A method and device for analyzing and supplementing atmospheric formaldehyde using domestic satellites in all time and space

Abstract

The present invention discloses a method and device for analyzing and completing atmospheric formaldehyde in a continuous manner across all time and space using a domestic satellite. The method comprises the following steps: obtaining formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained by inversion through observation of a domestic satellite, and constructing earth's time and space information data after matching and aligning these data; performing partial random masking on the formaldehyde concentration data in the earth's time and space information data and inputting the data into a neural operator designed according to the data type; completing the formaldehyde concentration data with the random mask through neural operator calculation; performing semi-supervised learning based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data, and optimizing the neural operator parameters; and analyzing and completing the observed formaldehyde concentration data using the neural operator with optimized parameters, thereby achieving efficient and reliable prediction of the observed formaldehyde concentration data.

Description

Technical Field

[0001] The present invention belongs to the technical field of environmental monitoring, and in particular relates to a method and device for analyzing and supplementing atmospheric formaldehyde continuously over the entire time and space period using a domestic satellite. Background Art

[0002] Formaldehyde (HCHO) is a significant reactive carbon gas in the atmosphere, and its concentration is considered a key indicator for assessing air quality and monitoring photochemical pollution. Accurately monitoring the spatiotemporal distribution of atmospheric formaldehyde is crucial for environmental science and atmospheric chemistry research. While traditional ground-based sampling is constrained by limited spatial coverage, emerging satellite remote sensing technologies offer new opportunities for global formaldehyde observations.

[0003] Hyperspectral satellites already have the capability to observe atmospheric gases like formaldehyde. However, some hyperspectral satellites still need to improve their spatiotemporal resolution and spectral accuracy. Due to operational limitations of some hyperspectral satellites, cloud cover, and fluctuating atmospheric conditions, the formaldehyde data obtained by these satellites often suffer from significant gaps and discontinuities. Even in areas where observations are available, the daily formaldehyde inversion products contain numerous pixels with noise and outliers. This data gap not only impacts daily monitoring but also significantly complicates trend analysis and source tracing of long-term data series.

[0004] While some methods attempt to fill data gaps through statistical interpolation or numerical model simulation and assimilation, these approaches often require additional assumptions or extensive auxiliary data and may not fully capture complex atmospheric chemical processes. Anthropogenic volatile organic compounds (AVOCs) and biogenic volatile organic compounds (BVOCs) are major sources of formaldehyde. Formaldehyde's chemical reactions involve complex thermodynamics and kinetics that are difficult to fully understand and simulate. Furthermore, these methods can suffer from high computational costs and low efficiency when processing large datasets.

[0005] Deep learning technology has made significant progress in data processing and pattern recognition, and has already demonstrated remarkable success in satellite data restoration. However, because formaldehyde data is stored in spherical coordinates, current research on formaldehyde is far less in-depth than that on nitrogen dioxide and ozone. Most importantly, there is a lack of comprehensive labels to enable supervised learning in neural networks.

[0006] The emergence of neural operators (NOs) has brought neural network fitting of partial differential equations to a new level of accuracy. For example, the Fourier neural operator (FNO) leverages the properties of the Fourier transform to combine a linear global integral operator with a nonlinear local activation function to learn complex nonlinear mapping relationships. This unique structure enables excellent performance when dealing with problems involving infinite-dimensional space, and is particularly suitable for operations in function space. The ability of neural operators is not only reflected in learning complex mappings and combining linearity and nonlinearity, but also in extracting and integrating physical knowledge, strong generalization capabilities, and the potential to adapt to new advances. It has demonstrated outstanding capabilities in multiple scientific and engineering problems, opening up new possibilities for the application of deep learning technology. Summary of the Invention

[0007] In view of the above, the purpose of the present invention is to provide a method and device for analyzing and completing the atmospheric formaldehyde concentration of domestic satellites in full time and space, using semi-supervised learning and neural operators to solve the problem of missing formaldehyde data, and to achieve high-precision analysis of the full spatial domain distribution of atmospheric formaldehyde concentration of domestic hyperspectral satellites, and obtain a seamless global concentration distribution.

[0008] To achieve the above-mentioned purpose of the invention, the embodiment provides a method for analyzing and completing the atmospheric formaldehyde content of a domestic satellite in a continuous manner over all time and space, comprising the following steps:

[0009] Obtain formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained from domestic satellite observation inversion, and align these data to construct Earth's spatiotemporal information data;

[0010] The formaldehyde concentration data in the Earth's spatiotemporal information data is partially randomly masked and then input into a neural operator designed according to the data type. The random mask is then used to complete the formaldehyde concentration data through neural operator calculation. Semi-supervised learning is then performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters.

