A method and system for remote sensing inversion of ozone concentration by region and time period

Abstract

The application discloses a regional and periodical ozone concentration remote sensing inversion method and system, and relates to the fields of data processing and ozone monitoring.The method comprises the following steps: acquiring ozone concentration measured values of each ground station in a target region in a preset historical time period and multi-source remote sensing and meteorological data, extracting surface, meteorological and precursor concentration characteristic parameters to form a multi-dimensional characteristic parameter set after preprocessing, dividing a plurality of spatial regions by spatial clustering based on the characteristic set, dividing a monitoring time period into a concentration increasing section and a concentration decreasing section based on a daily variation curve drawn based on the ozone concentration measured values, extracting key characteristic factors from the multi-dimensional characteristic parameter set corresponding to each region and each time period respectively and constructing an inversion model, and finally extracting real-time key characteristic factors of the target region and inputting the corresponding model to generate a spatial distribution result.The application can improve the inversion estimation precision of the ozone concentration.

Description

Technical Field

[0001] This application relates to the fields of data processing and ozone monitoring, and in particular to a method and system for remote sensing inversion of ozone concentration by region and time period. Background Technology

[0002] Ground-level ozone is a major component of photochemical smog, and high concentrations of ozone have adverse effects on human health and crop growth. In recent years, ozone has surpassed PM2.5 to become the primary pollutant in many regions of my country during periods of excessive pollution. Achieving high-precision, continuous spatial monitoring of ground-level ozone concentrations has become a pressing technical challenge in the field of atmospheric environmental monitoring. Ground-based monitoring can obtain accurate time-series data on ozone concentrations, but the number of stations is limited and their spatial distribution is uneven, making it difficult to meet the needs of large-scale continuous spatial monitoring. While satellite remote sensing technology has advantages such as wide coverage and low cost, there is a significant discrepancy between the atmospheric column concentration retrieved and the actual near-ground concentration.

[0003] In existing technologies, researchers have obtained atmospheric composition data such as total ozone column and tropospheric NO2 column concentration obtained from satellite remote sensing inversion, as well as meteorological elements such as temperature and humidity provided by meteorological reanalysis data. They have combined these data with ozone concentration data measured at ground stations to form a training sample set, and used machine learning to train a globally unified inversion model, thereby achieving spatial gridded estimation of near-surface ozone concentration.

[0004] However, existing technologies, when constructing the aforementioned multi-source data fusion models, typically only use instantaneous meteorological elements at the monitoring moment (such as current temperature, humidity, and wind speed) as model input features, rarely considering the impact of the persistent effects of meteorological conditions over a preset historical time period on ozone concentration. Furthermore, the formation, diffusion, and dissipation of near-surface ozone are significantly influenced by factors such as topography, surface cover, and diurnal variations. The response mechanisms of ozone concentration to its influencing factors differ across different regions (e.g., urban and suburban areas) and time periods (e.g., morning and afternoon). Under the aforementioned globally unified modeling framework, model parameters are shared across the entire region and time period. When a certain influencing factor has a weak correlation with ozone concentration in a specific region or time period, the noise introduced by that factor reduces the model's inversion accuracy within that region or time period. This reduces the accuracy of near-surface ozone concentration estimation in complex geographical environments. Summary of the Invention

[0005] This application provides a method and system for remote sensing inversion of ozone concentration by region and time period, which is used to solve the problem of how to improve the accuracy of inversion estimation of near-ground ozone concentration in complex geographical environments.

[0006] In a first aspect, embodiments of this application provide a method for remote sensing inversion of ozone concentration by region and time period, including: Acquire measured ozone concentration values, remote sensing image data, meteorological reanalysis data, and atmospheric composition remote sensing data of various ground stations within the target area over a preset historical time period. The preset historical time period ends at the current monitoring time. The remote sensing image data, meteorological reanalysis data and atmospheric composition remote sensing data are preprocessed to obtain a preprocessed dataset, in which all data have the same spatial resolution. Surface feature parameters, meteorological feature parameters, and precursor concentration feature parameters are extracted from the dataset to form a multidimensional feature parameter set. Meteorological feature parameters and precursor concentration feature parameters have a continuous impact on ozone concentration within a preset historical time period. Using a multidimensional feature parameter set as clustering samples, the target area is spatially clustered to obtain multiple spatial regions. Based on the measured ozone concentration values ​​in each spatial region, the daily variation curves of ozone concentration in each spatial region are plotted. Based on the times of the lowest and highest concentrations in the daily variation curve of ozone concentration, the monitoring period is divided into a concentration increase segment and a concentration decrease segment. Key feature factors for the concentration increase segment and the concentration decrease segment are extracted from the multidimensional feature parameter set corresponding to the spatial region, respectively. Using measured ozone concentrations as dependent variables and key characteristic factors as independent variables, ozone concentration inversion models were constructed for each spatial region under the corresponding concentration increase and concentration decrease segments. The real-time key feature factors of the target area in each spatial region are input into the corresponding ozone concentration inversion model, and the ozone concentration inversion value in raster format is output to generate the spatial distribution result of ozone concentration in the target area.

[0007] Optionally, the remote sensing image data, meteorological reanalysis data, and atmospheric composition remote sensing data are preprocessed to obtain a preprocessed dataset. This includes: quantitative preprocessing of the remote sensing image data; format unification and anomaly handling of the meteorological reanalysis data and atmospheric composition remote sensing data to obtain preprocessed raster data; using the highest spatial resolution in each raster data as the target spatial resolution; and resampling each raster data to the target spatial resolution using spatial interpolation methods to obtain a resampled dataset, so that the coordinates of each raster data in the dataset are aligned on the spatial grid.

[0008] Optional meteorological parameters include the average 2-meter temperature, average relative humidity, average 10-meter zonal wind speed, average 10-meter meridional wind speed, maximum boundary layer height, cumulative net solar radiation, and cumulative rainfall over a preset historical period; precursor concentration parameters include the average ozone column concentration and average nitrogen dioxide column concentration over a preset historical period; and surface parameters include the normalized difference vegetation index, nighttime light index, digital elevation model, slope, and aspect.

[0009] Optionally, a multidimensional feature parameter set is used as clustering samples to spatially cluster the target area, resulting in multiple spatial regions. This includes: determining clustering samples for each grid point based on the multidimensional feature parameter set, where each clustering sample is a feature vector composed of the parameter values ​​of the corresponding grid point in the dataset, including surface feature parameters, meteorological feature parameters, and precursor concentration feature parameters; randomly selecting initial cluster centroids based on the K-means clustering algorithm, updating the initial cluster centroids and redistributing each clustering sample for multiple iterations with the optimization objective of minimizing the sum of squared errors within the clusters, until the sum of squared errors within the clusters converges; plotting a scatter plot of the sum of squared errors within the clusters as a function of the number of clusters, determining the number of clusters corresponding to the point with the largest rate of change in the scatter plot as the number of categories for spatial division, and dividing the target area into multiple spatial regions based on the number of categories.

[0010] Optionally, key feature factors for the concentration increase segment and concentration decrease segment are extracted from the multidimensional feature parameter set corresponding to the spatial region, including: collecting all corresponding multidimensional feature parameters of the spatial region in the concentration increase segment and concentration decrease segment, respectively, as sample subsets of the spatial region in the concentration increase segment and concentration decrease segment; performing principal component analysis on each sample subset to obtain multiple principal component factors of the sample subset, and calculating the variance contribution value of each principal component factor in the sample subset; calculating the total value of all variance contribution values ​​in the sample subset, and accumulating the first variance contribution value to the current variance contribution value of the current principal component factor in descending order of variance contribution value to obtain the cumulative contribution value of the current principal component factor; using the ratio of the cumulative contribution value to the total value as the cumulative contribution rate of the current principal component factor; when the cumulative contribution rate first reaches the preset contribution threshold, using the first principal component factor corresponding to the first variance contribution value to the current principal component factor in descending order of variance contribution value as the key feature factors corresponding to the sample subset.

[0011] Optionally, using measured ozone concentrations as the dependent variable and key feature factors as independent variables, ozone concentration inversion models are constructed for each spatial region under the concentration increase and decrease phases, respectively. This includes: matching the key feature factors corresponding to each sample subset with the measured ozone concentrations of the corresponding spatial region to construct inversion sample sets for each sample subset; dividing each inversion sample set into a training sample set and a test sample set according to a preset ratio; and removing discrete samples from the training sample set based on the feature value distribution of the key feature factors corresponding to each sample in the training sample set, obtaining the removed samples. The training sample set is then used. Key feature factors from each removed training sample set are used as input variables, and the corresponding measured ozone concentration is used as the output variable. Model fitting training is then performed to obtain the inversion model corresponding to each sample subset. Key feature factors from the test sample set are input into the corresponding inversion model to obtain the estimated ozone concentration for the test sample set. The root mean square error (RMSE) and average accuracy are calculated based on the measured ozone concentration in the test sample set. When the RMS and average accuracy meet the preset accuracy requirements, the corresponding inversion model is used as the ozone concentration inversion model for the sample subset.

