A spatial matching method for cross-scale methane remote sensing data

Abstract

The application provides a spatial matching method for cross-scale methane remote sensing data, relates to the technical field of atmospheric remote sensing monitoring, and comprises the following steps: constructing a prior profile library and an average kernel set on a pressure grid, taking the prior profile library and the average kernel set as a benchmark to update the methane dry air column average mixing ratio of a pixel, calculating an optimal parameter set of an on-orbit point spread function of each sensor, and then mapping the updated methane dry air column average mixing ratio to a grid point of the pressure grid. The application introduces a common prior profile and an average kernel sensitivity benchmark, replaces the inversion results of different sensors to eliminate the difference in vertical detection capability, and outputs the methane dry air column average mixing ratio after vertical matching. Meanwhile, the optimal parameter set of the on-orbit point spread function of each sensor is calculated, and the optimal parameter set is converted into a discretized horizontal observation operator, so that the horizontal matching can be realized on each pixel of each sensor. The data standardization of the vertical sensitivity and the horizontal spatial response is realized.

Description

Technical Field

[0001] This invention relates to the field of atmospheric remote sensing monitoring technology, and more specifically, to a spatial matching method for cross-scale methane remote sensing data. Background Technology

[0002] Methane (CH4), the world's second-largest greenhouse gas, has a global warming potential approximately 28 times that of carbon dioxide on a centennial timescale. Multi-source satellite remote sensing provides crucial data support for global methane monitoring, including the European Tropomi sensor, Japan's GOSAT satellite, the US OCO series, and my country's Gaofen-5 satellite. However, the inherent heterogeneity of these multi-source data sources poses a significant obstacle to data fusion and application. Existing spatial matching techniques for cross-scale methane remote sensing data face the following key problems: 1) Mismatch of physical characteristics in the vertical dimension: Different sensors use different prior profiles and mean kernels in their inversion algorithms, resulting in significant differences in sensitivity to methane (CH4) in the atmospheric vertical layer. Furthermore, differences in observation geometry further affect the optical path of the signal. Directly comparing or fusing these physically inequivalent methane column concentration (XCH4) data introduces systematic bias.

[0003] 2) Spatial scale mismatch in the horizontal dimension: Sensors such as TROPOMI (5.5km × 7km quadrilateral), GOSAT (circle with a diameter of approximately 10.5km), and GOSAT-2 (circle with a diameter of approximately 9.7km) have vastly different spatial resolutions and varying observation footprint shapes. Existing mathematical resampling methods, such as bilinear interpolation and nearest-neighbor interpolation, smooth out high-frequency spatial information such as methane point source emission plumes and completely ignore the crucial physical process of the sensor's point spread function (PSF), leading to severe representativeness errors and information distortion.

[0004] In summary, existing technologies urgently require a comprehensive solution that can achieve data standardization across multiple physical dimensions, including vertical sensitivity, horizontal spatial response, and temporal consistency. Summary of the Invention

[0005] The purpose of this invention is to provide a spatial matching method for cross-scale methane remote sensing data to improve the aforementioned problems. To achieve this objective, the technical solution adopted by this invention is as follows: This application provides a spatial matching method for cross-scale methane remote sensing data, including: A spatial matching method for cross-scale methane remote sensing data, characterized by comprising: Acquire multiple methane inversion datasets corresponding to at least two types of sensors. One methane inversion dataset includes methane inversion data of multiple first pixels measured by one of the sensors. The methane inversion data includes the average mixing ratio of methane dry air column, the prior profile of methane concentration, and the average kernel. On a preset pressure grid, a priori profile library is constructed based on a preset reference profile field to obtain the priori profile library; An average kernel set is obtained, the average kernel set including a first average kernel of a plurality of grid cells of the pressure grid, and a first average kernel is determined by a plurality of preset first sensors in the average kernel of the grid cells; Based on the methane inversion data, the prior profile library and the average kernel set, update the average methane dry air column mixing ratio of the first pixel to obtain the first methane dry air column mixing ratio. The first parameter set of the on-orbit point spread function of each sensor is calculated based on a preset hyperspectral remote sensing dataset, resulting in multiple first parameter sets; The average mixing ratio of the second methane dry air column is calculated based on multiple sets of the first parameter and multiple average mixing ratios of the first methane dry air column, resulting in multiple average mixing ratios of the second methane dry air column.

[0006] Furthermore, the construction of the prior profile library based on the preset reference profile field includes: The three-dimensional methane concentration field output by the preset atmospheric chemical transport model is used as the reference field, and multiple pressure layers of the pressure grid are set as vertical layers simulated by the atmospheric chemical transport model to obtain the reference profile field. According to the preset time window and spatial grid, the reference profile field is subjected to spatiotemporal mean processing to obtain the prior profile library. The prior profile library includes first prior profiles of multiple spatiotemporal units. The spatiotemporal unit includes the observation time and geographical location. One spatiotemporal unit corresponds to one grid unit.

[0007] Furthermore, obtaining the average kernel set includes: The average value of the average kernel of the multiple first sensors in the first grid cell is calculated to obtain the first average kernel of the first grid cell; Calculate the first average kernel of multiple grid cells to obtain multiple first average kernels; The average kernel set is constructed based on multiple first average kernels.