[0011] The observed formaldehyde concentration data is analyzed and completed using a neural operator with optimized parameters to obtain the completed formaldehyde concentration data.

[0012] Preferably, the relevant substance emission data includes formaldehyde emission data and emission data of other relevant species that are interrelated with the chemical conversion process of formaldehyde in the atmosphere;

[0013] The human geography data includes population, leaf area index, and topographic data.

[0014] Preferably, the meteorological data includes temperature, humidity, pressure, and wind speed.

[0015] Preferably, the neural operator includes an encoder, a neural operator layer, and a decoder connected in sequence, wherein the neural operator layer adopts a spherical Fourier neural operator (SFNO), an orthogonal neural operator (ONO), a transolver or a general neural operator transformer (GNOT).

[0016] Preferably, the reconstruction error adopts mean square error, L2 norm or absolute error.

[0017] Preferably, when performing semi-supervised learning on the neural operator, the earth's spatiotemporal information data is coarsely sampled with a coarse resolution grid, and the neural operator is learned using the coarsely sampled sample data. At the same time, some random mask areas are filled with constants, and the null value areas are skipped when calculating the reconstruction error.

[0018] Preferably, when performing semi-supervised learning on the neural operator, after learning the neural operator using coarsely sampled sample data, fine sampling of fine-resolution grids is performed on the earth's spatiotemporal information data, and the neural operator is learned using the finely sampled sample data.

[0019] To achieve the above-mentioned purpose of the invention, the embodiment further provides a domestic satellite full-time and space continuous atmospheric formaldehyde analysis and supplement device, comprising:

[0020] The data acquisition and processing module is used to obtain formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained by domestic satellite observation inversion, and to match and align these data to construct Earth's spatiotemporal information data;

[0021] A model building and training module, which is used to randomly mask the formaldehyde concentration data in the Earth's spatiotemporal information data and input it into a neural operator designed according to the data type. The neural operator then calculates the formaldehyde concentration data based on the random mask. Semi-supervised learning is then performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters.

[0022] The parsing and completion module is used to parse and complete the observed formaldehyde concentration data using a neural operator with optimized parameters to obtain the completed formaldehyde concentration data.

[0023] To achieve the above-mentioned purpose of the invention, an embodiment also provides a computing device, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method.

[0024] To achieve the above-mentioned purpose of the invention, the embodiment also provides a computer-readable storage medium on which a program is stored. When the program is executed by the processor, the above-mentioned domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method is implemented.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] By purposefully introducing gaps in the dataset and using neural operators to learn and predict these gaps, the model can be trained without the need for real labels. This process does not rely on a complete formaldehyde concentration dataset, but instead makes full use of the data's inherent structure and pattern information to achieve efficient and reliable prediction of data gaps, providing strong support for environmental monitoring, weather forecasting, and climate change research. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 This is a flow chart of a method for analyzing and completing atmospheric formaldehyde in a domestic satellite in all time and space continuous manner provided by an embodiment;

[0029] Figure 2 This is a flowchart of a method for analyzing and completing atmospheric formaldehyde in a domestic satellite in all time and space continuous manner provided by an embodiment;

[0030] Figure 3 is a structural diagram of a neural operator model provided in an embodiment;

[0031] Figure 4 is a distribution diagram of the original formaldehyde tropospheric column concentration monitored by satellite provided in the embodiment;

[0032] Figure 5 This is the formaldehyde tropospheric column concentration distribution diagram after analysis and completion provided in the embodiment;

[0033] Figure 6 It is the effect verification of filling the vacancy formaldehyde tropospheric column concentration provided by the embodiment;

[0034] Figure 7 This is the ground-based MAX-DOAS instrument verification for reconstructing the formaldehyde tropospheric column concentration provided in the embodiment, wherein the left side shows the verification result of the original satellite data, and the right side shows the verification result after analysis and completion;

[0035] Figure 8The present invention is a structural diagram of a domestic satellite full-time and space continuous atmospheric formaldehyde analysis device provided in an embodiment. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0037] The inventive concept of the present invention is: when the existing technology reconstructs the spatiotemporal distribution of formaldehyde, the accuracy of the results is low, especially when the data is missing or incomplete. Based on this, in order to solve the technical problem of missing formaldehyde data in satellite observations, the embodiment of the present invention provides a domestic satellite full-time and space continuous atmospheric formaldehyde analysis method and device.