[0012] Optionally, the real-time key feature factors of the target area in each spatial region are input into the corresponding ozone concentration inversion model, and the ozone concentration inversion values ​​in raster format are output to generate the spatial distribution results of ozone concentration in the target area. This includes: obtaining the multi-dimensional feature parameters corresponding to each spatial region at the current inversion time; determining the real-time key feature factors of each spatial region at the current inversion time based on each spatial region and the time period in which the current inversion time is located; determining the target ozone concentration inversion model that matches the corresponding spatial region based on each spatial region and the time period in which the current inversion time is located, inputting the corresponding real-time key feature factors into the target ozone concentration inversion model, and calculating all ozone concentration inversion values ​​in raster format under the spatial region; and combining the ozone concentration inversion values ​​according to the spatial geographic coordinates of each spatial region to generate the spatial distribution results of ozone concentration in the target area.

[0013] Secondly, embodiments of this application provide a regional and time-segmented ozone concentration remote sensing inversion system, which includes: one or more processors and a memory; the memory is coupled to one or more processors, and the memory is used to store computer program code, which includes computer instructions. One or more processors call the computer instructions to cause the regional and time-segmented ozone concentration remote sensing inversion system to perform the method described in the first aspect and any possible implementation thereof.

[0014] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a regional and time-segmented ozone concentration remote sensing inversion system, cause the regional and time-segmented ozone concentration remote sensing inversion system to perform the method described in the first aspect and any possible implementation thereof.

[0015] Fourthly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a regional and time-segmented ozone concentration remote sensing inversion system, cause the regional and time-segmented ozone concentration remote sensing inversion system to execute the method described in the first aspect and any possible implementation thereof.

[0016] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: 1. By acquiring multi-source data within a preset historical time period and constructing a multi-dimensional feature parameter set, and then using this feature set for spatial clustering to divide the area into multiple spatial regions, and based on the daily variation curve of ozone concentration, the monitoring period is divided into concentration growth segment and concentration decrease segment. Then, key feature factors are extracted from the feature sets of each region and time period, and inversion models are constructed separately. This solves the technical problem in traditional global unified modeling methods that cannot characterize the differences in the response mechanism of ozone concentration to influencing factors under different terrains and time periods due to the sharing of model parameters across the entire region and time period. This significantly improves the spatial estimation accuracy of near-surface ozone concentration in complex geographical environments.

[0017] 2. By unifying the format and resampling of multi-source remote sensing and meteorological data, and incorporating the cumulative meteorological statistics, ozone column concentration, and nitrogen dioxide column concentration mean values ​​within a preset historical time period into the feature parameter set, the model simultaneously characterizes the impact of the time-cumulative effect of ozone concentration itself and the continuous effect of meteorological conditions on the current ozone concentration. This overcomes the problem of missing feature information caused by existing technologies that only use instantaneous meteorological data and ignore the continuous influence of historical meteorological conditions. At the same time, the introduction of surface feature parameters such as normalized vegetation index, nighttime light index, digital elevation model, slope, and aspect enables the model to fully reflect the influence of underlying surface conditions on ozone generation and diffusion, further enhancing the model's ability to characterize the spatiotemporal variation of ozone concentration.

[0018] 3. K-means clustering was used to iteratively cluster the ozone concentration with the goal of minimizing the sum of squared errors within each cluster. The number of clusters corresponding to the point with the largest rate of change in ozone concentration was used as the number of spatial partitions, thus achieving objective spatial partitioning of the study area and avoiding the uncertainty caused by subjective manual partitioning. At the same time, principal component analysis was performed on the sample subsets of each spatial region in the concentration increase and decrease phases. The variance contribution values ​​were accumulated one by one from largest to smallest until the cumulative contribution rate first reached the preset threshold. Key feature factors were extracted, which enabled adaptive screening of feature factors under different spatiotemporal categories. This ensured that the key feature factors retained in each category could accurately reflect the dominant influencing factors of ozone concentration under that category, effectively eliminating the noise interference introduced by irrelevant variables, and thus further improving the accuracy of the inversion model under each spatiotemporal category.

[0019] 4. By matching the key feature factors of each sample subset with the measured ozone concentration values ​​at the corresponding locations to construct an inversion sample set, which is then divided into a training set and a test set, discrete samples are removed based on the feature value distribution in the training set before model training. The root mean square error and average precision are calculated using the test set for dual verification, thus ensuring that the inversion models for each region and time period have undergone rigorous quality control and guaranteeing the reliability of the models in practical applications. Simultaneously, by dividing the target area into grids to be inverted and matching the corresponding inversion model based on the spatial region to which the grid belongs and the time period at the current moment, adaptive spatial continuous inversion of the entire region and time period is achieved, thereby providing reliable spatial distribution data support for the precise control of ozone pollution. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the remote sensing inversion method for ozone concentration by region and time period provided in the embodiments of this application; Figure 2 This is a schematic diagram of a scatter plot of SSE versus K value provided in an embodiment of this application. Figure 3 This is a schematic diagram of the spatial region division provided in an embodiment of this application; Figure 4 This is a schematic diagram of time period division provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of a remote sensing inversion system for ozone concentration by region and time period provided in an embodiment of this application.

[0021] Explanation of reference numerals in the attached figures: 201, Central Processing Unit; 202, Read-Only Memory; 203, Random Access Memory; 204, Bus; 205, Input / Output Interface; 206, Input Section; 207, Output Section; 208, Storage Section; 209, Communication Section; 210, Driver; 211, Removable Media. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0023] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0024] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0025] This application provides a method and system for remote sensing inversion of ozone concentration by region and time period, which is used to solve the problem of how to improve the accuracy of near-surface ozone concentration inversion estimation under complex geographical environments.

[0026] Figure 1 This is a flowchart illustrating the remote sensing inversion method for ozone concentration by region and time period provided in the embodiments of this application.

[0027] This application discloses a remote sensing inversion method for ozone concentration by region and time period, such as... Figure 1 As shown, the steps include the following.

[0028] S101. Obtain measured ozone concentration values, remote sensing image data, meteorological reanalysis data, and atmospheric composition remote sensing data of various ground stations within the target area during a preset historical time period. The preset historical time period ends at the current monitoring time.

[0029] Specifically, the geographical scope of the target area is determined, and hourly ozone concentration measurements from all national and provincial air quality monitoring stations within the area are obtained over a preset historical time period. Each data record includes the station number, monitoring time, and ozone concentration value. Remote sensing imagery covering the target area is acquired, including high-resolution multispectral satellite imagery (e.g., Sentinel-2) and nighttime light remote sensing data (e.g., VIIRS). Multispectral imagery is used to calculate the normalized vegetation index, and nighttime light data is used to characterize the intensity of human activities. Meteorological reanalysis data (e.g., ERA5) covering the target area is acquired. This data provides hourly gridded meteorological element data, including temperature, humidity, wind speed, boundary layer height, net surface solar radiation, and precipitation. Simultaneously, atmospheric composition remote sensing data (e.g., TROPOMI) covering the target area is acquired. This data provides daily or orbital products of ozone column concentration and tropospheric nitrogen dioxide column concentration. All the aforementioned time-series data (measured ozone concentration, meteorological reanalysis data, and remote sensing data of atmospheric composition) were acquired by looking back over a preset historical time period (24 hours in this embodiment) from the current monitoring time as the end time, ensuring that each data point has a consistent start and end boundary in the time dimension. Remote sensing image data were acquired from cloudless images available on the monitoring day or the most recent available image. If no image was available on the monitoring day, the best quality image from several days before and after the monitoring day could be selected as a substitute.

[0030] The target area refers to the research area where ozone concentration inversion is currently being conducted, including but not limited to cities, counties, and districts. Ground stations are ground-based ozone monitoring stations deployed within the target area by national or local environmental monitoring departments; their measured values ​​are actual observational data of near-surface ozone concentration, used as the true values ​​of the dependent variable for subsequent model training. Remote sensing image data includes multispectral optical image data, used to extract surface characteristic parameters. Meteorological reanalysis data are global or regional-scale atmospheric reanalysis gridded data, used to extract meteorological characteristic parameters. Atmospheric composition remote sensing data are satellite-retrieved trace gas column concentration data, used to extract precursor concentration characteristic parameters. The preset historical time period ends at the current monitoring time; its duration is determined based on the average lifetime of ozone and its precursors in the atmosphere and the duration of the continuous influence of meteorological conditions, and can be selected as 12 hours, 24 hours, or 48 hours, etc. In this embodiment, 24 hours is preferred to cover one complete photochemical cycle.

[0031] S102. Preprocess the remote sensing image data, meteorological reanalysis data, and atmospheric composition remote sensing data to obtain a preprocessed dataset, in which all data have the same spatial resolution.

[0032] Specifically, corresponding preprocessing operations are performed for different types of raw data: remote sensing image data needs to undergo radiometric calibration to convert the raw signal values ​​into physical radiance, atmospheric correction to eliminate the influence of atmospheric scattering and absorption effects on the surface signal, and geometric correction to establish an accurate mapping relationship between pixel coordinates and geographic coordinates; meteorological reanalysis data and atmospheric composition remote sensing data need to undergo format conversion to support unified processing, and outlier identification and removal to ensure data quality. After the above preprocessing is completed, all data are converted to raster format. The target spatial resolution is determined by comparing the spatial resolution of various types of preprocessed data. Usually, the original resolution of multispectral remote sensing imagery is used as the benchmark, because this resolution is usually the highest. If the original resolution of meteorological reanalysis data or atmospheric composition remote sensing data is lower than this benchmark, spatial interpolation methods are used to upsample it to the benchmark resolution; if the original resolution of individual data is higher than the benchmark, it is downsampled to the benchmark resolution. The spatial interpolation method is selected according to the data properties: for continuously changing meteorological elements and column concentration data, bilinear interpolation is used; for exponential data with discrete distribution characteristics, the nearest neighbor method is used. After resampling, all raster data are consistent in terms of cell size, number of rows and columns, and spatial reference frame of the spatial grid, thus completing spatial coordinate alignment.

