Full-coverage daily scale AOD inversion method, device and system and storage medium

Abstract

The invention discloses a full-coverage daily scale AOD inversion method, device and system and a storage medium, and the method comprises the steps: processing site data, a meteorological data set, a topographic data set and population data through employing a linear interpolation method, and carrying out the space-time matching with MAICAOD, and obtaining a modeling data set; triangular irregular network interpolation is carried out on the MAICAOD, and initial daily AOD data and annual average AOD data are obtained; based on the preliminary daily AOD data and the annual average AOD data, constructing a multilayer LGBM model, and performing multi-step simulation to correct a filling error so as to obtain an adjustment optimal prediction model; and based on the modeling data set, performing estimation by using the optimal prediction model to obtain a full-coverage daily AOD product in the region. The daily AOD product obtained by the method is basically consistent with the actually measured data of the monitoring station, meanwhile, the product with 100% full coverage can be obtained, and the method has good time-space continuity. And meanwhile, the recovered AOD product is consistent with the original data, so that the accuracy and the reliability are relatively high.

Description

Technical Field

[0001] The present invention belongs to the technical field of aerosol optical thickness inversion, and in particular, provides a full coverage daily scale AOD inversion method, device, system and storage medium. Background Art

[0002] Atmospheric aerosols are solid or liquid particles suspended in the air. They have a significant impact on human health, air quality, ecosystems, and climate change by changing the Earth's atmospheric circulation and radiation balance. Therefore, the spatiotemporal distribution of aerosols has become the focus of current research. As one of the important optical characteristics of aerosols, aerosol optical depth (AOD) is defined as the integral of the aerosol extinction coefficient in the vertical direction. It can characterize the aerosol load level and distribution. It is a key indicator for quantifying aerosol content and has been widely used in atmospheric environmental research.

[0003] There are two main methods for measuring AOD data: ground measurement and satellite measurement. Ground measurement mainly uses the Aerosol Robotic Network (AERONET) to provide relatively accurate AOD values ​​and regional analysis using a solar spectrophotometer. The data obtained lacks continuity and coverage over a large range of space. Therefore, AOD products based on satellite inversion are increasingly used as the main data source for AOD research and are considered to be the best data source for research on a global scale.

[0004] Common AOD products come from various sensors, such as the Advanced Very High Resolution Radiometer (AVHRR), the Moderate Resolution Imaging Spectroradiometer (MODIS), the Ozone Monitoring Instrument (OMI), and so on. Among them, the global high-resolution (36 spectral channels, 1-2 days of repetition period, 250m, 500m, 1000m spatial resolution) daily AOD inversion product provided by MODIS using the Advanced Multi-angle Atmospheric Correction (MAIAC) algorithm can better identify the AOD information in cloud and snow areas, and is widely used in research and applications. However, due to the limited detection capabilities of sensors in areas with high altitudes, mountainous areas, or high surface reflectivity and high aerosol concentrations, the original MAICAAOD product has a large amount of missing data, and its average coverage is only 30.42%. Since the MAIAC AOD product is only available under clear skies, frequent rainy or cloudy weather will also cause poor spatial continuity of the product, limiting its application at the regional scale, especially in specific areas, and the accuracy of subsequent prediction and analysis of atmospheric pollutants. At the same time, sunny days can also cause data loss due to unfavorable surface conditions (snow accumulation) or errors in data preprocessing (misclassification of heavy aerosol layers as clouds). Therefore, restoring the gaps in the MAICA AOD data to improve the quality and availability of AOD product data, and to enhance the spatial continuity of AOD and expand its application range has important significance and research prospects.

[0005] So far, many methods have been proposed to restore AOD data products. The earliest and most widely used method is based on spatial statistical methods such as Kriging, which uses the local spatial information of the AOD product itself to restore the blanks of AOD. However, the AOD restored by this method has high spatiotemporal variability and is not suitable for predicting missing data in AOD. Another common method is to improve data quality by fusing multi-source AOD data products, but this method often cannot unify the uncertainty between different AOD products, and cannot solve the impact of cloud pollution well. . In view of the continuous advancement and gradual improvement of machine learning technology, many scholars have begun to train models to restore missing AOD values. From the recovery results, it can be found that the algorithm developed based on machine learning technology can not only obtain spatially continuous AOD products but also achieve high accuracy, so it is increasingly used to restore AOD products.

