOD of the built-up area and surrounding areas obtained by inversion in S2.5, select pixels with AOD greater than 1 as high-AOD pixels;

[0029] S3.3 Based on the potential high AOD pixels extracted in step S3.2, the red band surface reflectance obtained in step S2.2 and the high-resolution cloud-free land pixel AOD obtained in step S2.5, for the high AOD pixels, compare the pixel AOD, red band surface reflectance, and vegetation index with the background pixels to determine whether they are potential dust source pixels. If the following equations (1) to (3) are satisfied, the pixel is determined to be a potential dust source pixel; otherwise, it is a non-dust source pixel:

[0030]

[0031] Wherein, AOD is the AOD value of the potential dust source pixel, dimensionless; is the average AOD of pixels within 1 km around the potential dust source pixel, dimensionless; is the mean absolute deviation of AOD within 1 km around the potential dust source pixel, dimensionless; is the red band surface reflectance of the potential dust source pixel retrieved by satellite, dimensionless; is the average surface reflectance in the red band within 1 km of the potential dust source pixel, dimensionless; is the average deviation of the red band surface reflectance within 1 km of the potential dust source pixel, dimensionless; NDVI is the vegetation cover index of the potential dust source pixel, calculated by the red band surface reflectance and the near-infrared surface reflectance, dimensionless; is the average value of vegetation coverage index within 1 km of the potential dust source pixel, is the average deviation of vegetation cover index within 1 km of the pixel;

[0032] S3.4 Based on all potential dust source pixels extracted in step S3.3, use the ramp function to calculate the dust source credibility of the pixel. The calculation is shown as follows:

[0033]

[0034] in,

[0035] C1=S(AOD,1,5) (5)

[0036]

[0037] C3=S(0.3-NDVI,0,0.2) (7)

[0038]

[0039] S3.5 Extract the pixel center coordinates of potential dust source pixels with a credibility greater than 60, and create a potential dust source vector grid based on the pixel size.

[0040] S4 includes:

[0041] S4.1 Superimpose all potential dust source vector grids extracted in step S3.4 on high-resolution satellite images better than 2 meters from GF-1, GF-2, and GF-6, and select satellite images with an interval of no more than six months from the Huanwu-2 image;

[0042] S4.2 Visually interpret the high-resolution imagery of the underlying surface of potential dust source grids. Grids containing more than 80% of the land used for farmland, residential areas, industrial plants, etc., and no obvious bare land, shall be classified as false grids. All other potential dust source grids shall be classified as true dust source grids.

[0043] S4.3 Perform detailed interpretation of the high-resolution image of the underlying surface of the actual dust source grid, and conduct comprehensive analysis and classification of the sources of the actual dust source grid based on the underlying surface. The classification method is as follows: if bare land and industrial plants are found on the underlying surface, it is judged to be an industrial construction dust source; if bare land and sheds and houses are found on the underlying surface, it is judged to be a construction dust source; if bare land and mountains are found on the underlying surface, it is judged to be a mining dust source; if bare land and multiple traffic arteries are found on the underlying surface, it is judged to be a transportation dust source;

[0044] S4.4 performs erosion and expansion processing on the dust source grid, eliminates discrete independent grids, summarizes information such as coordinates, credibility, and dust source type, and forms a dust source list.

[0045] Compared with the existing technology, the present invention constructs a remote sensing rapid monitoring technology method for dust sources through machine learning calculation of surface reflectance, aerosol optical thickness inversion and dust source identification and classification to comprehensively and accurately reflect the changes in dust source pollution. It can not only quickly identify dust sources at high frequency, but also finely classify dust sources, and efficiently guide the precise control of dust sources, thereby making up for the limitations of traditional prediction methods such as insufficient monitoring coverage, high investment cost, and inaccurate control support, which lead to the inability to accurately control dust sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The present invention is an automatic monitoring method for atmospheric dust sources based on the environmental satellite No. 2 DETAILED DESCRIPTION

[0047] The present invention will now be further described with reference to the accompanying drawings.

[0048] See also Figure 1 , a method for automatically monitoring atmospheric dust sources based on the Environmental Satellite No. 2, characterized in that the method comprises the steps of:

[0049] S1. Based on the CCD image data of the Environmental Satellite-2 and combined with the aerosol products of MODIS, a random forest method was used to establish a remote sensing estimation model for the surface reflectance of the Environmental Satellite-2.