[0038] like Figure 1 and Figure 2 As shown, the embodiment provides a domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method, including the following steps:

[0039] S1, obtains formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained by inversion of domestic satellite observations, and aligns these data to construct Earth's spatiotemporal information data.

[0040] In an embodiment, observed formaldehyde concentration data is obtained and cleaned and standardized. Specifically, observation data can be obtained from a trace gas differential absorption spectrometer carried on a satellite, and outliers in the formaldehyde concentration data are removed and standardized before being sorted in chronological order.

[0041] In this embodiment, meteorological data is obtained from a high-resolution global reanalysis of meteorological fields, including wind speed, temperature, air pressure, humidity, and geopotential. This data is then interpolated to the same resolution as satellite data. The present invention uses wind speed, temperature, humidity, air pressure, and geopotential height at 1000 hPa, 850 hPa, 700 hPa, and 500 hPa as meteorological input.

[0042] In the embodiment, the relevant substance emission data includes formaldehyde emission data, wherein the formaldehyde emission includes natural sources (vegetation, wildfire emissions) and anthropogenic sources (industry, transportation, and daily life emissions). According to the characteristics of the study area, selecting a suitable emission inventory for replacement can more accurately reflect the formaldehyde emission situation in the region. When making the emission inventory, vegetation emissions, biomass combustion emissions, and anthropogenic source emissions of formaldehyde are added. At the same time, in the simulation of formaldehyde emissions and atmospheric chemistry, in addition to formaldehyde itself, it is also necessary to consider the emissions of other related species, such as volatile organic compounds (VOCs) emissions, nitrogen oxides (NOx ) and particulate matter emissions. Emission inventories of these species usually need to be used together with formaldehyde emission inventories because their chemical transformation processes in the atmosphere are interconnected and jointly determine the concentration and distribution of formaldehyde.

[0043] In this way, the real chemical processes in the atmosphere can be more accurately reflected, thereby improving the reliability and predictive ability of the simulation. These emission inventories are processed into multi-channel two-dimensional data so that they can be merged with meteorological data. The emission flux over the ocean is mostly 0 and is usually not considered a major emission source. When normalizing, the magnitude of the emission inventory needs to be considered to avoid a large number of low values ​​affecting the normalization effect. A constant can be multiplied at the same time to make the emission flux reach a relative distribution of 0-1000. At the same time, human geography data such as population, leaf area index, and altitude data can also be added and processed to the same resolution as auxiliary data.

[0044] The embodiment also performs data point alignment and matching on formaldehyde concentration data, meteorological data, related substance emission data, and human geography data according to time sequence and spatial position to obtain earth spatiotemporal information data.

[0045] S2, the formaldehyde concentration data in the earth's spatiotemporal information data is partially randomly masked and input into the neural operator model designed according to the data type. The formaldehyde concentration data of the random mask is completed through neural operator calculation, and semi-supervised learning is performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters.

[0046] In the embodiment, the constructed neural operator model consists of three main parts: encoder, neural operator layer and decoder. It is a mapping from input function space x to output function space y. Neural operator This mapping is learned by parameters θ:

[0047]

[0048] Among them, the encoder maps the input function u to a high-dimensional feature space, Indicates passing Processing of layer neural operator layers.

[0049] Among them, the neural operators in the neural operator layer convert the data into a spherical grid suitable for earth coordinates. The spherical Fourier neural operator (SFNO) can be used. SFNO takes advantage of the advantages of Fourier transform in processing periodic and global data, and is suitable for simulating continuous space systems such as the atmosphere. When designing a neural operator model, you can choose the most suitable operator network according to the needs of the specific problem. It is not limited to Fourier neural operators, and other operator networks can be selected. Neural operators such as orthogonal neural operators (ONO) and general neural operator transformers (GNOT) can also project and transform data into a spherical coordinate system to meet the spatial continuity characteristics of the three-dimensional atmosphere. When selecting and adjusting neural operators, researchers need to consider the characteristics of the data, the complexity of the model, the requirements for computing resources, and the goals of model performance, and reasonably select the number of features and layers of the model. Specifically, the designed neural operator model is such as Figure 3 As shown, it includes the earth's spatiotemporal information input layer, encoding layer, earth's spatiotemporal feature extraction layer, neural operator layer, nonlinear activation layer, decoding layer, and earth's spatiotemporal information output layer.