[0006] Generally speaking, the changes in aerosol concentration in a short period of time are drastic and rapid, so it is most suitable for practical applications to consider combining machine learning and the continuous time information of the original AOD product itself to develop an algorithm to restore the single-day AOD product. However, due to the limitation of the small amount of information in a single-day image, the currently developed algorithm has defects in generating full-coverage AOD products in large areas, so linear functions are generally used in large-scale applications. Therefore, it is necessary to propose a new algorithm that fully utilizes the single-day AOD information to restore the full-coverage, high-precision AOD over a large area, so as to break the limitation of scarce complementary information. Summary of the invention

[0007] The present invention is provided to solve the above problems existing in the prior art. Therefore, a full coverage daily scale AOD inversion method, device, system and storage medium are needed.

[0008] According to the first technical solution of the present invention, a full coverage daily scale AOD inversion method is provided, the method comprising:

[0009] The station data, the meteorological data set, the terrain data set and the population data are processed by a linear interpolation method to obtain preprocessed data, and the preprocessed data are spatially and temporally matched with MAICAAOD to obtain a modeling data set;

[0010] The triangular irregular network interpolation of MAICAAOD was performed to obtain preliminary daily AOD data and annual average AOD data;

[0011] Based on the preliminary daily full coverage AOD data and annual average AOD data, a multi-layer LGBM model is constructed, and multi-step simulation is performed to correct the filling error and obtain the adjusted optimal prediction model;

[0012] Based on the modeling data set, the optimal prediction model is used to perform estimation to obtain a daily AOD product that fully covers the area.

[0013] Furthermore, the site data, meteorological data set, terrain data set and population data are processed by a linear interpolation method to obtain preprocessed data, and the preprocessed data is spatially and temporally matched with MAICAAOD to obtain a modeling data set, including:

[0014] Get site data using The index interpolates the two adjacent bands of 670nm and 440nm to obtain the AOD parameters at 550nm, so that the site data can meet the consistency with MAIAC AOD when comparing the accuracy;

[0015] Obtain NDVI data, DEM data, and population datasets;

[0016] The site data, NDVI data, DEM data, and population data were matched with the MAICAAOD data accuracy as the modeling dataset.

[0017] Furthermore, the temporal and spatial resolutions of the NDVI data are 16 days and 250M respectively, the resolution of the DEM data is 30m, and the resolution of the population data set is 1km.

[0018] Furthermore, triangular irregular network interpolation is performed on MAICAAOD to obtain full coverage AOD data and annual average AOD data, including:

[0019] Get MAICAAOD;

[0020] According to the quality assurance mark and valid AOD range provided by the MAIAC AOD product, the MAIACAOD data was cleaned and AOD values ​​greater than 3 were excluded;

[0021] The existing daily MAICAAOD data are used to perform TIN interpolation on the missing areas in MAIACAOD so as to make use of the existing spatial information for large-scale restoration and thus obtain the daily AOD with sufficient data volume.

[0022] Furthermore, the temporal and spatial resolutions of the MAICAAOD are 1 d and 1 km×1 km respectively.

[0023] Furthermore, based on the full coverage AOD data and the annual average AOD data, a multi-layer LGBM model was constructed, and multi-step simulations were performed to correct the filling errors and obtain the adjusted optimal prediction model, including:

[0024] A multi-layer LGBM model was constructed; in the multi-layer LGBM model, the first layer used the initially obtained full coverage AOD products and annual average AOD as the main explanatory variables to fill in the seasonal average value, the second layer used annual and seasonal AOD products as explanatory variables to calculate the monthly average AOD value, the third layer used annual, seasonal and monthly AOD products to fill in weekly AOD, and the last layer used annual, seasonal, monthly and weekly AOD products as explanatory variables to fill in daily AOD on a large scale; the calculation formula of the last layer is expressed as:

[0025]

[0026] Among them, AOD Daily (i, j) represents the actual daily AOD level of the specified pixel (i, j); LGBM represents the last layer of the LGBM model; AOD Year (i, j) represents the preliminary annual average AOD; AOD Quarterly (i, j) represents the seasonal average AOD obtained by the first-layer model; AOD Monthly(i, j) represents the monthly average AOD obtained by the second-layer model; AOD Weekly (i, j) represents the weekly AOD obtained by the third-layer model; NDVI (i, j) represents the normalized vegetation index value of the specified pixel point (i, j); DEM (i, j) represents the elevation value of the specified pixel point (i, j); POP (i, j) represents the population data of the specified pixel point (i, j).

[0027] Furthermore, based on the modeling data set, the optimal prediction model is used to perform estimation to obtain a daily AOD product covering the entire region, including:

[0028] All modeling data are processed into standard spatial grids, and the optimal prediction model is used to restore the daily AOD values ​​in the region to obtain a daily AOD product that covers the entire region.