[0033] Based on the above embodiments, as an optional embodiment, for Figure 1 Step S102 shown can be implemented through the following steps, which are explained in detail below.

[0034] Quantitative preprocessing is performed on remote sensing image data, and format unification and anomaly processing are performed on meteorological reanalysis data and atmospheric composition remote sensing data to obtain preprocessed raster data. The highest spatial resolution in each raster data is taken as the target spatial resolution. Spatial interpolation method is used to resample each raster data to the target spatial resolution to obtain a resampled dataset, so that the coordinates of each raster data in the dataset are aligned on the spatial grid.

[0035] Specifically, radiometric calibration, atmospheric correction, and geometric correction are performed sequentially on the remote sensing image data: radiometric calibration converts the raw digital quantization values ​​recorded by the satellite sensor into radiance values ​​at the top of the atmosphere; atmospheric correction uses a radiative transfer model (such as 6S or MODTRAN) to eliminate the scattering and absorption effects of atmospheric molecules and aerosols on the surface reflectance signal, outputting the true surface reflectance; geometric correction uses the geometric positioning parameters or ground control points inherent in the image to establish a mapping relationship between pixel coordinates and geographic coordinates, eliminating geometric distortion of the image. Simultaneously, the meteorological reanalysis data and atmospheric composition remote sensing data underwent format unification and anomaly processing: Format unification converted standard formats such as NetCDF or GRIB to a common raster data format (e.g., GeoTIFF), and extracted the variable layers corresponding to the target meteorological elements or trace gas components; Anomaly processing checked each raster pixel to ensure that the values ​​fell within a physically reasonable range (e.g., temperature between -50℃ and 50℃, ozone column concentration between 100-500 DU), marking pixels outside the reasonable range as outliers and correcting them using neighborhood interpolation or temporal interpolation. At the same time, time periods with severe data loss were identified and removed. After the above processing, all data were unified into a raster format, and each data point maintained its original state in terms of spatial range and pixel size.

[0036] The process iterates through all preprocessed raster data, compares the cell sizes of each data point, and takes the smallest value as the target resolution (e.g., 10m for Sentinel-2 imagery). Spatial interpolation is then used to resample each raster data to the target spatial resolution: for continuous raster data (such as temperature, humidity, and bar density), bilinear interpolation is used, locating the four nearest surrounding cells based on the target cell's center point's position in the original raster, and calculating the target cell value using a weighted average of the inverse distance; for discrete or exponential raster data (such as nighttime light data), the nearest neighbor method is used, directly assigning the value of the original cell closest to the target cell's center point to the target cell. After resampling, all raster data have a uniform cell size, number of rows and columns, and spatial reference system on the same spatial grid. Each cell position strictly corresponds to the same geographic spatial location across different data points, achieving spatial grid coordinate alignment and forming the resampled dataset.

[0037] For example, the Sentinel-2 10m resolution multispectral imagery (which has the highest spatial resolution) will be used as a reference, and the remaining data will be uniformly resampled to 10m. Radiometric calibration, atmospheric correction, and geometric correction were performed based on the Sentinel-2 imagery's metadata file and pre-calibrated sensor calibration parameters, using the Sen2Cor atmospheric correction module. Geometric correction utilized the imagery's built-in geometric positioning information. ERA5 reanalysis data was converted from GRIB to GeoTIFF format using the CDO tool. Each meteorological element value was checked pixel-by-pixel to ensure it fell within a physically reasonable range (e.g., temperature between -50°C and 50°C), and then resampled to 10m using bilinear interpolation. Tropomi data was converted from NetCDF to GeoTIFF format, and column concentration values ​​were similarly checked to ensure they fell within a reasonable range (ozone column concentration between 100-500 DU, nitrogen dioxide column concentration between 0-0.05 mol / m²), and then resampled to 10m using bilinear interpolation. VIIRS nighttime light data was resampled to 10m using the nearest neighbor method to ensure the numerical integrity of the nighttime light index. After resampling, all raster data were aligned pixel-by-pixel on the same spatial grid to form the resampled dataset.

[0038] Through the above embodiments, unified matching of multi-source heterogeneous data at a spatial scale was achieved. Quantitative preprocessing eliminated radiation distortion and geometric deformation introduced by sensors and the atmospheric environment, ensuring the accuracy of surface feature parameter extraction; format unification and anomaly handling eliminated interference from data storage format differences and outliers on subsequent modeling; resampling with the highest spatial resolution as the target ensured that all data were aligned on a unified pixel grid, providing a spatial matching basis for each pixel location in the subsequent multi-dimensional feature parameter set to simultaneously associate three types of features: surface, meteorological, and precursor concentration, avoiding the problem of feature vector misalignment due to inconsistent spatial scales.

[0039] S103. Extract surface feature parameters, meteorological feature parameters, and precursor concentration feature parameters from the dataset to form a multidimensional feature parameter set. The meteorological feature parameters and precursor concentration feature parameters have a continuous impact on ozone concentration within a preset historical time period.

[0040] Specifically, band operations are performed on the optical remote sensing images in the dataset to extract spectral indices representing surface vegetation cover. Elevation information is directly read from the topographic data layer in the dataset, and surface slope and aspect are calculated based on the elevation data, with this information serving as surface feature parameters. From the meteorological reanalysis grid data in the dataset, time-series data of each meteorological element within a preset historical time period are extracted for each pixel. Based on the influence mechanism of each element on ozone concentration, statistical compression is performed in the time dimension, using the mean value to represent the average state of the element, the maximum value to represent extreme conditions, and the cumulative value to represent the total effect. The compressed statistics are used as meteorological feature parameters. From the atmospheric composition remote sensing data in the dataset, time-series data of trace gas column concentrations within a preset historical time period are extracted for each pixel, and their mean concentration within the time window is calculated, serving as precursor concentration feature parameters. Finally, the above three types of feature parameters are correlated and stitched pixel by pixel according to spatial location, so that each pixel location corresponds to a comprehensive feature vector containing surface attributes, meteorological conditions, and precursor levels. The feature vectors of all pixels together constitute a multi-dimensional feature parameter set.

[0041] The table below provides an example table of multidimensional feature parameters provided in the embodiments of this application:

[0042] Meteorological characteristic parameters include the average 2-meter temperature, average relative humidity, average 10-meter zonal wind speed, average 10-meter meridional wind speed, maximum boundary layer height, cumulative net solar radiation, and cumulative rainfall over a preset historical time period (e.g., 24 hours). Precursor concentration characteristic parameters include the average ozone column concentration and average tropospheric nitrogen dioxide column concentration over a preset historical time period (e.g., 24 hours). Surface characteristic parameters include the normalized difference vegetation index (NDVI), nighttime light index, digital elevation model (DEM), slope, and aspect.

[0043] Specifically, the normalized vegetation index (NDI) is calculated pixel-by-pixel from the red and near-infrared reflectance of multispectral remote sensing images; the brightness value is extracted pixel-by-pixel from nighttime light remote sensing data as the nighttime light index; and the elevation value is read pixel-by-pixel from digital elevation model data, and slope and aspect are calculated using a 3×3 window neighborhood operation. These five types of parameters constitute the surface characteristic parameters. From the meteorological reanalysis data, for each pixel, data on the continuous impact of ozone concentration at all times within a preset historical time period (ending at the current monitoring time) are extracted. The arithmetic mean of 2-meter temperature, the arithmetic mean of relative humidity, the arithmetic mean of 10-meter zonal wind speed, the arithmetic mean of 10-meter meridional wind speed, the maximum value of boundary layer height, the cumulative value of net surface solar radiation, and the cumulative value of rainfall are calculated. These seven types of statistics constitute the meteorological characteristic parameters. From the ozone and nitrogen dioxide column concentration fields of atmospheric composition remote sensing data, the column concentration values ​​that have a continuous impact on ozone concentration at each time point within a preset historical time period are extracted for each pixel, and the arithmetic mean is calculated for each time point to form the precursor concentration characteristic parameters.

[0044] Here, 2-meter temperature refers to the atmospheric temperature at a height of 2 meters above the ground, and relative humidity is the percentage of actual water vapor pressure in the air at that height to saturated water vapor pressure at the same temperature. The average of these two values ​​characterizes the average thermal and humidity conditions during this time period. 10-meter zonal wind speed and 10-meter meridional wind speed refer to the east-west and north-south components of wind speed at a height of 10 meters above the ground, respectively. Their averages characterize the average wind field characteristics during this time period. The maximum boundary layer height characterizes the optimal state of atmospheric vertical diffusion during this time period. The cumulative net solar radiation at the surface characterizes the total solar radiation energy reaching the Earth's surface during this time period, reflecting the energy supply for photochemical reactions. The cumulative precipitation characterizes the total amount of ozone and its precursors removed by wet deposition during this time period. The precursor concentration characteristic parameters include the average ozone column concentration and the average nitrogen dioxide column concentration within a preset historical time period before the monitoring time. The arithmetic mean of the ozone column concentration and nitrogen dioxide column concentration at each moment within the preset historical time period is calculated from the atmospheric composition remote sensing data. The column concentration represents the total amount of target gas in the vertical column of the atmosphere, and its mean is used to characterize the average cumulative level of precursors within that time period.