[0008] Furthermore, updating the average mixing ratio of the methane dry air column of the first pixel based on the methane inversion data, the prior profile library, and the average kernel set includes: The scaling factor of the first pixel is calculated based on the prior profile of the methane column concentration of the first pixel and the average mixing ratio of the methane dry air column. The a priori profile of methane column concentration of the first pixel and the mean kernel are interpolated onto the pressure grid to obtain the interpolated a priori profile of methane column concentration and the mean kernel of the first pixel. The first vertical profile is calculated based on the scaling factor of the first pixel and the interpolated prior profile of the methane column concentration. The first methane dry air column average mixing ratio of the first pixel is calculated based on the first vertical profile of the first pixel, the average mixing ratio of the methane dry air column, the second prior profile, and the second average kernel. The second prior profile is the first prior profile corresponding to the second grid cell in the prior profile library. The second average kernel is the first average kernel corresponding to the second grid cell in the average kernel set. The second grid cell is determined based on the observation time and geographical location of the first pixel.

[0009] Furthermore, the first set of parameters for calculating the on-orbit point spread function for each of the sensors includes: Based on the sensor, a scan is performed, and multiple second pixels that cover multiple field angles in a single scan are selected to obtain multiple second pixels; The apparent reflectance observation value of the second pixel is obtained, and a first data block that is consistent with the observation date of the second pixel and is covered by the footprint of the second pixel is cropped from the hyperspectral remote sensing dataset to obtain the first data block. The first data block includes multiple first apparent reflectance values, and one first data block corresponds to one second pixel. Construct a paired sample set, which includes multiple paired samples, and each paired sample includes a second cell and its corresponding first data block; Based on the paired sample set, the first parameter set of the on-orbit point spread function for each of the sensors is calculated to obtain multiple first parameter sets.

[0010] Furthermore, the first set of parameters for calculating the on-orbit point spread function of each sensor based on the paired sample set includes: The on-orbit diffusion function of the second sensor is constructed based on a preset two-dimensional elliptic super-Gaussian function, wherein the second sensor is one of the sensors; Based on multiple first paired samples, the on-orbit diffusion function is calculated to obtain multiple simulated values ​​under a preset second parameter set, resulting in a simulated value group. Each simulated value group includes multiple simulated values, each simulated value group corresponds to a second parameter set, and each simulated value corresponds to a first paired sample. The first paired sample is the paired sample corresponding to the second sensor in the paired sample set. Based on multiple first paired samples, the simulated value set of the on-orbit diffusion function under multiple second parameter sets is calculated to obtain multiple simulated value sets; The total sample residual is calculated based on the simulated value set and multiple first paired samples to obtain the total sample residual, and one total sample residual corresponds to one simulated value set. Based on multiple sets of simulated values ​​and multiple first paired samples, multiple total sample residuals are calculated to obtain multiple total sample residuals; The second parameter set corresponding to the first sample total residual is used as the first parameter set of the on-orbit diffusion function of the second sensor, and the first sample total residual is the minimum sample total residual.

[0011] Furthermore, the calculation of the average mixing ratio of the second methane dry air column at multiple grid points of the pressure grid includes: Calculate the first coordinates of the first grid point of the pressure grid in the pixel coordinate system to obtain the first coordinates; Substitute the first parameter set corresponding to the first coordinate and the third pixel into the on-orbit point diffusion function to obtain the spatial response weight. One spatial response weight corresponds to one third pixel, and the third pixel is the first pixel covering the first grid point. Based on the first coordinate and the plurality of third pixels, a plurality of spatial response weights are calculated to obtain a plurality of spatial response weights; The average mixing ratio of the second methane dry air column at the first grid point is calculated based on the multiple spatial response weights and the average mixing ratio of the first methane dry air column at the multiple third pixels, thus obtaining the average mixing ratio of the second methane dry air column at the first grid point.

[0012] Furthermore, the average mixing ratio of the second methane dry air column at the first grid point is expressed as: ; in, The average mixing ratio of the second methane dry air column at the first grid point. For the first The spatial response weights corresponding to the third pixel. For the first The average mixing ratio of the first methane dry air column of the third pixel.

[0013] Furthermore, after obtaining the average mixing ratio of multiple second methane dry air columns, the process further includes: The methane inversion dataset is divided into multiple time slices according to a preset time window. Within each time slice, a KD-tree is constructed using the two-dimensional geographic coordinates of the first pixel as the key. When it is necessary to find spatiotemporally adjacent matching data for a target observation, the relevant one or more time slices are first located based on the timestamp. A range query is performed in the KD-tree of each time slice, and data including the average mixing ratio of the second methane dry air column is returned.

[0014] The beneficial effects of this invention are as follows: This invention introduces a shared prior profile and average kernel sensitivity benchmark to map and replace the inversion results of different sensors, thereby eliminating differences in vertical detection capabilities and outputting the average mixing ratio of a vertically matched methane dry air column. Simultaneously, the optimal on-orbit point diffusion function parameters are calculated for each sensor and transformed into a discretized horizontal observation operator, which is applied to each pixel of each sensor to achieve horizontal matching. This achieves data standardization in both vertical sensitivity and horizontal spatial response.