[0029] According to the second technical solution of the present invention, a full coverage daily scale AOD inversion device is provided, the device comprising:

[0030] The spatiotemporal matching module is configured to process the site data, the meteorological data set, the terrain data set and the population data using a linear interpolation method to obtain preprocessed data, and perform spatiotemporal matching between the preprocessed data and MAICAAOD to obtain a modeling data set;

[0031] The data construction module is configured to perform triangular irregular network interpolation on MAICAAOD to obtain full coverage AOD data and annual average AOD data;

[0032] The model inversion module is configured to construct a multi-layer LGBM model based on the full coverage AOD data and the annual average AOD data, and perform multi-step simulation to correct the filling error to obtain the adjusted optimal prediction model; based on the modeling data set, the optimal prediction model is used to perform estimation to obtain the daily AOD product with full coverage in the area.

[0033] According to the third technical solution of the present invention, a full coverage daily scale AOD inversion system is provided, the system comprising:

[0034] Memory for storing computer programs;

[0035] A processor is used to execute the computer program to implement the method as described above.

[0036] According to a fourth technical solution of the present invention, a non-transitory computer-readable storage medium storing instructions is provided. When the instructions are executed by a processor, the method described above is executed.

[0037] The full coverage daily scale AOD inversion method, device, system and storage medium according to each solution of the present invention have at least the following technical effects:

[0038] The present invention can obtain regional full coverage daily-scale AOD products. After verification, the matching rate between the restored results and the AERONET measured AOD reached 96.8%, R=0.86, and RMSE=0.33. After restoration, MAICAAOD increased from an average daily coverage of about 40% to 100% full coverage, and has good spatiotemporal continuity. In addition, the spatial distribution characteristics of the annual and monthly averages of the restored AOD products are consistent with the original MAICAAOD. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same reference numerals with letter suffixes or different letter suffixes may represent different instances of similar parts. The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the specification and claims, are used to illustrate the embodiments of the invention. When appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0040] Figure 1 A flowchart of a full coverage daily-scale AOD inversion method according to an embodiment of the present invention is shown;

[0041] Figure 2 A diagram showing the accuracy comparison results between the multi-layer LGBM model according to an embodiment of the present invention and the traditional model;

[0042] Figure 3 A monthly accuracy verification diagram of a multi-layer LGBM model according to an embodiment of the present invention is shown;

[0043] Figure 4 It shows the time series distribution diagram of AOD products restored by the model according to the embodiment of the present invention in the target area and typical areas;

[0044] Figure 5 A structural diagram of a full-coverage daily-scale AOD inversion device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present invention. For the various steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that they can be adjusted in order, as long as the logic between them is not destroyed, resulting in the inability to implement the entire process.

[0046] The embodiment of the present invention provides a full coverage daily scale AOD inversion method, such as Figure 1 As shown, the method includes the following steps S10 to S40, which are described in detail as follows.

[0047] S1. Use linear interpolation method to process site data, meteorological data set, terrain data set and population data to obtain pre-processed data, and perform spatiotemporal matching between the pre-processed data and MAICAAOD to obtain a modeling data set.

[0048] In this embodiment, the linear interpolation method is used to process the site data, meteorological data set, terrain data set and population data, and these data are matched with MAICAAOD in time and space. Specifically, the present invention uses MAICAAOD data, site data, NDVI, DEM and population data.

[0049] The site data used is the 1.5 and 2.0 level data of the AERONET site version 3. AERONET uses a sun photometer to provide AOD spectral measurements with high time resolution (15min) in the 0.34-1.06μm band. The network has three quality levels of data sets: level 1.0 (unscreened), level 1.5 (cloud screened), and level 2.0 (cloud screened and quality assured). Among them, since the uncertainty of level 1.5 and level 2.0 in version 3 is lower, the level 1.5 and level 2.0 data are used as ground truth for accuracy verification. Since AERONET does not measure AOD at 550nm, it is used The index interpolates the two adjacent bands of 670nm and 440nm to obtain the AOD parameters at 550nm, so that it meets the consistency with MAIACAOD when comparing the accuracy.

[0050] The global Normalized Difference Vegetation Index (NDVI) with a spatial resolution of 500 meters provided every 16 days by the MODIS MOD13A1 product; the 30-meter digital elevation model (DEM) data provided by the Shuttle Radar Topography Mission (SRTM); and the 1-km LandScan population dataset developed by the Oak Ridge National Laboratory (ORNL) and provided by East View Cartographic (https: / / landscan.ornl.gov / ).

[0051] The MAICAAOD data provided by NASA were used, with a temporal and spatial resolution of 1 d and 1 km × 1 km, respectively. According to the quality assurance (QA) mark and valid AOD range provided by the MAICAAOD product, the data were cleaned (QAcloud_mask = clear) and AOD values ​​greater than 3 were excluded.