[0045] For example, based on the aforementioned 10m×10m pixel grid of the target area, feature parameters of each pixel are extracted. The Normalized Difference Vegetation Index (NDV) is calculated pixel-by-pixel using the red and near-infrared bands of Sentinel-2 imagery; the nighttime light index is directly extracted from the brightness values ​​of each pixel in the VIIRS data; DEM, slope, and aspect are extracted from SRTM data. Meteorological feature parameters are calculated using a 24-hour historical time period. Data on 2-meter temperature, relative humidity, 10-meter zonal wind speed, 10-meter meridional wind speed, boundary layer height, net surface solar radiation, and precipitation are extracted pixel-by-pixel from ERA5 data for each time point within the 24-hour period. The mean values ​​of temperature, humidity, and wind speed are calculated, and the maximum value of the boundary layer height is taken to calculate the cumulative values ​​of net surface solar radiation and precipitation. Precursor concentration feature parameters are extracted from the TROPOMI data for each time point within the 24-hour period, and the mean values ​​are calculated. Each pixel obtains the above 14 feature parameters, forming a multi-dimensional feature parameter set, with each pixel corresponding to a 14-dimensional feature vector.

[0046] S104. Using the multidimensional feature parameter set as clustering samples, the target area is spatially clustered to obtain multiple spatial regions. Based on the measured ozone concentration values ​​in each spatial region, the daily variation curve of ozone concentration in each spatial region is plotted.

[0047] Specifically, the feature vector corresponding to each pixel in the multidimensional feature parameter set is used as an independent clustering sample. The feature dimension of the sample is the total number of three-dimensional feature parameters. The distance between samples in the feature space is used to measure the similarity of their underlying surface conditions, meteorological background, and precursor levels. Then, a clustering algorithm is used to divide all samples into several clusters according to feature similarity. Samples within the same cluster are closer in the feature space, while samples in different clusters are farther apart. Pixels in the target area are classified according to the cluster labels of each sample. Pixels with the same cluster label constitute a spatial region, thus spatially segmenting the target area into multiple sub-regions with different feature combinations. At the same time, for each spatial region obtained after clustering, the measured ozone concentration values ​​of all ground stations in the region are collected at each time during the monitoring period. The concentration values ​​are grouped according to the monitoring time, and the average concentration of all stations in the region at the same time is calculated. A curve is plotted with time as the x-axis and concentration as the y-axis. Each point on the curve represents the average ozone concentration level of the region at that time. The lowest and highest points on each curve are identified, and their corresponding times are recorded for subsequent time period division.

[0048] In this context, a spatial region refers to a sub-region spatially divided from the target region through cluster analysis. Pixels within the same spatial region exhibit similarity in the feature space, while pixels from different spatial regions show differences in the feature space. Each spatial region consists of multiple spatially continuous or discrete pixels. The diurnal variation curve is a curve plotted with the monitoring time as the x-axis and ozone concentration as the y-axis. It characterizes the variation of ozone concentration over time within a day, typically exhibiting a unimodal distribution. The diurnal variation curves for ozone concentration in each spatial region are individually plotted for each region, rather than using a uniform diurnal variation curve for the entire region. The shapes of the diurnal variation curves may differ between spatial regions, reflecting the variations in the diurnal variation patterns of ozone concentration in different areas.

[0049] Based on the above embodiments, as an optional embodiment, for Figure 1 The step of dividing the spatial region in step S104 shown can be achieved through the following steps, which are explained in detail below.

[0050] Cluster samples for each grid point are determined based on a multidimensional feature parameter set. The cluster samples are feature vectors composed of the parameter values ​​of the corresponding grid point in the dataset, including surface feature parameters, meteorological feature parameters, and precursor concentration feature parameters. Initial cluster centroids are randomly selected based on the K-means clustering algorithm. With minimizing the sum of squared errors within the cluster as the optimization objective, the initial cluster centroids are updated and the cluster samples are redistributed for multiple iterations until the sum of squared errors within the cluster converges. A scatter plot of the sum of squared errors within the cluster versus the number of clusters is plotted. The number of clusters corresponding to the point with the largest rate of change in the scatter plot is determined as the number of categories for spatial division. The target area is divided into multiple spatial regions based on the number of categories.

[0051] Specifically, the number of clusters is set to k. K samples are randomly selected from the cluster sample set as initial cluster centroids. Each cluster centroid is a feature vector with the same dimension as the cluster samples. For each sample in the cluster sample set, the Euclidean distance to each initial cluster centroid is calculated. The sample is then assigned to the cluster corresponding to the nearest centroid, completing the first round of assignment. After the first round of assignment, for each cluster, the arithmetic mean of the feature vectors of all samples within the cluster is taken as the new cluster centroid, replacing the original centroid. The distance from each sample to each cluster centroid is recalculated using the updated centroid, and the samples are reassigned to the nearest cluster, updating the cluster centroid again. This process of "assigning samples and updating centroids" is repeated. After each iteration, the sum of squared errors (SSE) within the cluster is calculated, and the SSE of the current iteration is compared with the SSE of the previous iteration. When the absolute change in SSE between two adjacent iterations is less than a preset convergence threshold (e.g., a change rate less than 0.01% or an absolute change less than 10), the cluster centroid is considered closed. -6When the clustering reaches convergence, the iteration stops and the final cluster label of each sample is output.

[0052] To determine the optimal number of spatial partitioning categories, the number of clusters k is set from 2 to 10 (the upper limit can be adjusted according to the area and complexity of the target region). The K-means clustering process described above is performed for each k value, and its corresponding SSE is recorded. A scatter plot is drawn with k as the x-axis and SSE as the y-axis. This plot typically shows a monotonically decreasing trend, but the rate of decrease varies across different k value ranges. The rate of change of the rate of decrease at each point on the curve is calculated, i.e., the ratio of the decrease in SSE to the increase in k between two adjacent points. The change in this ratio between adjacent intervals is further calculated, and the point with the largest change is taken as the inflection point of the curve. This inflection point indicates that after the number of clusters exceeds this value, the marginal contribution of further increasing the number of clusters to reducing SSE significantly decreases. Therefore, the number of clusters corresponding to this inflection point is taken as the optimal number of spatial partitioning categories. K-means clustering is re-executed with the same number of clusters. Based on the clustering results, each grid point in the target region is classified according to its cluster label. Grid points belonging to the same cluster label and spatially adjacent or non-adjacent together constitute a spatial region. Thus, the target region is spatially divided into multiple sub-regions with different combinations of features.

[0053] In this context, a grid point refers to a basic spatial unit with a uniform spatial resolution in the preprocessed dataset. Each grid point has unique geographic coordinates and corresponds to the same geographic location across all raster data, serving as the smallest spatial unit for feature extraction and cluster analysis. A cluster sample is a feature vector composed of all feature parameter values ​​belonging to a single grid point, extracted from a set of multidimensional feature parameters, and used as input for the clustering algorithm. A feature vector is a one-dimensional numerical sequence formed by arranging the values ​​of each feature parameter in a fixed feature order, representing the grid point's position in the feature space. The sum of squared intra-cluster errors is the sum of the squares of the Euclidean distances from each cluster sample to the centroid of the cluster, used to measure the density of samples within the cluster; the clustering algorithm aims to minimize this value.

[0054] Example, reference Figure 2 , Figure 2This is a schematic diagram of the scatter plot of SSE versus K value provided in the embodiments of this application. The horizontal axis represents the K value, and the vertical axis represents SSE. During K-means clustering, it iterates from k=1 to 8, performs clustering for each k value, and records the SSE value. The point with the largest rate of change in the SSE curve is calculated to be k=3 (SSE decreases by about 50% from k=2 to k=3, and by about 20% from k=3 to k=4, showing the largest rate of change). k=3 is taken as the optimal number of clusters. The number of clusters is set to 3, and 3 initial cluster centroids are randomly selected. Samples are iteratively allocated and centroids are updated. Convergence occurs when the change in SSE between two consecutive iterations is less than 0.01%. After clustering is completed, refer to... Figure 3 , Figure 3 This is a schematic diagram of spatial area division provided in the embodiments of this application. The target area is divided into three spatial areas: urban high-value area, suburban transition area, and suburban low-value area according to the category labels of each grid point.

[0055] Through the above embodiments, objective spatial partitioning of the study area was achieved. K-means clustering groups similar grid points in the feature space into the same category, so that grid points in the same spatial region have similar underlying surface conditions, meteorological background and precursor levels, and there are significant differences between different spatial regions, avoiding the uncertainty brought about by manual subjective division of regions. The SSE inflection point method adaptively determines the optimal number of partitions, rather than setting a fixed number of partitions in advance, so that the partitioning results objectively reflect the distribution structure of the data itself, providing a reasonable spatial partitioning basis for the subsequent independent extraction of key feature factors and construction of inversion models for each spatial region.