[0015] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart illustrating a spatial matching method for cross-scale methane remote sensing data as described in an embodiment of the present invention. Figure 2 This is another flowchart illustrating a spatial matching method for cross-scale methane remote sensing data as described in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0020] Example 1: It should be noted that existing spatial matching techniques for cross-scale methane remote sensing data have the following key problems: how to achieve physical consistency and coordination of the vertical sensitivity characteristics of different sensors, and eliminate systematic biases caused by differences in the mean kernel and prior profile; and how to construct spatial response operators based on the physical characteristics of sensors to achieve true physical matching of cross-scale data rather than simple mathematical interpolation.

[0021] To address the aforementioned technical problems, this embodiment provides a spatial matching method for cross-scale methane remote sensing data.

[0022] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4, S5 and S6.

[0023] S1. Obtain multiple methane inversion datasets corresponding to at least two types of sensors. One methane inversion dataset includes methane inversion data of multiple first pixels measured by one of the sensors. The methane inversion data includes the average mixing ratio of methane dry air column, the prior profile of methane concentration, and the average kernel. Understandably, this embodiment begins with the input and standardization of raw, multi-scale remote sensing data. An example is the raw methane inversion dataset from the Tropospheric Monitoring Instrument (TROPOMI) and the Greenhouse Gases Observing Satellite (GOSAT).

[0024] Based on data standardization, this embodiment sequentially executes the following three core technical modules, aiming to systematically transform raw data with varying physical definitions and spatiotemporal characteristics into Analysis Ready Data (ARD) with a unified physical benchmark and complete uncertainty quantification: like Figure 2 As shown, step S100 is for vertical sensitivity adjustment, achieving physical vertical matching. This module introduces a shared prior profile and average kernel sensitivity benchmark to map and replace the inversion results of different sensors, eliminating differences in vertical detection capabilities and outputting the vertically matched column concentration. Step S200 is for spatial response operatorization, achieving physical horizontal matching. This module inverts the on-orbit high-resolution point spread function (PSF) of the sensor and uses PSF convolution for mapping and replacement, constructing an accurate horizontal observation operator for each pixel to solve the problem of inconsistency between spatial resolution and pixel footprint morphology. Step S300 is for efficient spatial indexing and retrieval based on KD-tree. This module constructs an efficient multidimensional index for large-scale datasets and achieves fast spatiotemporal query and matching through pruning mechanisms, optimizing data processing performance. Finally, this method outputs an ARD product containing the average mixing ratio of methane dry air column and the quantification results of uncertainty, and provides API services.

[0025] S2. On a preset pressure grid, a priori profile library is constructed based on a preset reference profile field to obtain the priori profile library. Specifically, step S2 includes: S21. Using the three-dimensional methane concentration field output by the preset atmospheric chemical transport model as the reference field, and setting multiple pressure layers of the pressure grid as vertical layers simulated by the atmospheric chemical transport model, the reference profile field is obtained. S22. According to the preset time window and spatial grid, the reference profile field is subjected to spatiotemporal mean processing to obtain the prior profile library. The prior profile library includes the first prior profile of multiple spatiotemporal units. The spatiotemporal unit includes the observation time and geographical location. One spatiotemporal unit corresponds to one grid unit.

[0026] S3. Obtain an average kernel set, wherein the average kernel set includes a first average kernel of a plurality of grid cells of the pressure grid, and a first average kernel is determined by a plurality of preset first sensors in the average kernel of the grid cells; Specifically, step S3 includes: S31. Calculate the average value of the average kernel of the multiple first sensors in the first grid cell to obtain the first average kernel of the first grid cell; S32. Calculate the first average kernel of multiple grid cells to obtain multiple first average kernels; S33. Construct the average kernel set based on multiple first average kernels.

[0027] S4. Based on the methane inversion data, the prior profile library and the average kernel set, update the average methane dry air column mixing ratio of the first pixel to obtain the first methane dry air column average mixing ratio. Specifically, step S4 includes: S41. Calculate the scaling factor of the first pixel based on the prior profile of the methane column concentration of the first pixel and the average mixing ratio of the methane dry air column. S42. Interpolate the prior profile of methane column concentration of the first pixel and the mean kernel onto the pressure grid to obtain the interpolated prior profile of methane column concentration and the mean kernel of the first pixel. S43. Calculate the first vertical profile based on the scaling factor of the first pixel and the interpolated methane column concentration prior profile; S44. Calculate the first methane dry air column average mixing ratio of the first pixel based on the first vertical profile of the first pixel, the average mixing ratio of the methane dry air column, the second prior profile, and the second average kernel. The second prior profile is the first prior profile corresponding to the second grid cell in the prior profile library. The second average kernel is the first average kernel corresponding to the second grid cell in the average kernel set. The second grid cell is determined based on the observation time and geographical location of the first pixel.