[0050] S2. Based on the CCD images of the Huanhuan-2 satellite, the atmospheric radiation transfer model is used to invert and obtain high-resolution aerosol optical depth;

[0051] S3. Extract the aerosol spatial maximum grid cells based on the high-resolution aerosol optical depth and calculate the reliability of the high-value grid cells;

[0052] S4. Use high-resolution images to screen out dust sources from the underlying surface of high-value aerosol optical depth areas and interpret and classify high-value aerosol optical depth areas.

[0053] In this embodiment, step S1 further includes:

[0054] S1.1 Collect CCD image data of the Environmental Protection Satellite No. 2 in each quarter of the urban built-up area, extract the 550nm aerosol optical depth product data from the MODIS aerosol product (MCD19A2) of the corresponding date, and reproject it according to the projection method of the Environmental Protection Satellite No. 2 data.

[0055] S1.2 Extract calibration coefficients and satellite observation angles from the auxiliary XML file of the Huanhuan-2 CCD image. Calculate the apparent reflectance of the built-up area in the blue band (around 0.47 μm), green band (around 0.55 μm), red band (around 0.66 μm), near-infrared band (around 0.76 μm), and red-edge band of the CCD image. Simultaneously, extract the aerosol optical depth parameter at each pixel of the CCD image from the MODIS aerosol optical depth product.

[0056] S1.3 Use the 6SV atmospheric radiation transfer model to perform atmospheric correction on the red band (around 0.66 μm) and blue band (around 0.47 μm) of the CCD image based on the observation geometry, observation date, band spectral response function apparent reflectance, and aerosol optical depth to obtain the surface reflectance of the two bands of the CCD image;

[0057] S1.4 adopts a stratified sampling method to randomly obtain the solar zenith angle, satellite observation zenith angle, relative azimuth, blue band (near 0.47μm), green band (near 0.55μm), red band (near 0.66μm), near-infrared band (near 0.83μm) and red edge band (near 0.71μm) apparent reflectance of 1000 pixels in urban built-up areas in each season, and simultaneously extract the surface reflectance of the blue wave (near 0.47μm) and red band (near 0.66μm) of the corresponding pixels to form a training sample data set.

[0058] S1.5 is based on the training samples established in S1.4. The solar zenith angle, satellite observation zenith angle, relative azimuth, blue band (near 0.47μm), green band (near 0.55μm), red band (near 0.66μm), near-infrared band (near 0.83μm) and red edge band (near 0.71μm) apparent reflectance are used as independent variables, and the surface reflectance is used as the dependent variable. The random forest method is used for fitting training to establish the surface reflectance inversion model of the red and blue bands of the CCD camera of the Environmental-2 satellite.

[0059] In this embodiment, step S2 further includes:

[0060] S2.1: Based on the real-time environmental No. 2 CCD camera image, extract the calibration coefficients from the auxiliary file and calculate the apparent reflectance of the blue band (near 0.47 μm), green band (near 0.55 μm), red band (near 0.66 μm), near-infrared band (near 0.83 μm), and red-edge band (near 0.71 μm) according to the calibration coefficients. Areas with an apparent reflectance of the red band greater than 0.2 are considered cloud-covered areas.

[0061] S2.2 Read the solar zenith angle, satellite zenith angle, and relative azimuth from the auxiliary file of the Real-Time Environment II CCD camera image, match the apparent reflectance data in step S2.1 to form an input parameter dataset, and obtain the red and blue band surface reflectances of the cloud-free land pixels of the Real-Time Environment II CCD camera based on the surface reflectance inversion model of the Real-Time Environment II CCD camera established in step S1.5;

[0062] S2.3, based on the 6SV atmospheric radiation transfer model and the red and blue band surface reflectances in S2.2, set the AODs to 0, 0.3, 0.6, 0.9, 1.2, 1.5, 1.8, 2.1, 2.4, 2.7, and 3.0, respectively. The aerosol model is set to dust aerosol. The red and blue band apparent reflectances at different AODs under specific observation conditions such as solar zenith angle, satellite zenith angle, and relative azimuth angle are simulated by reading the auxiliary file of the real-time environmental No. 2 CCD camera image.

[0063] S2.4 Calculate the simulated apparent reflectance errors for the red and blue bands based on the simulated apparent reflectance at different AODs in S2.3. The calculation method is as follows:

[0064]

[0065] in, represents the apparent reflectance simulation error, is the apparent reflectance of the blue band observed by satellite, To simulate and calculate the apparent reflectance of the blue band, is the apparent reflectance of the red band observed by satellite, Calculate the apparent reflectance of the red band for simulation;

[0066] S2.5 uses the least squares method to fit the relationship between the apparent reflectance error and AOD for each pixel using a quadratic polynomial based on the apparent reflectance error at different AODs in S2.4. The AOD at which the simulated apparent reflectance error is minimized is calculated as the inverted AOD for that pixel, thereby obtaining the AOD of the cloud-free land pixels in the entire image.