[0056] S105. Based on the time of the lowest and highest concentration points in the daily variation curve of ozone concentration, the monitoring period is divided into a concentration increase segment and a concentration decrease segment.

[0057] For details, please refer to Figure 3 , Figure 3 This is a schematic diagram of time period division provided in the embodiments of this application. Figure 3 The diurnal variation curve of ozone concentration is displayed through discrete concentration points, exhibiting a unimodal variation characteristic. This means that within a day, there is a minimum concentration point and a maximum concentration point. The minimum concentration point refers to the moment when the ozone concentration on the curve reaches its lowest value for the entire day, typically occurring around sunrise. Figure 3 t1 in the curve refers to the moment when the ozone concentration reaches its maximum value for the entire day, typically occurring between afternoon and evening. Figure 3The time period is divided into two phases, with the lowest and highest points as time boundaries: The period from the lowest point (t1) to the highest point (t2) is the concentration increase phase, during which ozone concentration shows a continuous upward trend over time, corresponding to a stage where photochemical reactions gradually intensify and the ozone generation rate exceeds the consumption rate; the period from the highest point (t2) to the next day's lowest point (t1) is the concentration decrease phase, during which ozone concentration shows a continuous downward trend over time, corresponding to a stage where photochemical reactions gradually weaken and the ozone consumption rate exceeds the generation rate. It should be noted that the time period division is performed independently for each spatial region. Due to differences in underlying surface conditions and meteorological backgrounds in different spatial regions, the extreme point times of their diurnal variation curves may differ. Therefore, the time boundaries of the concentration increase and decrease phases in each spatial region may not be entirely consistent.

[0058] For example, taking the aforementioned target area as an example, if the diurnal variation curve of the high-value area in the urban area shows that the lowest ozone concentration in the area occurs at 06:00 (approximately 40 μg / m³) and the highest occurs at 17:00 (approximately 120 μg / m³), then the concentration increase period in this area is from 06:00 to 17:00, and the concentration decrease period is from 17:00 to 06:00 the next day. The diurnal variation curve of the suburban transition area shows that the lowest point occurs at 06:30 (approximately 35 μg / m³) and the highest point occurs at 16:30 (approximately 105 μg / m³), with the increase period from 06:30 to 16:30 and the decrease period from 16:30 to 06:30 the next day. The diurnal variation curve for the low-value area in the suburbs shows the lowest point at 07:00 (approximately 30 μg / m³) and the highest point at 16:00 (approximately 90 μg / m³). The growth period is from 07:00 to 16:00, and the decline period is from 16:00 to 07:00 the next day. The differences in the timing of extreme values ​​in different areas reflect the differences in the influence of urban heat island, precursor emissions, and local meteorological conditions on each area. If the current monitoring time is 10:00, then the high-value area in the urban area, the transitional area in the near suburbs, and the low-value area in the suburbs are all in their respective concentration growth phases.

[0059] Through the above embodiments, the extreme points of the diurnal variation curves of each spatial region are independently identified. This allows each region to determine the time boundaries of the growth and decline phases based on its own diurnal ozone concentration variation patterns, accurately delineating the generation and accumulation phase dominated by photochemical reactions and the consumption and dissipation phase after the photochemical reactions weaken. This provides a precise temporal basis for subsequently extracting key feature factors and constructing inversion models for each spatial region at different time periods. The time period division results, together with the spatial division results, constitute a two-dimensional classification framework of "spatial region × time period," resulting in samples within each combined category exhibiting more consistent ozone concentration response characteristics.

[0060] S106. Extract key feature factors from the multidimensional feature parameter set corresponding to the spatial region for the concentration increase segment and the concentration decrease segment respectively.

[0061] Specifically, based on the spatial region and time period division results obtained in the preceding steps, a two-dimensional classification framework of "spatial region × time period" is formed. For each spatial region, all grid point samples within that region are further divided into two subsets according to the monitoring time: samples falling within the concentration increase segment of that region are assigned to the increase segment subset, and samples falling within the concentration decrease segment are assigned to the decrease segment subset. Thus, each spatial region generates two sample subsets, for a total of 2k sample subsets. For each sample subset, using the multidimensional feature parameters of all grid points in that subset as input, a feature extraction algorithm is used to extract the comprehensive variable or representative feature subset from the original high-dimensional feature space that can retain the original feature information to the greatest extent and has the lowest redundancy between variables, as the key feature factor of that subset. The selection of feature extraction algorithms is based on the data characteristics of the sample subsets: when there is a strong linear correlation between the feature parameters in the sample subset, linear transformation methods can be used to map the original features to a low-dimensional space; when there is a nonlinear relationship between the feature parameters, nonlinear dimensionality reduction methods can be used to extract representative features; when it is necessary to retain the physical interpretability of the original features, feature selection methods based on correlation analysis or importance ranking can be used to directly select key original features. The extracted key feature factors are used to replace the original multidimensional feature parameters as input variables for the subsequent inversion model. The above extraction process is performed independently for each sample subset, and the extraction results are only applicable to the spatial region and time period corresponding to that sample subset. At the same time, the number of features extracted or the information retention standard (such as a preset percentage of the total amount of original information retained after extraction) can be preset to ensure that the number of key feature factors extracted from each sample subset is appropriate and the information is sufficient.

[0062] The above scheme is a general feature extraction framework, suitable for application scenarios where the original feature parameters of different spatiotemporal categories have varying degrees of correlation. Its core advantage lies in its ability to adaptively select appropriate feature extraction methods and extraction quantities based on the data distribution characteristics of each sample subset. If, in practical applications, the data distribution of sample subsets under different spatiotemporal categories is relatively regular, the feature parameters have strong linear correlations, and it is necessary to ensure the consistency and reproducibility of the feature extraction process for each sample subset, then, as an optional embodiment based on the above embodiments, [further details are needed]. Figure 1 Step S106 shown can be achieved through the following steps, which are explained in detail below.

[0063] All multidimensional feature parameters corresponding to the concentration increase and decrease segments of the spatial region are collected separately, serving as sample subsets for the spatial region in the concentration increase and decrease segments respectively. Principal component analysis is performed on each sample subset to obtain multiple principal component factors, and the variance contribution value of each principal component factor in the sample subset is calculated. The total value of all variance contribution values ​​in the sample subset is calculated, and the cumulative contribution value of the current principal component factor is accumulated from the first variance contribution value to the current variance contribution value in descending order of variance contribution value. The ratio of the cumulative contribution value to the total value is used as the cumulative contribution rate of the current principal component factor. When the cumulative contribution rate reaches the preset contribution threshold for the first time, the first principal component factor corresponding to the first variance contribution value to the current principal component factor is used as the key feature factor corresponding to the sample subset in descending order of variance contribution value.

[0064] Specifically, based on the 2k combined categories obtained in the preceding steps (k being the number of categories in the spatial region), for each combined category, feature vectors of all grid points belonging to that spatial region and whose monitoring time falls within that time period are selected from the multidimensional feature parameter set. This constitutes a sample subset for that combined category. Principal component analysis is performed on each sample subset. The original feature data in the sample subset is standardized to eliminate dimensional differences between the feature parameters. Then, the covariance matrix of the standardized data is calculated, and the eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved. After the eigenvalues ​​are sorted from largest to smallest, the eigenvector corresponding to each eigenvalue is a principal component factor. Thus, multiple principal component factors for that sample subset are obtained. These principal component factors are uncorrelated with each other. Arranged in descending order of variance contribution value, the variance contribution value is the eigenvalue corresponding to the principal component factor. Its magnitude reflects the degree of contribution of the principal component factor to the total variance of the original data; a larger value indicates that the principal component factor carries more original information. The sum of the variance contribution values ​​of all principal component factors in the sample subset is calculated, and this sum equals the total variance of the original feature parameters. Starting from the first principal component factor, the variance contribution values ​​are accumulated sequentially in descending order. Each accumulated value yields the cumulative contribution value of the current principal component factor. Dividing the current cumulative contribution value by the sum of the variance contribution values ​​of all principal component factors gives the cumulative contribution rate of the current principal component factor. This ratio measures the degree to which the accumulated principal component factors retain the total information of the original data. Accumulation stops when the cumulative contribution rate first reaches or exceeds a preset contribution threshold. All principal component factors accumulated at this point are then considered as the key feature factors corresponding to this sample subset. The preset contribution threshold is typically set to 85%, but can be adjusted within the range of 80% to 95% depending on the data characteristics and accuracy requirements of the study area.

[0065] The sample subset refers to the set of grid point eigenvectors selected from the multidimensional feature parameter set that simultaneously satisfy specific spatial region constraints and specific time period constraints, used as the input dataset for principal component analysis. Principal component factors are comprehensive variables obtained by linearly transforming the original feature parameters through principal component analysis. Each principal component factor is a linear combination of all original feature parameters, with the combination coefficients determined by the eigenvectors of the covariance matrix. The principal component factors are linearly independent of each other. The variance contribution value refers to the eigenvalue corresponding to the principal component factor, used to measure the contribution of that principal component factor to the total variance of the original data; its value is equal to the variance of that principal component factor over all samples. The total value is the sum of the variance contribution values ​​of all principal component factors in the sample subset, equal to the total variance of the original feature parameters. The cumulative contribution value is the sum of the variance contribution values ​​of all principal component factors, arranged in descending order of variance contribution value, from the first principal component factor to the current principal component factor. The cumulative contribution rate is the ratio of the cumulative contribution value to the total variance contribution value of all principal component factors, used to measure the degree to which the first few principal component factors retain the total information of the original data. The preset contribution threshold refers to the minimum standard that the cumulative contribution rate should reach, typically set at 85%. It determines the number of principal component factors to retain. A higher threshold results in more retained principal component factors, less information loss, but reduced dimensionality compression; conversely, a lower threshold results in fewer retained principal component factors, higher dimensionality compression, but increased information loss. Key feature factors refer to the first m principal component factors corresponding to the first time the cumulative contribution rate reaches the preset contribution threshold. These factors are used to replace the original multidimensional feature parameters for subsequent inversion modeling. The specific value of m is determined by the data itself.