[0028] Understandably, this embodiment selects at least two types of methane remote sensing payloads with different observation characteristics for the input multi-source satellite products, such as: wide-swath medium / high-resolution payload TROPOMI, high-precision point or narrow-swath payload GOSAT, and other payloads with column-averaged methane inversion capabilities, such as GF-5 / AHSI. For each sensor, its methane satellite remote sensing inversion product is obtained, including the following fields: methane dry air column average mixing ratio, prior methane vertical profile, column averaging kernel (AK) vector, vertical pressure layer information, instrument observation and inversion uncertainty, pixel center latitude and longitude, corner latitude and longitude, observation timestamp, solar / observation geometry information, quality control marks, etc. Then, the geographic coordinates of each product are unified to the World Geodetic System 1984 (WGS84); and the observation time is uniformly converted to Coordinated Universal Time (UTC).

[0029] Please see Figure 2 Steps S1-S4 correspond to Figure 2 The main objective of step S100 is to adjust the average mixing ratio of methane dry air columns from different sensors to a common prior vertical profile and average nuclear sensitivity benchmark on a standardized vertical pressure grid, thereby achieving physical comparability of multi-source products in the vertical dimension.

[0030] First, a unified vertical calculation benchmark is defined as the basis for this step. During the data standardization phase, the prior profiles of methane column concentrations from each sensor and the column average kernel need to be interpolated to a unified pressure layer grid. The interpolated prior profile of methane column concentration is obtained above. and average nuclei In this embodiment, a sensor with a coarser vertical resolution, such as the 12-layer barometric grid used by Tropomi, is selected as the uniform grid. Other sensors are interpolated from finer vertical grids to this grid to reduce interpolation errors and ensure consistency in comparison.

[0031] Then, a shared prior vertical profile library and a shared average kernel are constructed and assigned. To eliminate the bias introduced by the differences in prior information of each inversion system, this embodiment constructs and uses a shared prior vertical profile library. First, the three-dimensional methane concentration field output by the high-resolution Atmospheric Chemistry Transport Model (CTM) is selected as the reference field, and the aforementioned unified pressure layer is set as the vertical layer output by the CTM simulation, thus obtaining the three-dimensional reference profile field. Secondly, the reference profile field is subjected to spatiotemporal mean normalization according to a preset spatiotemporal scale (e.g., monthly average, latitude zone) to generate a shared prior profile library. ,in For a specific spatiotemporal unit. Then, for any cell to be processed. Based on its observation time and geographical location In a shared prior profile library Match the nearest spatiotemporal unit and assign the corresponding shared methane concentration prior profile. Assigned to the corresponding grid cell of this spatiotemporal unit. Simultaneously, the average kernel of multiple preset sensors in this spatiotemporal unit is... After interpolating to a uniform vertical layer and taking the arithmetic mean, a common average kernel is obtained. This is also assigned to the grid cell. Specifically, the sensors used to calculate the shared average kernel are multi-source satellite sensors (such as Tropomi, GOSAT, etc.) participating in the cooperative inversion. Thus, the pixel is completed. Common methane concentration prior profile and shared average kernel The assignment, for other pixels, also utilizes a shared prior profile library. The system assigns corresponding prior profiles and averaging kernels to multiple pre-defined sensors to eliminate biases introduced by differences in prior information from various inversion systems. Furthermore, in the above formulas, the superscript (ref) represents reference, and (c) represents common.

[0032] Next, a common prior benchmark conversion is performed. This step converts the original inversion results of each pixel into the average mixing ratio of a methane dry air column under the same prior and average kernel benchmark.

[0033] First, the prior profiles of the methane dry air column average mixing ratio and methane column concentration provided by the original dataset are utilized. Calculate the ratio of the two as the scaling factor. : ; in, For the first Scaling factor per pixel, For the first Average mixing ratio of methane dry air column per pixel For the first The pixel in the first Prior profile of methane column concentration in the atmosphere scalar form, This represents the total number of vertical layers in the original dataset. This is a vertical hierarchical index, with values ​​ranging from 1 to... , For the first The first pixel corresponds to the first pixel in the original dataset. The thickness weight of the atmospheric layer can be obtained by the following formula:

[0034] in, For the first The pixel in the vertical layering The air pressure value at the interface. Specifically, when... hour, It is 0.

[0035] Then, the above scaling factor A priori profile of methane column concentration interpolated to a uniform pressure layer This yields an approximate true vertical profile, which can be represented as: ; in, For the first Approximate true vertical profile of each pixel.

[0036] Finally, using common reference parameters, the average mixing ratio of the methane dry air column provided by the original data is corrected to eliminate differences in instrument vertical sensitivity, thus obtaining the average mixing ratio of the methane dry air column under the common vertical reference: ; At this point, the output is... This eliminates the systematic differences between prior assumptions and vertical sensitivity in the original inversion, achieving physical comparability.