[0052] The site data, NDVI, DEM and population data were matched with the MAICAAOD data accuracy and used as the modeling dataset.

[0053] S2. Perform triangular irregular network interpolation on MAICAAOD to obtain preliminary full-coverage AOD data and annual average AOD data.

[0054] In this embodiment, the MAICAAOD data provided by NASA is obtained, and its temporal and spatial resolutions are 1d and 1km×1km respectively. According to the quality assurance (QA) mark and effective AOD range provided by the MAICAAOD product, the data is cleaned (QAcloud_mask=clear), and AOD values ​​greater than 3 are excluded.

[0055] The existing daily MAICAAOD data is used to perform TIN interpolation on the missing areas to use the existing spatial information for large-scale restoration, thereby obtaining daily AOD with sufficient data volume. With the help of TIN interpolation, the spatial information contained in the daily AOD is retained, and the annual average AOD that reflects the interannual information of AOD is obtained more accurately, providing good spatiotemporal information for the establishment of the model.

[0056] S3. Based on the full coverage AOD data and the annual average AOD data, a multi-layer LGBM model is constructed, and multi-step simulation is performed to correct the filling error and obtain the adjusted optimal prediction model.

[0057] In the present embodiment, the processed variable data are composed into a modeling data set, and the matching data set is modeled using a multi-layer LGBM model. Specifically, the model selected by the present invention is a multi-layer machine learning model. It is believed that the AOD levels under different scales reflect the different spatiotemporal variation characteristics of AOD. The AOD under the annual and seasonal scales reflects the overall variation trend of AOD over a whole year: the year reflects the overall trend of AOD over a year, and the season characterizes the fluctuations that may occur in AOD over this year. The AOD under the monthly and weekly scales reflects the overall trend of AOD in a local time period: the month reflects the overall variation of AOD in this time period, and the week reflects the random variation of AOD in this time period. The final daily AOD represents the dynamic variation state of AOD every day. They are coupled and affect each other. Therefore, the accuracy of each scale result and the completeness of spatiotemporal information extraction within the scale have a great influence on the accuracy of AOD recovery. Therefore, based on this mechanism, the LGBM model is used to perform multi-step simulation to recover AOD on a large scale.

[0058] Due to the complex relationship between the temporal and spatial information of AOD, it is difficult to describe it with a single model. Therefore, based on the above mechanism, a multi-layer LGBM model was constructed: the first layer uses the initially obtained full coverage AOD products and annual average AOD as the main explanatory variables to fill in the seasonal average value, the second layer uses annual and seasonal AOD products as the main explanatory variables to calculate the monthly average AOD value, the third layer uses annual, seasonal, and monthly AOD products to fill in weekly AOD, and finally uses annual, seasonal, monthly, and weekly as the main explanatory variables to fill in daily AOD on a large scale:

[0059]

[0060] Among them, AOD Daily (i, j) refers to the actual daily AOD level of the specified pixel (i, j); a series of explanatory variables include the multi-scale AOD spatiotemporal information and related auxiliary factors simulated by the previous layer LGBM. LGBM represents the last layer LGBM model; AOD Year (i, j) represents the preliminary annual average AOD; AOD Quarterly (i, j) represents the seasonal average AOD obtained by the first-layer model; AOD Monthly (i, j) represents the monthly average AOD obtained by the second-layer model; AOD Weekly (i, j) represents the weekly AOD obtained by the third-layer model; NDVI (i, j) represents the normalized vegetation index value of the specified pixel point (i, j); DEM (i, j) represents the elevation value of the specified pixel point (i, j); POP (i, j) represents the population data of the specified pixel point (i, j).

[0061] Exemplarily, this embodiment evaluates the performance of the proposed model in two ways. First, the original MAICAAOD, the AOD after TIN filling, and the AOD restored by the MLL-MST model are evaluated with the ground AOD measured at the AERONET site. Secondly, the data restored by the model in the blank area of ​​the original MAICAAOD are selected for evaluation with the AERONET site measurement data adjacent to the area. Thereafter, four statistical indicators including the correlation coefficient (R), the mean relative error (MRE), the root mean square error (RMSE), and the expected error (EE) envelope are used to evaluate the accuracy of the proposed method. Among them, according to the definition of the EE envelope, the proportion within the EE envelope is also calculated to evaluate the percentage of qualified data.

[0062]

[0063] EE=±(0.05+0.15×AOD)

[0064] AOD-EE≤AOD Sat ≤AOD+EE

[0065] Among them, AOD Aero is the AOD measured at the AERONET site. Sat The daily AOD products representing the original MAICAAOD, TIN-filled AOD, and model-restored AOD, respectively; and is the corresponding mean value; N is the number of matches between the restored day AOD product and the ground-measured AOD.