[0066] For example, taking the aforementioned sample subset of the high-value growth segment in the urban area as an example, this sample subset contains feature vectors of all grid points belonging to the high-value area of ​​the urban area and monitored between 06:00 and 17:00. Each feature vector contains 14 feature parameters. Principal component analysis is performed on this sample subset to obtain 14 principal component factors and their corresponding variance contribution values ​​(i.e., eigenvalues), and the sum of variance contribution values ​​is calculated. Arranged from largest to smallest variance contribution value, the variance contribution value of PC1 accounts for approximately 40% of the total, PC2 approximately 28%, and PC3 approximately 17.2%. The cumulative contribution value of the first three is the sum of their respective contribution values, with a cumulative contribution rate of (40% + 28% + 17.2%) = 85.2%. With a preset contribution threshold of 85%, the cumulative contribution rate of the first three principal component factors, 85.2%, reaches the threshold for the first time. Therefore, PC1, PC2, and PC3 are considered as key feature factors of the high-value growth segment in the urban area. Similarly, if the cumulative contribution rate of the top 5 principal component factors among the 14 principal component factors in the low-value growth segment of the suburbs reaches 86.3%, then 5 key feature factors are extracted for this category.

[0067] Through the above embodiments, adaptive selection of feature factors under each spatiotemporal category was achieved. Due to differences in underlying surface conditions and photochemical characteristics across different spatial regions and time periods, the dominant factors influencing ozone concentration vary across different combination categories. Principal component analysis (PCA) is used to process each combination category separately, compressing the original high-dimensional feature space into a low-dimensional principal component space. This ensures that the retained key feature factors can represent the original features with minimal information loss within their respective categories. Accumulation stops when the cumulative contribution rate reaches a preset threshold, avoiding dimensional redundancy caused by retaining too many principal component factors and preventing the loss of key information due to retaining too few principal component factors. This provides accurate and low-dimensional input features for the subsequent independent construction of inversion models for each combination category.

[0068] S107. Using the measured ozone concentration values ​​as dependent variables and key characteristic factors as independent variables, construct ozone concentration inversion models for each spatial region under the corresponding concentration increase and decrease segments.

[0069] Specifically, the key feature factors extracted in the preceding steps are organized according to their respective combination categories (spatial region × time period). Each combination category corresponds to a set of key feature factors, and each key feature factor is a comprehensive variable obtained after dimensionality reduction of the original features. Its dimension is independently determined based on the data distribution characteristics of each combination category. For each combination category, the key feature factor values ​​at various surface monitoring stations within that category are correlated and matched with the measured ozone concentration values ​​at the same location and time, forming several sample pairs consisting of independent variables (key feature factors) and dependent variables (measured ozone concentration values). Using the key feature factors in the sample pairs as model inputs and the measured ozone concentration values ​​as model outputs, a regression analysis method is used to establish a mapping function between the two. This function can estimate the corresponding ozone concentration value based on the input key feature factor values. The specific form of the model is selected based on the sample size and the complexity of the variable relationships: when the sample size is sufficient and the relationship between variables is approximately linear, a linear regression model can be used; when the sample size is large and the relationship between variables is complex, a nonlinear regression model or an ensemble learning model can be used. The models for each combination category are constructed independently, and the model parameters are optimized independently, ensuring that different combination categories do not affect each other. After the model is built, its performance on unseen data needs to be evaluated to ensure that the model has good generalization ability before it can be put into use for estimating ozone concentration at unknown grid points in this combination category.

[0070] For example, taking the aforementioned high-value growth segment in the urban area as an example, this combination category extracted three key feature factors (PC1, PC2, PC3). Measured ozone concentration values ​​Y were obtained from ground station data for each station belonging to this area and whose monitoring time fell within the growth segment. The key feature factor values ​​(PC1, PC2, PC3) of each station location were paired with the measured value Y to form a sample set. Each sample was (PC1, PC2, PC3) → Y. A model Y = β0 + β1 × PC1 + β2 × PC2 + β3 × PC3 was established using multiple linear regression, and the regression coefficients were estimated using the least squares method. Reserved test samples (approximately 20% of the total samples) that were not used in training were input into the model, and the error between the estimated and measured values ​​was calculated to evaluate the model's accuracy. The model's root mean square error was verified to be 6.126 μg / m³, with an average accuracy of 90%, meeting the usage requirements. The other combination categories independently underwent the same modeling process to obtain their respective inversion models.

[0071] The above scheme is a general modeling framework, suitable for application scenarios where the sample size of each combination category is sufficient and the data quality is good. Its core advantage lies in its ability to flexibly select appropriate regression modeling methods based on the actual distribution characteristics of the data for each combination category. If the data distribution of each combination category is relatively regular in practical applications, it is necessary to ensure that the construction process of each model has a consistent degree of standardization. Based on the above embodiments, as an optional embodiment, for... Figure 1 Step S107 shown can be achieved through the following steps, which are explained in detail below.

[0072] The key feature factors corresponding to each sample subset are matched with the measured ozone concentration values ​​of the corresponding spatial regions to construct inversion sample sets for each sample subset. Each inversion sample set is divided into a training sample set and a test sample set according to a preset ratio. Based on the feature value distribution of the key feature factors corresponding to each sample in the training sample set, discrete samples are removed from the training sample set to obtain the removed training sample set. Using the key feature factors in each removed training sample set as input variables and the corresponding measured ozone concentration values ​​as output variables, model fitting training is performed to obtain the inversion model corresponding to each sample subset. The key feature factors in the test sample set are input into the corresponding inversion model to obtain the estimated ozone concentration values ​​for the test sample set. The root mean square error and average accuracy are calculated based on the measured ozone concentration values ​​in the test sample set. When the root mean square error and average accuracy meet the preset accuracy requirements, the corresponding inversion model is used as the ozone concentration inversion model for the sample subset.

[0073] Specifically, for each combination category (spatial region × time period), key feature factor values ​​are extracted for each surface station location within that category. Ozone concentration values ​​for the same station at the same time period are obtained from measured ozone concentration values. These two values ​​are then correlated spatially and temporally to form a one-to-one corresponding sample pair (key feature factor vector → measured ozone concentration value). All sample pairs constitute the inversion sample set for that combination category. Each inversion sample set is divided into a training sample set and a test sample set according to a preset ratio, typically 80% training and 20% test. That is, 80% of the samples in the inversion sample set are randomly selected for model training, and the remaining 20% ​​are used for model validation. The selection process should ensure that the samples in the training and test sets are representative in terms of spatial region and time period. Statistical analysis is performed on the feature values ​​of each key feature factor in the training sample set, calculating the mean and standard deviation of each key feature factor. Samples whose feature values ​​deviate from the mean by more than three times the standard deviation are considered discrete samples and are removed to avoid interference from abnormal samples in model training.

[0074] Using key feature factors from each removed training sample set as input variables and the corresponding measured ozone concentration as output variables, model fitting training is performed to obtain inversion models corresponding to each sample subset. Model fitting training is implemented using algorithms such as multiple linear regression or random forest regression, aiming to minimize the error between predicted and measured values ​​on the training sample set. The model parameters are iteratively optimized until convergence. Key feature factors from the test sample set are input into the corresponding inversion model to obtain the estimated ozone concentration for the test sample set. This estimated value is the model's predicted output on unseen data. The root mean square error (RMSE) and average precision (AMP) are calculated by combining the measured ozone concentration values ​​from the test sample set. The RMSE measures the average deviation between the estimated and measured values, while the AMP measures the relative accuracy of the estimated value relative to the measured value. When the RMSE is less than a preset threshold and the AMP is greater than a preset threshold, the model accuracy is deemed to meet the requirements. The corresponding inversion model is then used as the ozone concentration inversion model for that sample subset and is used to predict the ozone concentration at unknown grid points under that combination category.