[0037] S5. Calculate the first parameter set of the on-orbit point spread function for each sensor based on a preset hyperspectral remote sensing dataset, and obtain multiple first parameter sets; Specifically, step S5 includes: S51. Based on the scanning observation geometry of the sensor, select multiple second pixels that cover multiple field angles in a single scan of the sensor to obtain multiple second pixels; S52. Obtain the apparent reflectance observation value of the second pixel, and cut out the first data block from the hyperspectral remote sensing dataset that is consistent with the observation date of the second pixel and is covered by the footprint of the second pixel, to obtain the first data block. The first data block includes multiple first apparent reflectance values, and one first data block corresponds to one second pixel. It is understood that in this embodiment, "cropping" from a high spatial resolution hyperspectral remote sensing dataset, such as GF-5 / AHSI, refers to extracting a small patch of data from a specific location within that dataset. The radiance data from the hyperspectral remote sensing dataset is obtained and converted into apparent reflectance based on preset solar irradiance and observation geometric parameters; alternatively, the apparent reflectance product provided by the dataset can be directly obtained. The corresponding conversion formula for radiance to apparent reflectance is as follows: ; in, Pi; This represents the apparent reflectance at the top of the atmosphere (TOA Reflectance). It is a dimensionless value (typically between 0 and 1) that represents the proportion of sunlight reflected by the Earth's atmosphere. Indicates the satellite sensor in the band The observed spectral radiance is a raw physical quantity directly measured by the satellite, and its unit is usually 1. ; The Earth-Sun distance at the time of observation is expressed in astronomical units (AU). Indicates band Mean Solar Exoatmospheric Irradiances at the top of the atmosphere is a constant for a specific wavelength range, representing the amount of energy flux projected by the sun to the top of the atmosphere at the average Earth-Sun distance. The solar zenith angle at the time of observation.

[0038] S53. Construct a paired sample set, the paired sample set including multiple paired samples, and each paired sample including a second cell and its corresponding first data block; S54. Calculate the first parameter set of the on-orbit point spread function for each of the sensors based on the paired sample set, to obtain multiple first parameter sets.

[0039] Specifically, step S54 includes: S541. Construct the on-orbit diffusion function of the second sensor based on a preset two-dimensional elliptic super-Gaussian function, wherein the second sensor is one of the sensors; S542. Calculate multiple simulated values ​​of the on-orbit diffusion function under a preset second parameter set based on multiple first paired samples to obtain a simulated value group. One simulated value group includes multiple simulated values. One simulated value group corresponds to one second parameter set. One simulated value corresponds to one first paired sample. The first paired sample is the paired sample corresponding to the second sensor in the paired sample set. S543. Calculate the simulated value set of the on-orbit diffusion function under multiple second parameter sets based on multiple first paired samples, and obtain multiple simulated value sets; S544. Calculate the total sample residual based on the simulated value group and multiple first paired samples to obtain the total sample residual, where one total sample residual corresponds to one simulated value group; S545. Calculate multiple total sample residuals based on multiple sets of simulated values ​​and multiple first paired samples to obtain multiple total sample residuals; S546. The second parameter set corresponding to the total residual of the first sample is used as the first parameter set of the on-orbit diffusion function of the second sensor, and the total residual of the first sample is the minimum total residual of the sample.

[0040] S6. Calculate the average mixing ratio of the second methane dry air column at multiple grid points of the pressure grid based on multiple sets of the first parameter and multiple average mixing ratios of the first methane dry air column, and obtain multiple average mixing ratios of the second methane dry air column.

[0041] Specifically, step S6 includes: S61. Calculate the first coordinates of the first grid point of the pressure grid in the pixel coordinate system to obtain the first coordinates; S62. Substitute the first parameter set corresponding to the first coordinate and the third pixel into the on-orbit point diffusion function to obtain the spatial response weight. One spatial response weight corresponds to one third pixel. The third pixel is the first pixel covering the first grid point. S63. Calculate multiple spatial response weights based on the first coordinates and multiple third pixels to obtain multiple spatial response weights; S64. Calculate the second methane dry air column average mixing ratio of the first grid point based on the multiple spatial response weights and the first methane dry air column average mixing ratio of the multiple third pixels, and obtain the second methane dry air column average mixing ratio of the first grid point.

[0042] Understandably, please refer to Figure 2 Steps S5 and S6 correspond to Figure 2Step S200 primarily aims to construct a high- and low-resolution spatiotemporal pairing sample set covering different scanning geometries, and to perform a global empirical inversion of the on-orbit point spread function (PSF) of the medium- and low-resolution greenhouse gas sensor. This method aims to calculate a set of optimal PSF parameters suitable for the sensor and transform them into a discretized horizontal observation operator, which is then applied to each pixel of the sensor to achieve horizontal matching. In other words, based on these parameters, a discretized spatial response weight is constructed. This can be understood as the proportion of contribution of each small pixel with fine resolution within a large, coarse-resolution pixel to the total signal; the set of these weights constitutes the aforementioned discretized horizontal observation operator.

[0043] First, high- and low-resolution spatiotemporal matching and pairing samples are constructed. Since the on-orbit PSF of the sensor is affected by factors such as optical system aging, attitude changes, and atmospheric scattering, deviations often occur compared to laboratory calibration. This method employs a strategy of using auxiliary high-resolution data to simulate low-resolution data, and inverts the actual PSF of the methane sensor by constructing high- and low-resolution paired samples. To approximate real surface and atmospheric conditions, a product with a significantly higher spatial resolution than the target sensor is typically selected as a reference. In this embodiment, the high-resolution apparent reflectance of GF-5 / AHSI is used as a truth surrogate.