[0066] from Figure 2 It can be seen that the multi-layer LGBM model has the best agreement with the AERONET measurement results, with an R of 0.90. Compared with the verification results of the original AOD, the performance of the full coverage AOD results has slightly decreased, with MRE and RMSE increased by 0.004 and 0.003 respectively. However, it can be seen that after the AOD is restored by the multi-layer LGBM model, the number of matching points has increased to 8200, and the matching rate has reached 96.8%. It is worth noting that the ratio of within EE in the multi-layer LGBM model has increased from the original 55% to 60%, and the ratios of above and below EE have both decreased.

[0067] Based on the model validation, this example counted the data distribution of the original MAICAAOD, TIN MAICAAOD and multi-layer LGBM model AOD products in the study area and calculated the monthly averages of the three products to verify the accuracy of the overall recovery results ( Figure 3). Overall, the performance trends of the three products are generally consistent, that is, they all show an upward trend from January to April, and a gradually downward trend from April to December. By comparing the lengths of the boxes, it can be found that the multi-layer LGBM model product can better restore the fluctuation degree of AOD than the TIN model product, which improves the credibility of the product data. At the same time, by comparing the position differences of the horizontal lines in the boxes, it is found that the average level trend of MAICAAOD restored by the multi-layer LGBM model is more consistent with the original MAICAAOD. Moreover, the monthly average value shows the same phenomenon, that is, although the multi-layer LGBM model and the TIN model both show an overestimation phenomenon, this is also related to the filling of a large number of missing values. However, the multi-layer LGBM model presents a mean level that is more in line with the actual site than the TIN model, and does not show an excessively high overestimation level. Moreover, its change trend is more similar to the actual situation than TINAOD.

[0068] In the long-term changes, the ability to fit the overall trend and the ability to capture local features are equally important in measuring the overall expressiveness of the dataset. The measured data of the three AERONET stations in the target area were used to compare with the original MAICAAOD, TIN model AOD, and multi-layer LGBM model AOD products, and the AOD time series distribution map for 2023 was drawn ( Figure 4 ) to further evaluate the accuracy of the multi-layer LGBM model results. Overall, the changing trend of the AOD level of the multi-layer LGBM model is not much different from the changing trend of the measured data at the site, and it can basically simulate the real annual AOD level trend. At the same time, by observing the fitting of the multi-layer LGBM model when the AOD is at a low level, it is found that the multi-layer LGBM model can sensitively capture local small fluctuations, which is approximately the same as the actual situation of AERONET, indicating the reliability of the model recovery results. However, for different sites, due to the influence of factors such as the quality of the original MAICAAOD data and the different AOD levels in the region, there are certain differences between the restored AOD results. But on the whole, the AOD product restored by the multi-layer LGBM model shows better accuracy and reliability than the AOD restored by TIN, and has higher precision to support scientific research.

[0069] From the perspective of time, the average time coverage of the original MAICAAOD product in the target area in 2023 is 46.9%, which is about 171 days. Among them, the time coverage rate in most areas in the north is relatively high, basically reaching more than half a year, while the time coverage in the southern region is relatively low, and even some areas have zero coverage. This phenomenon has certain limitations on the application of the product in the south, so it is necessary to update the MAICA AOD product to ensure the universality and reliability of the product. From the improvement effect, it can be seen that the results obtained by the TIN and multi-layer LGBM models have greatly improved the time coverage, and for the zero coverage area in the south, both methods can achieve full coverage of the zero coverage area, which reflects that both models have good performance in restoring the time characteristics of the data. However, since AOD itself contains rich spatiotemporal information, only measuring the time completion cannot comprehensively evaluate the AOD product. Therefore, the spatial coverage and improvement effect are calculated. It can be seen that the spatial coverage of the original MAICAAOD product in the target area in 2023 is basically around 43.9%, the highest can reach about 58.7%, and the lowest is only 24.6%. Compared with the original product, in 2023, the full-coverage AOD product restored by the multi-layer LGBM model improved by an average of 127.5%, providing good data support for the application analysis of AOD on the daily scale.

[0070] S4. Based on the modeling data set, the optimal prediction model is used to perform estimation to obtain a daily AOD product that fully covers the area.