[0067] In this embodiment, step S3 further includes:

[0068] S3.1: For the built-up area of ​​the target region, a buffer zone with a radius of 3 km is established. The high-resolution cloud-free land pixel AOD obtained by inversion in S2.5 is clipped using the buffer zone to obtain the cloud-free AOD of the built-up area and surrounding areas.

[0069] S3.2 Based on the high-resolution cloud-free AOD of the built-up area and surrounding areas obtained by inversion in S2.5, select pixels with AOD greater than 1 as high-AOD pixels;

[0070] S3.3 Based on the potential high AOD pixels extracted in step S3.2, the red band surface reflectance obtained in step S2.2 and the high-resolution cloud-free land pixel AOD obtained in step S2.5, for the high AOD pixels, compare the pixel AOD, red band surface reflectance, and vegetation index with the background pixels to determine whether they are potential dust source pixels. If the following equations (1) to (3) are satisfied, the pixel is determined to be a potential dust source pixel; otherwise, it is a non-dust source pixel:

[0071]

[0072] Wherein, AOD is the AOD value of the potential dust source pixel, dimensionless; is the average AOD of pixels within 1 km around the potential dust source pixel, dimensionless; is the mean absolute deviation of AOD within 1 km around the potential dust source pixel, dimensionless; is the red band surface reflectance of the potential dust source pixel retrieved by satellite, dimensionless; is the average surface reflectance in the red band within 1 km of the potential dust source pixel, dimensionless; is the average deviation of the red band surface reflectance within 1 km of the potential dust source pixel, dimensionless; NDVI is the vegetation cover index of the potential dust source pixel, calculated by the red band surface reflectance and the near-infrared surface reflectance, dimensionless; is the average value of vegetation coverage index within 1 km of the potential dust source pixel, is the average deviation of vegetation cover index within 1 km of the pixel;

[0073] S3.4 Based on all potential dust source pixels extracted in step S3.3, use the ramp function to calculate the dust source credibility of the pixel. The calculation is shown as follows:

[0074]

[0075] in,

[0076] C1=S(AOD,1,5) (5)

[0077]

[0078] C3=S(0.3-NDVI,0,0.2) (7)

[0079]

[0080] S3.5 extracts the pixel center coordinates of potential dust source pixels with a credibility greater than 60, and creates a potential dust source vector grid based on the pixel size.

[0081] S4 includes:

[0082] S4.1 Superimpose all potential dust source vector grids extracted in step S3.4 on high-resolution satellite images better than 2 meters from GF-1, GF-2, and GF-6, and select satellite images with an interval of no more than six months from the Huanwu-2 image;

[0083] S4.2 Visually interpret the high-resolution imagery of the underlying surface of potential dust source grids. Grids containing more than 80% of the land used for farmland, residential areas, industrial plants, etc., and no obvious bare land, shall be classified as false grids. All other potential dust source grids shall be classified as true dust source grids.

[0084] S4.3 Perform detailed interpretation of the high-resolution image of the underlying surface of the actual dust source grid, and conduct comprehensive analysis and classification of the sources of the actual dust source grid based on the underlying surface. The classification method is as follows: if bare land and industrial plants are found on the underlying surface, it is judged to be an industrial construction dust source; if bare land and sheds and houses are found on the underlying surface, it is judged to be a construction dust source; if bare land and mountains are found on the underlying surface, it is judged to be a mining dust source; if bare land and multiple traffic arteries are found on the underlying surface, it is judged to be a transportation dust source;

[0085] S4.4 performs erosion and expansion processing on the dust source grid, eliminates discrete independent grids, summarizes information such as coordinates, credibility, and dust source type, and forms a dust source list.

[0086] It's important to note that the rapid dust source monitoring method based on the Environmental Satellite-2 is a comprehensive monitoring approach. It provides a rapid detection and identification technology for dust sources, reflecting their spatial distribution, pollution impact, and primary source types. Therefore, this method overcomes the shortcomings of traditional source tracking monitoring, which suffers from incomplete coverage, slow monitoring timelines, high investment costs, and inaccurate management support. This makes dust source monitoring based on Environmental Satellite-2 remote sensing more accurate and comprehensive.