[0044] For each sensor to be calibrated (GOSAT, TROPOMI), select multiple low-resolution pixels covering different field of view angles in a single scan, denoted as i=1,2,…,N; For each pixel i, obtain its apparent reflectance observation. Then, a block of GF-5 / AHSI local high-resolution apparent reflectance data that matches the observation date and is completely covered by the footprint of pixel i is cropped out. (m represents the number of high-resolution data points in the data block that match low-resolution pixel i). All The paired data of each pixel are aggregated to form a sample dataset for subsequent joint fitting. .

[0045] Furthermore, the parameters of the on-orbit point spread function (PSF) are fitted. Based on the paired samples mentioned above, spatial convolution is performed on high-resolution image patches to simulate low-resolution observations, and the difference between the simulated and real observations is used as the objective function to invert and fit the on-orbit PSF parameters.

[0046] For a specific sensor, assume its PSF shape is determined by a set of parameters. The decision is to utilize all samples. The information from these parameters collectively determines this set of parameters. Here, a rotatable two-dimensional elliptic hypergaussian function is used to describe the PSF: ; in, , where is the global parameter set to be inverted: Normalization factor; and The spatial scale represents the direction of the principal axis of the ellipse, and the characteristic widths of the spread function of the ellipse point in the directions of the two orthogonal principal axes (x-axis and y-axis) are respectively. and This represents the relative coordinates of the center of the high-resolution data grid point relative to the center of the low-resolution data cell. Represents the super-Gaussian exponent, controlling for tail decay; Relative coordinates After rotation angle The transformed coordinate values, specifically... and The relationship is as follows: ; ; For each sample in the sample set In the current parameters The following calculates its simulated value: ; Then, construct a global objective function that includes all samples, and find the set of parameters that minimizes the total residual of all samples: ; The final calculated optimal parameters This can represent the average on-orbit optical characteristics of the sensor.

[0047] Furthermore, it is understandable that GOSAT has a coarser resolution and larger pixel size, and it is used to determine how many small, high-resolution GF5 / AHSI pixels are contained within it; then, the optimal PSF parameters are determined by comparing the spectral results of this large pixel location calculated under different weights.

[0048] Because the observation value of a single pixel in a low-resolution sensor (such as GOSAT or TROPOMI) is essentially the result of a weighted average of the true surface and atmospheric fields within its footprint, calculated using the sensor's PSF (Pressure Field Strength). Therefore, low-resolution pixels record a "spatially blurred true field." To infer the PSF, it is necessary to know the "high-resolution true field" that perfectly corresponds to the spatial range of the low-resolution pixel's true observation value. Therefore, it is necessary to pair low-resolution observations with matched high-resolution reference fields one by one to form the basic sample for PSF fitting.

[0049] Furthermore, a gridded mapping of the horizontal observation operator is performed, based on the optimal PSF parameters obtained from the previous stage for each target sensor. A spatial response weighting operator is constructed to physically weight sample low-resolution methane dry air column average mixing ratio data to generate high-fidelity gridded data products.

[0050] For any standard grid point within the target area and satellite pixels covering that grid point. First, calculate the relative coordinates of the grid points in the cell coordinate system. Then, substitute the optimal parameters. Calculate the spatial response weight of the cell to the grid point. This weight reflects the satellite pixels. What proportion of the observed values ​​should be physically distributed across the grid points? Location. For all covered grid points. Effective satellite pixel set Using the aforementioned spatial response weights, the weighted average method is employed to calculate the average methane dry air column mixing ratio at this grid point. : ; in, The satellite pixels after conversion in step S100 The average mixing ratio of methane dry air column.

[0051] Overall, this means that we use the sensor optical properties statistically learned from a large number of samples to correct the spatial response of each observation, ensuring the physical consistency of the inversion process.

[0052] After calculating the average methane-dry air column mixing ratio at each grid point of the pressure grid, further, as... Figure 2 As shown, step S300 performs efficient spatial indexing and retrieval based on KD-tree. Specifically, within each time slice, the two-dimensional geographic coordinates of the cell are used as the index key, and the methane mixing ratio and related quality identifiers calculated under a uniform pressure grid for that cell are used as the data value to construct a KD-tree spatial index.

[0053] This step targets large-scale datasets that have undergone physical coordination, enabling efficient spatiotemporal matching and the generation of the final ARD product.

[0054] First, a multidimensional index structure is constructed. To efficiently process long-term series data, the dataset is first pieced together according to fixed time windows (e.g., days). Within each time piece, the index is generated using the two-dimensional geographic coordinates of the pixels. Using the key, a KD-tree spatial index is constructed. A KD-tree is a spatially partitioned data structure that recursively divides data points into different dimensions to support efficient nearest neighbor search. In this way, when subsequent queries require methane concentration at any geographical location, the system can quickly retrieve physically consistent mixing ratio data from the KD-tree, thus supporting subsequent high-precision mapping or multi-source data fusion applications.