[0071] In this embodiment, all modeling data are processed by standard spatial gridding, and the established multi-layer LGBM model is used to restore the daily AOD values ​​in the region to obtain a daily AOD product with full coverage in the region. Specifically, in order to better show the recovery results, the multi-layer LGBM model built by the present invention is used to restore the AOD of the target area. According to the comparison between the restored daily AOD product and the traditional model, it can be seen that this is a day with low coverage of the original MAICAAOD product, only 26.8%. The areas covered are mostly concentrated in the western region, and only very few cities have achieved full coverage, which has very large restrictions on the AOD research of hot spots and urban scales. For the restored daily AOD products, both models have reached the level of full coverage, and the changing trends of the areas covered by the original MAICAAOD products are generally consistent. Although the overall levels of the two models are not much different (the overall mean only differs by 0.01), the TIN model lacks more detailed expressions compared to the multi-layer LGBM model. From the perspective of local recovery, although the magnitude difference of the characteristic values ​​(maximum and mean) is not large, TIN smoothes too much detailed information, resulting in a more obvious overestimation phenomenon. The information on adjacent longitudes and latitudes within the region is very similar, which does not conform to the actual situation of AOD. The AOD product restored by the multi-layer LGBM model retains the spatiotemporal characteristics provided by the original MAICAAOD product, and with the help of AOD-related auxiliary variables, it restores more random characteristics of the AOD of the day.

[0072] In order to further observe the reliability of the spatial and temporal distribution of the restored AOD product in the target area, the restored AOD was compared with the annual average distribution of the original MAICAAOD in the target area in 2023. At the same time, the annual average values ​​of the two products in the vicinity of the AERONET station in the target area were calculated to analyze the consistency between the product and the station data.

[0073] According to the above comparison, we can conclude that: (1) The distribution effect of the AOD product restored by the multi-layer LGBM model is very consistent with the original MAICAAOD product. Both products show that the distribution in the east is higher than that in the west. And like the original MAICAAOD, the AOD product restored by the multi-layer LGBM model has a certain level of gradient: it decreases from the southeast to the northwest, and at the same time, it shows the boundary between high / low value areas in some areas. This shows that the change trend of the original AOD has not been covered up too much. The MLL-MST model can better retain the original spatiotemporal information while filling the original gaps, reflecting more realistic aerosol spatiotemporal distribution characteristics. (2) The phenomenon of a large number of missing AOD pixels is very common in the original MAICAAOD. Averaging the original MAICA AOD in the absence of sufficient AOD pixels will result in a lower annual average value of the original MAICAAOD (the average AOD value is 0.20). After filling the gaps, the mean of the AOD product restored by the multi-layer LGBM model only rose to 0.25. The deviation of 0.05 proves that the deviation of the multi-layer LGBM model is not large, which can better avoid overestimation and restore the missing AOD data more accurately. (3) Observing the annual mean effect near the AERONET station, both products show a high degree of consistency. Taking the ground monitoring data provided by the AERONET station as a reference, the AOD product restored by the multi-layer LGBM model shows a certain stability and accuracy in different geographical locations and meteorological conditions, which once again shows the reliability of the restored product in estimating the true AOD.

[0074] Although the AOD product restored by the multi-layer LGBM model performs very well in terms of overall trend, AOD, as one of the important optical properties of aerosols, is highly dynamic and random like aerosols, and exhibits significant variability in time and space. Therefore, it is not reliable to only evaluate the accuracy of the overall annual trend of the AOD product restored by the multi-layer LGBM model. Therefore, the monthly periodic changes before and after restoration were compared to verify the performance of the multi-layer LGBM model in characterizing the dynamic randomness of AOD.

[0075] Overall, the missing values ​​of the original MAICAAOD mainly appear in spring and winter, while both products can restore the integrity of AOD well. The original MAICAAOD generally shows a trend of spring>winter>summer>autumn. The AOD product restored by the multi-layer LGBM model is generally consistent with the original MAICAAOD. Although there is a certain positive deviation, the deviation is not particularly large. In detail, the multi-layer LGBM model can restore more local random features. The restoration results of MLL-MST not only retain the distribution of the original data, which is roughly consistent with the change trend of the original MAICAAOD, but also restore the AOD distribution with more dynamic levels and random gradients. At the same time, comparing the statistics between the two products, the minimum deviation between the eigenvalues ​​of the multi-layer LGBM model and the original MAICAAOD results can reach 0.02, and the maximum is only 0.07, which further verifies the reliability of the multi-layer LGBM model in restoring the dynamic random characteristics of the AOD product.

[0076] The embodiment of the present invention also provides a full coverage daily scale AOD inversion device, such as Figure 5 As shown, the device comprises:

[0077] The spatiotemporal matching module is configured to process the site data, the meteorological data set, the terrain data set and the population data using a linear interpolation method to obtain preprocessed data, and perform spatiotemporal matching between the preprocessed data and MAICAAOD to obtain a modeling data set;

[0078] The data construction module is configured to perform triangular irregular network interpolation on MAICAAOD to obtain full coverage AOD data and annual average AOD data;

[0079] The model inversion module is configured to construct a multi-layer LGBM model based on the full coverage AOD data and the annual average AOD data, and perform multi-step simulation to correct the filling error to obtain the adjusted optimal prediction model; based on the modeling data set, the optimal prediction model is used to perform estimation to obtain the daily AOD product with full coverage in the area.