[0075] Wherein, the inversion sample set refers to a set of one-to-one sample pairs composed of key feature factors and corresponding measured ozone concentration values, each sample pair comprising a key feature factor vector and a measured ozone concentration scalar, which is used as the basic dataset for model construction. The preset ratio refers to a pre-set allocation ratio of the number of samples when dividing the training set and the test set, usually the training set accounts for 80% and the test set accounts for 20%, and this ratio can be adjusted within the range of 70%-90% for the training set and 10%-30% for the test set according to the total number of samples. The higher the proportion of the training set, the more sufficient the model training but the fewer the verification samples; the higher the proportion of the test set, the more sufficient the verification but the fewer the training samples. The training sample set refers to a subset of samples divided from the inversion sample set for model parameter training, whose sample size usually accounts for the majority of the inversion sample set, and is used for learning the mapping relationship from input variables to output variables through iterative optimization. The test sample set refers to a subset of samples divided from the inversion sample set for model accuracy verification, whose samples do not participate in model training, and is used to evaluate the generalization performance of the model on unseen data. The discrete sample refers to a sample in the training sample set where the feature value of any key feature factor deviates from the mean value of this factor by more than three times the standard deviation. The feature value distribution of this sample is significantly different from that of other samples, which may introduce noise to interfere with model training, and removing such samples can make the training set more representative. The root mean square error refers to the square root of the mean of the squares of the deviations between the estimated values and the measured values on the test sample set, which is used to measure the average deviation amplitude between the estimated values and the measured values. A smaller value indicates higher model accuracy, and the acceptable preset threshold is usually set according to the order of magnitude of ozone concentration, for example, 5-15 μg / m³. The average accuracy refers to the average value of the relative degree of the deviations between the estimated values and the measured values on the test sample set. A larger value indicates higher model accuracy, and the acceptable preset threshold is usually 85%-95%. The preset accuracy requirement refers to a pre-set qualified standard for model accuracy, including the dual conditions that the root mean square error is less than the preset error threshold and the average accuracy is greater than the preset accuracy threshold, and the model is determined to be qualified only when both conditions are met at the same time.

[0076] For example, taking the aforementioned high-value growth segment in the urban area as an example, this category matched 200 station sample pairs. Randomly divided at an 80% / 20% ratio, this resulted in 160 training samples and 40 test samples. Statistical analysis was performed on the eigenvalues ​​of three key feature factors in the 160 training samples, calculating the mean and standard deviation of each factor. A sample was considered a discrete sample if its eigenvalue for any key feature factor deviated from the mean by more than three times the standard deviation. Assuming five discrete samples were removed, 155 valid training samples were obtained. Using PC1, PC2, and PC3 as input variables and the measured ozone concentration as the output variable, multiple linear regression was used to train the model, resulting in the inversion model Y=β0+β1×PC1+β2×PC2+β3×PC3. The PC1, PC2, and PC3 values ​​of the 40 test samples were input into this model to obtain 40 estimated ozone concentration values. The root mean square error and average precision were calculated by comparing these values ​​with the measured values ​​from the test samples. If the preset accuracy requirement is RMSE < 10 μg / m³ and average accuracy > 85%, the calculated RMSE is 6.126 μg / m³ and the average accuracy is 90%, which meets the preset requirements. Therefore, this model is used as the ozone concentration inversion model for the high-value growth segment of the urban area.

[0077] Through the above embodiments, the separation of the training set and the test set avoids model overfitting, enabling the test accuracy to truly reflect the model's generalization ability; the removal of discrete samples in the training set is based on statistical methods to identify and exclude outliers, improving the quality of training data; the root mean square error measures the estimation deviation from the perspective of absolute error, while the average precision measures the estimation reliability from the perspective of relative error, and the two indicators together constitute a comprehensive evaluation of the model accuracy; the model is put into use only after the accuracy meets the standard, ensuring the reliability of subsequent inversion output.

[0078] S108. Input the real-time key feature factors of the target area in each spatial region into the corresponding ozone concentration inversion model, output the ozone concentration inversion value in raster format, and generate the spatial distribution result of ozone concentration in the target area.

[0079] Specifically, based on the established and validated ozone concentration inversion models for each spatiotemporal category, the first step is to determine the multidimensional feature parameter values ​​corresponding to each spatial location within the target region at the current inversion time. Then, based on the spatial region to which each location belongs and the time period in which the current inversion occurs, the model category corresponding to each spatial location is determined. The multidimensional feature parameters are then projected and transformed using the feature extraction parameters corresponding to that category (i.e., the load matrix or feature selection rules saved during the modeling phase) to generate real-time key feature factors for each spatial location. Finally, model instances matching the current spatial region and time period affiliation are selected from the established and validated inversion model set. The model inputs real-time key feature factors for calculation, yielding the ozone concentration inversion value for that spatial location. The matching and calculation process is performed independently for each spatial location, without affecting each other. After calculations are completed for all spatial locations, the inversion values ​​are arranged into a two-dimensional matrix according to the row and column order of each location. Location information is assigned based on spatial geographic coordinates, forming geocoded raster data. The value of each pixel in this raster data represents the estimated ozone concentration at that location at the current moment. All pixels together constitute a complete spatial distribution result of ozone concentration, which is directly output in raster format for visualization, spatial statistical analysis, or pollution hotspot identification. This can be achieved through the following steps, detailed below.

[0080] Obtain the multidimensional feature parameters corresponding to each spatial region at the current inversion time; based on each spatial region and the time period in which the current inversion time is located, determine the real-time key feature factors of each spatial region at the current inversion time; based on each spatial region and the time period in which the current inversion time is located, determine the target ozone concentration inversion model that matches the corresponding spatial region, input the corresponding real-time key feature factors into the target ozone concentration inversion model, and calculate all ozone concentration inversion values ​​in raster format under the spatial region; combine the ozone concentration inversion values ​​according to the spatial geographic coordinates of each spatial region to generate the spatial distribution results of ozone concentration in the target region.

[0081] Specifically, the process involves obtaining multidimensional feature parameters for each spatial region at the current inversion time. A spatial region is a sub-region defined by cluster analysis in the previous steps. Each region consists of multiple spatial locations, and each location within a region corresponds to a set of independent multidimensional feature parameters. For each spatial region, the feature extraction parameters for the current time period (i.e., the load matrix or feature selection rules saved during the modeling phase) are used to perform projection transformations on the multidimensional feature parameters of each spatial location within that region, generating real-time key feature factors for each spatial location at the current inversion time. Simultaneously, each spatial region has independently constructed inversion models corresponding to different time periods in the aforementioned steps. Based on the time period to which the current time belongs (concentration increase segment or concentration decrease segment), the corresponding model instance is selected as the target model from the model set for that region. The corresponding real-time key feature factors are input into the target ozone concentration inversion model to calculate all ozone concentration inversion values ​​in raster format under the spatial region. That is, each spatial location in the region is calculated one by one to obtain the ozone concentration estimate corresponding to each location. The calculation results of all locations are the pixel values ​​in the same raster format data layer. The ozone concentration inversion values ​​are combined according to the spatial geographic coordinates of each spatial region. The inversion results of different regions are stitched together according to the actual geographic coordinates to make adjacent regions seamlessly connected in space, generating the spatial distribution result of ozone concentration covering the entire target region.

[0082] Among them, the real-time key feature factor refers to the comprehensive variable value obtained after projection transformation of the multidimensional feature parameters of each spatial location within the spatial region at the current inversion time. The projection transformation uses the loading matrix saved during the modeling stage to map the multidimensional feature parameters from the original feature space to the principal component space. The target ozone concentration inversion model refers to an inversion model instance selected from the model set of each spatial region that is applicable to the current time period. Each model instance in the model set corresponds to a specific time period. The raster format refers to the digital format that stores spatial data in regular grid cells. Each grid cell records the value at that location, and all inversion results within the same spatial region together constitute the raster data layer of that region.

[0083] For example, suppose three spatial regions have been defined by clustering (high-value urban area, suburban transition area, and low-value suburban area). Each region contains two inversion models: one for increasing concentration and one for decreasing concentration, resulting in a total of six model instances. The current inversion time is 10:00 AM on a certain day, which is in the concentration increasing phase. For the high-value urban area, obtain the 14 feature parameter values ​​of all spatial locations within the region at 10:00 AM. Using the loading matrix saved when modeling the region in the increasing phase, perform projection transformation on the feature vectors of each location to generate real-time key feature factors for each location. Select the inversion model corresponding to the increasing phase from the model set of the high-value urban area as the target model. Input the real-time key feature factors of each location into the model sequentially to calculate the ozone concentration inversion value for each location, forming a raster data layer for the high-value urban area. Perform the same operation for the suburban transition area and the low-value suburban area to obtain their respective raster data layers. Finally, stitch the raster data layers of the three regions together according to their respective geographic coordinates to generate ozone concentration spatial distribution raster data covering the entire target area.

[0084] Through the above embodiments, adaptive spatial continuous inversion by region and time period is achieved. Each spatial region independently acquires multidimensional feature parameters at the current moment, independently performs projection transformation to generate real-time key feature factors, and independently matches and calculates the corresponding target model, enabling each region to obtain accurate inversion results based on its own feature distribution and time period affiliation. All spatial locations within each region are calculated one by one, and finally, the spatial distribution results covering the entire region are output in raster format, which can be directly used for subsequent ozone pollution spatial distribution feature analysis, high-pollution area identification, and governance decision support.

[0085] The following describes the regional and time-segmented ozone concentration remote sensing inversion system in the embodiments of this invention from the perspective of hardware processing. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 This is a schematic diagram of the structure of a remote sensing inversion system for ozone concentration by region and time period provided in an embodiment of this application.

[0086] It should be noted that, Figure 4 The structure of the ozone concentration remote sensing inversion system shown in the illustration is merely an example and should not impose any limitations on the functionality and scope of application of the embodiments of this application.