[0055] After constructing the KD-tree, fast pruning and spatiotemporal matching are performed. When it is necessary to find spatiotemporally adjacent matching data for a target observation, one or more relevant time slices are first located based on the timestamp. Subsequently, a range query is performed in the KD-tree of each time slice. The pruning mechanism of the KD-tree can quickly eliminate a large number of spatial partitions that do not intersect with the query area, and only perform precise distance and time difference calculations on a small number of candidate pixels, thereby filtering out the final matching set that meets the preset spatiotemporal threshold, thus optimizing query performance.

[0056] The final output is analysis-ready data, which generates a fully physically coordinated average mixing ratio of a methane dry air column and provides an Application Programming Interface (API) for range, nearest neighbor, and most recent time retrieval. Thus, this invention, through a series of interconnected physical coordination and data processing steps, successfully transforms multi-source heterogeneous methane remote sensing data into a high-quality analysis-ready data product with unified physical meaning and complete uncertainty characterization.

[0057] Understandably, to address the inconsistencies between the original priors and mean kernels of different sensors, this embodiment introduces a unified prior and performs an analytical transformation on the retrieved methane concentration profile. This transformation calibrates the methane concentration profile in the inversion results to a common prior and sensitivity characterization, achieving cross-source vertical consistency. Without requiring a complex re-inversion of the atmosphere, it achieves consistency between cross-source profile data and columnar data, significantly reducing systematic biases caused by differences in the original priors and mean kernels, and providing comparable and usable high-quality input for subsequent spatial matching and data assimilation.

[0058] Furthermore, traditional simple interpolation methods neglect the true spatial response characteristics of satellites in orbit, as well as the rotation and deformation of their footprints. This leads to the loss of plume details and the introduction of gridding representativeness errors when performing cross-scale matching of observational data. This scheme utilizes a high-resolution reference field (such as apparent reflectivity) from the same near-temporal region to empirically estimate the true on-orbit point spread function (PSF) of the target observations. This significantly reduces the representativeness error of the data in spatial scale matching, providing a physically consistent forward operator for subsequent data fusion and assimilation.

[0059] Meanwhile, for massive, irregular, and cross-sensor satellite observation data, this solution constructs a multi-dimensional joint index based on geographic coordinates and timestamps (e.g., KD-tree combined with time sharding). This index supports range queries, K-nearest neighbor queries (KNN), and joint retrieval of the most recent time, with typical query complexity optimized to [value missing]. level.

[0060] It is worth noting that, in terms of spatial response, in addition to the super-Gaussian distribution, other forms of nonparametric kernel learning or geometric optics can be used to directly construct spatial response operators and replace empirical convolution; In terms of retrieval indexing, KD-tree can be replaced by algorithms such as rectangular tree (R-tree) or ball tree (Ball-tree), and combined with pre-aggregation and multi-level caching to accelerate high-dimensional space data queries.

[0061] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0062] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A spatial matching method for cross-scale methane remote sensing data, characterized in that, include: Acquire multiple methane inversion datasets corresponding to at least two types of sensors. One methane inversion dataset includes methane inversion data of multiple first pixels measured by one of the sensors. The methane inversion data includes the average mixing ratio of methane dry air column, the prior profile of methane concentration, and the average kernel. On a preset pressure grid, a priori profile library is constructed based on a preset reference profile field to obtain the priori profile library; An average kernel set is obtained, the average kernel set including a first average kernel of a plurality of grid cells of the pressure grid, and a first average kernel is determined by a plurality of preset first sensors in the average kernel of the grid cells; Based on the methane inversion data, the prior profile library and the average kernel set, update the average methane dry air column mixing ratio of the first pixel to obtain the first methane dry air column mixing ratio. The first parameter set of the on-orbit point spread function of each sensor is calculated based on a preset hyperspectral remote sensing dataset, resulting in multiple first parameter sets; The average mixing ratio of the second methane dry air column is calculated based on multiple sets of the first parameter and multiple average mixing ratios of the first methane dry air column, resulting in multiple average mixing ratios of the second methane dry air column.

2. The spatial matching method for multi-scale methane remote sensing data according to claim 1, characterized in that... The construction of the prior profile library based on the preset reference profile field includes: The three-dimensional methane concentration field output by the preset atmospheric chemical transport model is used as the reference field, and multiple pressure layers of the pressure grid are set as vertical layers simulated by the atmospheric chemical transport model to obtain the reference profile field. According to the preset time window and spatial grid, the reference profile field is subjected to spatiotemporal mean processing to obtain the prior profile library. The prior profile library includes first prior profiles of multiple spatiotemporal units. The spatiotemporal unit includes the observation time and geographical location. One spatiotemporal unit corresponds to one grid unit.

3. The spatial matching method for cross-scale methane remote sensing data according to claim 2, characterized in that... The acquisition of the average kernel set includes: The average value of the average kernel of the multiple first sensors in the first grid cell is calculated to obtain the first average kernel of the first grid cell; Calculate the first average kernel of multiple grid cells to obtain multiple first average kernels; The average kernel set is constructed based on multiple first average kernels.