[0080] In some embodiments, the spatiotemporal matching module is further configured to:

[0081] Get site data using The index interpolates the two adjacent bands of 670nm and 440nm to obtain the AOD parameters at 550nm, so that the site data can meet the consistency with MAIAC AOD when comparing the accuracy;

[0082] Obtain NDVI data, DEM data, and population datasets;

[0083] The site data, NDVI data, DEM data, and population data were matched with the MAICAAOD data accuracy as the modeling dataset.

[0084] In some embodiments, the temporal and spatial resolutions of the NDVI data are 16 days and 250M respectively, the resolution of the DEM data is 30m, and the resolution of the population dataset is 1km.

[0085] In some embodiments, the data construction module is further configured to:

[0086] Get MAICAAOD;

[0087] According to the quality assurance mark and valid AOD range provided by the MAIAC AOD product, the MAIACAOD data was cleaned and AOD values ​​greater than 3 were excluded;

[0088] The existing daily MAICAAOD data are used to perform TIN interpolation on the missing areas in the MAIAC AOD so as to make use of the existing spatial information for large-scale restoration and thus obtain the daily AOD with sufficient data volume.

[0089] In some embodiments, the temporal and spatial resolutions of the MAICAAOD are 1d and 1km×1km, respectively.

[0090] In some embodiments, the model inversion module is further configured to:

[0091] A multi-layer LGBM model was constructed; in the multi-layer LGBM model, the first layer used the initially obtained full coverage AOD products and annual average AOD as the main explanatory variables to fill in the seasonal average value, the second layer used annual and seasonal AOD products as explanatory variables to calculate the monthly average AOD value, the third layer used annual, seasonal and monthly AOD products to fill in weekly AOD, and the last layer used annual, seasonal, monthly and weekly AOD products as explanatory variables to fill in daily AOD on a large scale; the calculation formula of the last layer is expressed as:

[0092]

[0093] Among them, AOD Daily (i, j) represents the actual daily AOD level of the specified pixel (i, j); LGBM represents the last layer of the LGBM model; AOD Year (i, j) represents the preliminary annual average AOD; AOD Quarterly (i, j) represents the seasonal average AOD obtained by the first-layer model; AOD Monthly (i, j) represents the monthly average AOD obtained by the second-layer model; AOD Weekly(i, j) represents the weekly AOD obtained by the third-layer model; NDVI (i, j) represents the normalized vegetation index value of the specified pixel point (i, j); DEM (i, j) represents the elevation value of the specified pixel point (i, j); POP (i, j) represents the population data of the specified pixel point (i, j).

[0094] In some embodiments, the model inversion module is further configured to:

[0095] All modeling data are processed into standard spatial grids, and the optimal prediction model is used to restore the daily AOD values ​​in the region to obtain a daily AOD product that covers the entire region.

[0096] It should be noted that the modules involved in the embodiments of the present invention may be implemented by software or hardware, and the modules described may also be set in a processor. The names of these modules do not, in some cases, limit the modules themselves.

[0097] The aerosol optical thickness inversion device mentioned in the embodiment of the present invention belongs to the same technical concept as the method previously described, and the technical effects they achieve are basically the same, which will not be repeated here.

[0098] The embodiment of the present invention further provides an aerosol optical thickness inversion system, the system comprising:

[0099] Memory for storing computer programs;

[0100] A processor is used to execute the computer program to implement the full coverage daily scale AOD inversion method of any embodiment of the present invention.

[0101] An embodiment of the present invention further provides a non-transitory computer-readable medium storing instructions. When the instructions are executed by a processor, the full coverage daily-scale AOD inversion method according to any embodiment of the present invention is executed.

[0102] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present invention with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of this application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0103] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present invention. This should not be interpreted as a feature of an invention that is not claimed for protection being necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of the embodiments of a specific invention. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

Claims

1. A full coverage daily scale AOD inversion method, characterized in that: The method comprises: The station data, the meteorological data set, the terrain data set and the population data are processed by a linear interpolation method to obtain preprocessed data, and the preprocessed data are spatially and temporally matched with MAICAAOD to obtain a modeling data set; The triangular irregular network interpolation of MAICAAOD was performed to obtain preliminary daily AOD data and annual average AOD data; Based on the full coverage AOD data and the annual average AOD data, a multi-layer LGBM model is constructed, and multi-step simulation is performed to correct the filling error and obtain the adjusted optimal prediction model; Based on the modeling data set, the optimal prediction model is used to perform estimation to obtain a daily AOD product that fully covers the area.