[0087] The above are merely preferred embodiments of the present invention and are intended to help understand the method and core concept of this application. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the concept of the present invention fall within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

[0088] The present invention solves the problem of rough dust source identification based on manual inspection and high-resolution satellite visual interpretation in the existing technology as a whole, without considering the impact of dust sources on atmospheric pollution and source classification. It constructs a remote sensing rapid monitoring technology method for dust sources through machine learning calculation of surface reflectivity, aerosol optical thickness inversion and dust source identification and classification to comprehensively and accurately reflect the changes in dust source pollution. It can not only quickly identify dust sources at high frequency, but also finely classify dust sources, and efficiently guide the precise control of dust sources, thereby making up for the limitations of traditional prediction methods such as insufficient monitoring coverage, slow monitoring timeliness, high investment cost, and inaccurate control support, which lead to the inability to accurately control dust sources.

Claims

1. A remote sensing monitoring method for atmospheric dust sources based on the Environmental Satellite No. 2, characterized in that: The method comprises the steps of: S1. Based on the CCD image data of the Environmental Satellite-2 and combined with the aerosol products of MODIS, a random forest method was used to establish a remote sensing estimation model for the surface reflectance of the Environmental Satellite-2. S2. Based on the CCD images of the Huanhuan-2 satellite, the atmospheric radiation transfer model is used to invert the aerosol optical depth; S3. Extract the aerosol spatial maximum grid cells based on the aerosol optical depth and calculate the grid reliability; S4. Use images to screen out dust sources based on the underlying surface of the aerosol optical depth (AOD) and interpret and classify the AOD.

2. The method for remote sensing monitoring of atmospheric dust sources based on the Environmental Satellite No. 2 according to claim 1, characterized in that: Said S1 comprises: S1.1 Collect quarterly historical CCD imagery from the Huanhuan-2 satellite for urban built-up areas, extract 550nm aerosol optical depth product data from the MODIS aerosol product for the corresponding dates, and reproject the data using the Huanhuan-2 satellite projection method. S1.2 Extract calibration coefficients and satellite observation angles from the auxiliary XML file of the Huanhuan-2 CCD image, calculate the apparent reflectance of the blue band, green band, red band, near-infrared band, and red-edge band of the urban built-up area CCD image, and extract the aerosol optical depth parameters at each pixel of the CCD image from the MODIS aerosol optical depth product; S1.3 Use the 6SV atmospheric radiation transfer model to perform atmospheric correction on the red and blue bands of the CCD image based on the observation geometry, observation date, band spectral response function apparent reflectance, and aerosol optical depth to obtain the surface reflectance of the two bands of the CCD image; S1.4 uses a stratified sampling method to randomly obtain the solar zenith angle, satellite observation zenith angle, relative azimuth, blue band, green band, red band, near-infrared band, and red edge band apparent reflectance of 1000 pixels in urban built-up areas in each season. At the same time, the surface reflectance of the blue band and red band of the corresponding pixels is extracted to form a training sample dataset; S1.5 is based on the training samples established in S1.

4. The solar zenith angle, satellite observation zenith angle, relative azimuth, blue band, green band, red band, near-infrared band and red edge band apparent reflectance are independent variables, and the surface reflectance is the dependent variable. The random forest method is used for fitting training to establish the surface reflectance inversion model of the red and blue bands of the CCD camera of the Environmental-2 satellite.

3. The method for remote sensing monitoring of atmospheric dust sources based on the Environmental Satellite No. 2 according to claim 2, characterized in that: The S2 includes: S2.1 Extract calibration coefficients from auxiliary files based on the real-time environmental No. 2 CCD camera image, calculate the apparent reflectance of the blue band, green band, red band, near-infrared band, and red edge band based on the calibration coefficients, and define the area with the red band apparent reflectance greater than 0.2 as the cloud cover area; S2.2 Read the solar zenith angle, satellite zenith angle, and relative azimuth from the auxiliary file of the Real-Time Environment II CCD camera image, match the apparent reflectance data in step S2.1 to form an input parameter dataset, and obtain the red and blue band surface reflectances of the cloud-free land pixels of the Real-Time Environment II CCD camera based on the surface reflectance inversion model of the Real-Time Environment II CCD camera established in step S1.5; S2.3, based on the 6SV atmospheric radiation transfer model and the red and blue band surface reflectances in S2.2, set the AODs to 0, 0.3, 0.6, 0.9, 1.2, 1.5, 1.8, 2.1, 2.4, 2.7, and 3.0, respectively. The aerosol model is set to dust aerosol. The red and blue band apparent reflectances at different AODs under the observation conditions of solar zenith angle, satellite zenith angle, and relative azimuth angle are simulated by reading the auxiliary file of the real-time environmental No. 2 CCD camera image. S2.4 Calculate the simulated apparent reflectance errors for the red and blue bands based on the simulated apparent reflectance at different AODs in S2.