4. The spatial matching method for cross-scale methane remote sensing data according to claim 3, characterized in that... The step of updating the average mixing ratio of the methane dry air column of the first pixel based on the methane inversion data, the prior profile library, and the average kernel set includes: The scaling factor of the first pixel is calculated based on the prior profile of the methane column concentration of the first pixel and the average mixing ratio of the methane dry air column. The a priori profile of methane column concentration of the first pixel and the mean kernel are interpolated onto the pressure grid to obtain the interpolated a priori profile of methane column concentration and the mean kernel of the first pixel. The first vertical profile is calculated based on the scaling factor of the first pixel and the interpolated prior profile of the methane column concentration. The first methane dry air column average mixing ratio of the first pixel is calculated based on the first vertical profile of the first pixel, the average mixing ratio of the methane dry air column, the second prior profile, and the second average kernel. The second prior profile is the first prior profile corresponding to the second grid cell in the prior profile library. The second average kernel is the first average kernel corresponding to the second grid cell in the average kernel set. The second grid cell is determined based on the observation time and geographical location of the first pixel.

5. The spatial matching method for multi-scale methane remote sensing data according to claim 1, characterized in that... The first set of parameters for calculating the on-orbit point spread function for each of the sensors includes: Based on the sensor, a scan is performed, and multiple second pixels that cover multiple field angles in a single scan are selected to obtain multiple second pixels; The apparent reflectance observation value of the second pixel is obtained, and a first data block that is consistent with the observation date of the second pixel and is covered by the footprint of the second pixel is cropped from the hyperspectral remote sensing dataset to obtain the first data block. The first data block includes multiple first apparent reflectance values, and one first data block corresponds to one second pixel. Construct a paired sample set, which includes multiple paired samples, and each paired sample includes a second cell and its corresponding first data block; Based on the paired sample set, the first parameter set of the on-orbit point spread function for each of the sensors is calculated to obtain multiple first parameter sets.

6. The spatial matching method for cross-scale methane remote sensing data according to claim 5, characterized in that... The first parameter set for calculating the on-orbit point spread function for each of the sensors based on the paired sample set includes: The on-orbit diffusion function of the second sensor is constructed based on a preset two-dimensional elliptic super-Gaussian function, wherein the second sensor is one of the sensors; Based on multiple first paired samples, the on-orbit diffusion function is calculated to obtain multiple simulated values ​​under a preset second parameter set, resulting in a simulated value group. Each simulated value group includes multiple simulated values, each simulated value group corresponds to a second parameter set, and each simulated value corresponds to a first paired sample. The first paired sample is the paired sample corresponding to the second sensor in the paired sample set. Based on multiple first paired samples, the simulated value set of the on-orbit diffusion function under multiple second parameter sets is calculated to obtain multiple simulated value sets; The total sample residual is calculated based on the simulated value set and multiple first paired samples to obtain the total sample residual, and one total sample residual corresponds to one simulated value set. Based on multiple sets of simulated values ​​and multiple first paired samples, multiple total sample residuals are calculated to obtain multiple total sample residuals; The second parameter set corresponding to the first sample total residual is used as the first parameter set of the on-orbit diffusion function of the second sensor, and the first sample total residual is the minimum sample total residual.

7. The spatial matching method for multi-scale methane remote sensing data according to claim 1, characterized in that... The calculation of the average mixing ratio of the second methane dry air column at multiple grid points of the pressure grid includes: Calculate the first coordinates of the first grid point of the pressure grid in the pixel coordinate system to obtain the first coordinates; Substitute the first parameter set corresponding to the first coordinate and the third pixel into the on-orbit point diffusion function to obtain the spatial response weight. One spatial response weight corresponds to one third pixel, and the third pixel is the first pixel covering the first grid point. Based on the first coordinate and the plurality of third pixels, a plurality of spatial response weights are calculated to obtain a plurality of spatial response weights; The average mixing ratio of the second methane dry air column at the first grid point is calculated based on the multiple spatial response weights and the average mixing ratio of the first methane dry air column at the multiple third pixels, thus obtaining the average mixing ratio of the second methane dry air column at the first grid point.

8. The spatial matching method for multi-scale methane remote sensing data according to claim 1, characterized in that... The average mixing ratio of the second methane dry air column at the first grid point is expressed as: ； in, The average mixing ratio of the second methane dry air column at the first grid point. For the first The spatial response weights corresponding to the third pixel. For the first The average mixing ratio of the first methane dry air column of the third pixel.

9. The spatial matching method for multi-scale methane remote sensing data according to claim 1, characterized in that... After obtaining the average mixing ratio of multiple second methane dry air columns, the method further includes: The methane inversion dataset is divided into multiple time slices according to a preset time window. Within each time slice, a KD-tree is constructed using the two-dimensional geographic coordinates of the first pixel as the key. When it is necessary to find spatiotemporally adjacent matching data for a target observation, the relevant one or more time slices are first located based on the timestamp. A range query is performed in the KD-tree of each time slice, and data including the average mixing ratio of the second methane dry air column is returned.

Images