2. The full coverage daily scale AOD inversion method according to claim 1 is characterized in that: The site data, meteorological data set, terrain data set and population data are processed by linear interpolation method to obtain preprocessed data, and the preprocessed data is matched with MAICAAOD in time and space to obtain a modeling data set, including: Get site data using The index interpolates the two adjacent bands of 670nm and 440nm to obtain the AOD parameters at 550nm, so that the site data can meet the consistency with MAIAC AOD when comparing the accuracy; Obtain NDVI data, DEM data, and population datasets; The site data, NDVI data, DEM data, and population data were matched with the MAICAAOD data accuracy as the modeling dataset.

3. The full coverage daily scale AOD inversion method according to claim 2 is characterized in that: The temporal and spatial resolutions of the NDVI data are 16 days and 250M respectively, the resolution of the DEM data is 30m, and the resolution of the population dataset is 1km.

4. The full coverage daily scale AOD inversion method according to claim 1 is characterized in that: The triangular irregular network interpolation of MAICAAOD is performed to obtain the full coverage AOD data and annual average AOD data, including: Get MAICAAOD; According to the quality assurance mark and valid AOD range provided by the MAIAC AOD product, the MAIACAOD data was cleaned and AOD values ​​greater than 3 were excluded; The existing daily MAICAAOD data are used to perform TIN interpolation on the missing areas in MAIACAOD so as to make use of the existing spatial information for large-scale restoration and thus obtain the daily AOD with sufficient data volume.

5. The full coverage daily scale AOD inversion method according to claim 4 is characterized in that: The temporal and spatial resolutions of the MAICAAOD are 1 d and 1 km×1 km respectively.

6. The full coverage daily scale AOD inversion method according to claim 1 is characterized in that: Based on the preliminary daily AOD data and annual average AOD data, a multi-layer LGBM model was constructed, and multi-step simulation was performed to correct the filling errors and obtain the adjusted optimal prediction model, including: A multi-layer LGBM model was constructed; in the multi-layer LGBM model, the first layer used the initially obtained full coverage AOD products and annual average AOD as the main explanatory variables to fill in the seasonal average value, the second layer used annual and seasonal AOD products as explanatory variables to calculate the monthly average AOD value, the third layer used annual, seasonal and monthly AOD products to fill in weekly AOD, and the last layer used annual, seasonal, monthly and weekly AOD products as explanatory variables to fill in daily AOD on a large scale; the calculation formula of the last layer is expressed as: Among them, AOD Daily (i, j) represents the actual daily AOD level of the specified pixel (i, j); LGBM represents the last layer of the LGBM model; AOD Year (i, j) represents the preliminary annual average AOD; AOD Quarterly (i, j) represents the seasonal average AOD obtained by the first-layer model; AOD Monthly (i, j) represents the monthly average AOD obtained by the second-layer model; AOD Weekly (i, j) represents the weekly AOD obtained by the third-layer model; NDVI (i, j) represents the normalized vegetation index value of the specified pixel point (i, j); DEM (i, j) represents the elevation value of the specified pixel point (i, j); POP (i, j) represents the population data of the specified pixel point (i, j).

7. The full coverage daily scale AOD inversion method according to claim 1 is characterized in that: Based on the modeling data set, the optimal prediction model is used to estimate and obtain a daily AOD product covering the entire region, including: All modeling data are processed into standard spatial grids, and the optimal prediction model is used to restore the daily AOD values ​​in the region to obtain a daily AOD product that covers the entire region.

8. A full coverage daily scale AOD inversion device, characterized in that: The device comprises: The spatiotemporal matching module is configured to process the site data, the meteorological data set, the terrain data set and the population data using a linear interpolation method to obtain preprocessed data, and perform spatiotemporal matching between the preprocessed data and MAICAAOD to obtain a modeling data set; The data construction module is configured to perform triangular irregular network interpolation on MAICAAOD to obtain full coverage AOD data and annual average AOD data; The model inversion module is configured to construct a multi-layer LGBM model based on the full coverage AOD data and the annual average AOD data, and perform multi-step simulation to correct the filling error to obtain the adjusted optimal prediction model; based on the modeling data set, the optimal prediction model is used to perform estimation to obtain the daily AOD product with full coverage in the area.

9. A full coverage daily scale AOD inversion system, characterized in that: The system comprises: Memory for storing computer programs; A processor, configured to execute the computer program to implement the full coverage daily-scale AOD inversion method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, performs the full coverage daily-scale AOD inversion method according to any one of claims 1 to 7.

Images