[0087] like Figure 4As shown, the regional and time-segmented ozone concentration remote sensing inversion system includes a central processing unit 201, which can perform various appropriate actions and processes based on programs stored in read-only memory 202 or programs loaded from storage section 208 into random access memory 203, such as executing the methods described in the above embodiments. The random access memory 203 also stores various programs and data required for system operation. The central processing unit 201, read-only memory 202, and random access memory 203 are interconnected via bus 204. Input / output interface 205 is also connected to bus 204.

[0088] The following components are connected to the input / output interface 205: an input section 206 including audio input devices, push-button switches, etc.; an output section 207 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; a storage section 208 including a hard disk, etc.; and a communication section 209 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 209 performs communication processing via a network such as the Internet. A drive 210 is also connected to the input / output interface 205 as needed. A removable medium 211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 210 as needed so that computer programs read from it can be installed into the storage section 208 as needed.

[0089] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 209, and / or installed from removable medium 211. When the computer program is executed by central processing unit 201, it performs the various functions defined in the present invention. It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0090] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0091] Specifically, the regional and time-segmented ozone concentration remote sensing inversion system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the regional and time-segmented ozone concentration remote sensing inversion method provided in the above embodiment.

[0092] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the regional and time-segmented ozone concentration remote sensing inversion system described in the above embodiments; or it may exist independently and not assembled into the regional and time-segmented ozone concentration remote sensing inversion system. The storage medium carries one or more computer programs, which, when executed by a processor of the regional and time-segmented ozone concentration remote sensing inversion system, cause the system to implement the regional and time-segmented ozone concentration remote sensing inversion method provided in the above embodiments.

[0093] The above description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the present invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. This application aims to cover any variations, uses, or adaptations disclosed in this invention, which follow the general principles of the present invention and include common knowledge or customary techniques in the art not described in this invention. The specification and embodiments are considered exemplary only, and the scope and spirit of this invention are defined by the claims.

Claims

1. A method for remote sensing inversion of ozone concentration by region and time period, characterized in that, include: The system acquires measured ozone concentration values, remote sensing image data, meteorological reanalysis data, and atmospheric composition remote sensing data from various ground stations within the target area over a preset historical time period, with the current monitoring time as the end time. The remote sensing image data, the meteorological reanalysis data and the atmospheric composition remote sensing data are preprocessed to obtain a preprocessed dataset, wherein all data in the dataset have the same spatial resolution. Surface feature parameters, meteorological feature parameters, and precursor concentration feature parameters are extracted from the dataset to form a multidimensional feature parameter set. The meteorological feature parameters and the precursor concentration feature parameters have a continuous impact on ozone concentration within the preset historical time period. Using the multidimensional feature parameter set as clustering samples, the target region is spatially clustered to obtain multiple spatial regions, and the daily variation curve of ozone concentration in each spatial region is plotted based on the measured ozone concentration values ​​in each spatial region. Based on the times of the lowest and highest concentration points in the daily ozone concentration variation curve, the monitoring period is divided into a concentration increase segment and a concentration decrease segment. Key feature factors for the concentration increase segment and the concentration decrease segment are extracted from the multidimensional feature parameter set corresponding to the spatial region, respectively. Using the measured ozone concentration values ​​as dependent variables and the key characteristic factors as independent variables, ozone concentration inversion models for each spatial region under the corresponding concentration increase and concentration decrease segments are constructed respectively. The real-time key feature factors of the target region in each of the spatial regions are input into the corresponding ozone concentration inversion model, and the ozone concentration inversion value in raster format is output to generate the spatial distribution result of ozone concentration in the target region.

2. The method according to claim 1, characterized in that, The preprocessing of the remote sensing image data, the meteorological reanalysis data, and the atmospheric composition remote sensing data to obtain a preprocessed dataset includes: The remote sensing image data is quantitatively preprocessed, and the meteorological reanalysis data and the atmospheric composition remote sensing data are format unified and anomaly processed to obtain preprocessed raster data. The highest spatial resolution among the raster data is taken as the target spatial resolution. The raster data are resampled to the target spatial resolution using a spatial interpolation method to obtain a resampled dataset, so that the coordinates of the raster data in the dataset are aligned on the spatial grid.

3. The method according to claim 1, characterized in that, The meteorological characteristic parameters include the average 2-meter temperature, average relative humidity, average 10-meter zonal wind speed, average 10-meter meridional wind speed, maximum boundary layer height, cumulative net solar radiation, and cumulative rainfall, which have a continuous impact on ozone concentration within the preset historical time period. The precursor concentration characteristic parameters include the average ozone column concentration and the average nitrogen dioxide column concentration that have a continuous impact on ozone concentration within the preset historical time period. The surface characteristic parameters include normalized vegetation index, nighttime light index, digital elevation model, slope, and aspect.

4. The method according to claim 1, characterized in that, The multidimensional feature parameter set is used as a clustering sample to spatially cluster the target region, resulting in multiple spatial regions, including: Based on the multidimensional feature parameter set, cluster samples for each grid point are determined. The cluster sample is a feature vector composed of the parameter values ​​of the surface feature parameters, meteorological feature parameters, and precursor concentration feature parameters of the corresponding grid point in the dataset. Based on the K-means clustering algorithm, initial cluster centroids are randomly selected. With minimizing the sum of squared errors within the cluster as the optimization objective, the initial cluster centroids are updated and the cluster samples are redistributed for multiple iterations until the sum of squared errors within the cluster reaches a convergent state. Plot a scatter plot of the sum of squared errors within the clusters as a function of the number of clusters. Determine the number of clusters corresponding to the point with the largest rate of change in the decrease in the scatter plot as the number of categories for spatial partitioning. Divide the target region into multiple spatial regions based on the number of categories.

5. The method according to claim 1, characterized in that, The step of extracting key feature factors from the multidimensional feature parameter set corresponding to the spatial region for the concentration increase segment and the concentration decrease segment includes: Collect all multidimensional feature parameters of the spatial region in the concentration increase segment and the concentration decrease segment respectively, and use them as sample subsets of the spatial region in the concentration increase segment and the concentration decrease segment respectively; Principal component analysis is performed on each of the sample subsets to obtain multiple principal component factors of the sample subsets, and the variance contribution value of each principal component factor in the sample subsets is calculated. Calculate the total value of all variance contribution values ​​in the sample subset, and accumulate the first variance contribution value to the current variance contribution value of the current principal component factor in descending order of variance contribution values ​​to obtain the cumulative contribution value of the current principal component factor; The ratio of the cumulative contribution value to the total value is taken as the cumulative contribution rate of the current principal component factor; When the cumulative contribution rate reaches the preset contribution threshold for the first time, the first principal component factor corresponding to the first variance contribution value to the current principal component factor are taken as the key feature factors corresponding to the sample subset, in descending order of the variance contribution values.

6. The method according to claim 5, characterized in that, The method involves constructing ozone concentration inversion models for each spatial region under corresponding concentration increase and decrease segments, using the measured ozone concentration values ​​as dependent variables and the key characteristic factors as independent variables. This includes: The key feature factors corresponding to each of the sample subsets are matched with the measured ozone concentration values ​​of the corresponding spatial regions to construct the inversion sample sets corresponding to each of the sample subsets. The inversion sample sets are divided into training sample sets and test sample sets according to a preset ratio; Based on the feature value distribution of the key feature factors corresponding to each sample in the training sample set, discrete samples in the training sample set are removed to obtain the removed training sample set. Using the key feature factors in each of the removed training sample sets as input variables and the corresponding measured ozone concentration as output variables, model fitting training is performed to obtain the inversion model corresponding to each of the sample subsets. The key feature factors in the test sample set are input into the corresponding inversion model to obtain the ozone concentration estimate corresponding to the test sample set. The root mean square error and average accuracy are calculated by combining the measured ozone concentration values ​​in the test sample set. When the root mean square error and the average accuracy meet the preset accuracy requirements, the corresponding inversion model is used as the ozone concentration inversion model for the sample subset.

7. The method according to claim 1, characterized in that, The step of inputting the real-time key feature factors of the target area in each of the spatial regions into the corresponding ozone concentration inversion model, outputting raster-formatted ozone concentration inversion values, and generating the spatial distribution results of ozone concentration in the target area includes: Obtain the multidimensional feature parameters corresponding to each of the spatial regions at the current inversion time; Based on each of the spatial regions and the time period in which the current time to be inverted is located, determine the real-time key feature factors of each of the spatial regions at the current time to be inverted; Based on each of the spatial regions and the time period in which the current time to be inverted is located, a target ozone concentration inversion model matching the corresponding spatial region is determined, and the corresponding real-time key feature factors are input into the target ozone concentration inversion model to calculate all ozone concentration inversion values ​​in raster format under the spatial region. By combining the ozone concentration inversion values ​​with the spatial geographic coordinates of each of the aforementioned spatial regions, the spatial distribution result of ozone concentration in the target region is generated.

8. A remote sensing inversion system for ozone concentration by region and time period, characterized in that, The regional and time-segmented ozone concentration remote sensing inversion system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the regional and time-segmented ozone concentration remote sensing inversion system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is run on the regional and time-segmented ozone concentration remote sensing inversion system, the regional and time-segmented ozone concentration remote sensing inversion system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the regional and time-segmented ozone concentration remote sensing inversion system, the regional and time-segmented ozone concentration remote sensing inversion system performs the method as described in any one of claims 1-7.

Images