3. The calculation method is as follows: in, represents the apparent reflectance simulation error, is the apparent reflectance of the blue band observed by satellite, To simulate and calculate the apparent reflectance of the blue band, is the apparent reflectance of the red band observed by satellite, Calculate the apparent reflectance of the red band for simulation; S2.5 uses the least squares method to fit the relationship between the apparent reflectance error and AOD for each pixel using a quadratic polynomial based on the apparent reflectance error at different AODs in S2.

4. The AOD at which the simulated apparent reflectance error is minimized is calculated as the inverted AOD for that pixel, thereby obtaining the AOD of the cloud-free land pixels in the entire image.

4. The method for remote sensing monitoring of atmospheric dust sources based on the Environmental Satellite No. 2 according to claim 3, characterized in that: The S3 includes: S3.1: For the built-up area of ​​the target region, a buffer zone with a radius of 3 km is established. The AOD of the cloud-free land pixels obtained by inversion in S2.5 is clipped using the buffer zone to obtain the cloud-free AOD of the built-up area and surrounding areas. S3.2 Based on the cloud-free AOD of the built-up area and surrounding areas obtained by inversion in S2.5, select pixels with AOD greater than 1 as high-AOD pixels; S3.3 Based on the potential high AOD pixels extracted in step S3.2, the red band surface reflectance obtained in step S2.2 and the AOD of the cloud-free land pixels obtained in step S2.5, for the high AOD pixels, by comparing the pixel AOD, red band surface reflectance, and vegetation index with the background pixels, determines whether they are potential dust source pixels. If the following equations (1) to (3) are satisfied, the pixel is determined to be a potential dust source pixel; otherwise, it is a non-dust source pixel: Wherein, AOD is the AOD value of the potential dust source pixel, dimensionless; is the average AOD of pixels within 1 km around the potential dust source pixel, dimensionless; is the mean absolute deviation of AOD within 1 km around the potential dust source pixel, dimensionless; is the red band surface reflectance of the potential dust source pixel retrieved by satellite, dimensionless; is the average surface reflectance in the red band within 1 km of the potential dust source pixel, dimensionless; is the average deviation of the red band surface reflectance within 1 km of the potential dust source pixel, dimensionless; NDVI is the vegetation cover index of the potential dust source pixel, calculated by the red band surface reflectance and the near-infrared surface reflectance, dimensionless; is the average value of vegetation coverage index within 1 km of the potential dust source pixel, is the average deviation of vegetation cover index within 1 km of the pixel; S3.4 Based on all potential dust source pixels extracted in step S3.3, use the ramp function to calculate the dust source credibility of the pixel. The calculation is shown as follows: in, C1=S(AOD,1,5) (5) C3=S(0.3-NDVI,0,0.2) (7) S3.5 Extract the pixel center coordinates of potential dust source pixels with a credibility greater than 60, and create a potential dust source vector grid based on the pixel size.

5. The method for remote sensing monitoring of atmospheric dust sources based on the Environmental Satellite No. 2 according to claim 4, characterized in that: The S4 includes: S4.1 Superimpose all potential dust source vector grids extracted in step S3.4 on GF-1, GF-2, and GF-6 satellite images with a resolution better than 2 meters, and select satellite images with a time interval of no more than six months between GF-2 images; S4.2 Visually interpret the underlying surface imagery of potential dust source grids, and identify potential source grids where cultivated land, residential areas, or industrial plant buildings account for more than 80% of the grid, or where no bare land is found, as false grids; identify all other potential dust source grids as true dust source grids; S4.3 Interpret the underlying surface image of the real dust source grid, and conduct a comprehensive analysis and classification of the sources of the real dust source grid based on the underlying surface. The classification method is as follows: if bare land and industrial plants are found on the underlying surface, it is judged to be an industrial construction dust source; if bare land and sheds and houses are found on the underlying surface, it is judged to be a construction dust source; if bare land and mountains are found on the underlying surface, it is judged to be a mining dust source; if bare land and multiple traffic arterial road signs are found on the underlying surface, it is judged to be a transportation dust source; S4.4 performs erosion and expansion processing on the dust source grid, eliminates discrete independent grids, summarizes the coordinates, credibility and dust source type information, and forms a dust source list.

Images