[0050] The goal of supervised learning for the neural operator model is to learn a data representation by reconstructing missing portions of the formaldehyde concentration data within the Earth's spatiotemporal information. During supervised learning, a portion of the formaldehyde concentration data is randomly selected and masked. The masking percentage can be adjusted based on the actual situation, for example, between 25% and 75%. The neural operator model learns to fill in the missing portions of the formaldehyde data in the original dataset. This process involves optimizing the parameters of the neural operator so that the model can accurately fill in the data.

[0051] The specific steps of supervised learning are:

[0052] (a) Initial Data Sampling: Earth's spatiotemporal data is first coarsely sampled, gridded at a resolution of 1°. This step aims to create a coarse geospatial grid for subsequent modeling and analysis. The 1° resolution balances the need for spatial detail and computational resources, while ensuring rapid convergence of the neural network and the ability to learn the chemical characteristics of the coarse-resolution physical and relevant emission data.

[0053] (b) Model training: The reconstruction error is used as the loss function to measure the difference between the completed formaldehyde concentration data output by the network and the formaldehyde concentration data in the masked area of ​​the original image. The reconstruction error can be the mean square error or the absolute difference. For a set of samples, the L2 loss (mean square error, MSE) can be expressed as:

[0054]

[0055] Where n is the number of samples, y iis the true value of the i-th sample, is the predicted value of the i-th sample.

[0056] During masking, areas requiring masking are filled with constants, and regions with empty values ​​are skipped when calculating the loss function. The model is trained on the coarse-resolution dataset for 100 epochs (an epoch in neural operator model training refers to a complete iteration through all samples in the dataset). This training period was chosen based on experimentation and validation to ensure that the neural operator model can converge quickly without overfitting, while quickly capturing key features in the data.

[0057] (c) Fine-data training: After 100 epochs of training on the coarse-resolution dataset, the Earth's spatiotemporal information data is fine-sampled to a fine-resolution grid, specifically gridded at a resolution of 0.1°. This step enables the model to further learn more detailed features, thereby improving the resolution and prediction accuracy of the neural operator model. The neural operator model is further trained on the fine data to further improve its performance and generalization ability. This step helps the neural operator model better understand subtle changes in the data, thereby improving its ability to fill in gaps.

[0058] (d) Model Evaluation: After retraining is complete, the neural operator model should be evaluated, including its performance and generalization ability on an independent test set. The predictive performance of the neural operator model should be evaluated on the test set using appropriate evaluation metrics (e.g., correlation coefficient, mean squared error). The evaluation results should demonstrate that the model can accurately predict and simulate the distribution of formaldehyde in the atmosphere.

[0059] S3, using the parameter-optimized neural operator model to parse and complete the observed formaldehyde concentration data to obtain the completed formaldehyde concentration data.

[0060] In this embodiment, the parameter-optimized neural operator model is applied to the missing formaldehyde data in the original dataset. The prediction results are post-processed to match the statistical characteristics and spatial distribution of the dataset, thereby obtaining a long-term, continuous, and accurate satellite formaldehyde column concentration dataset.

[0061] To verify the effectiveness of the above method, a specific experimental example was conducted, using formaldehyde column concentration data obtained from observations from 2018 to 2022. Due to factors such as cloud cover and satellite orbits, the original data contained approximately 30% missing and invalid values.

[0062] Data preprocessing: First, the original satellite-observed formaldehyde column concentration data was cleaned, obvious outliers were removed, and the data were arranged in daily chronological order. Then, the temperature, humidity, air pressure, and wind speed parameters from 2018 to 2022 were extracted from the meteorological field, and the resolution interpolation was consistent with the formaldehyde column concentration data observed by satellite. When constructing the emission inventory, the CAMS formaldehyde emissions and wildfire emission inventories were combined, including anthropogenic sources of formaldehyde, biomass burning, and vegetation emissions. In addition, population distribution, LAI vegetation index, and terrain altitude were extracted as auxiliary data. Finally, all the above data were gridded and matched to a uniform spatiotemporal resolution. All data were standardized to suit model fitting. Due to the low values ​​of the emission inventory, they were multiplied by 10 8 Then standardize.

[0063] Model Construction: The experimental example uses a 6-layer Spherical Fourier Neural Operator (SFNO) as the core. Each SFNO layer consists of a Fourier transform and an activation function. The input data first passes through a 3D convolution layer to merge the multi-dimensional features into a single-channel spherical data. The SFNO is then used to extract features, and finally a decoding layer outputs the prediction results. The Adam optimizer is used, with an initial learning rate of 1e. -3 , and the cosine annealing strategy is adopted. The loss function is the square error loss. During the training process, the batch size is 8 and the weight decay coefficient is 1e -5 .

[0064] Semi-supervised training: When sampling training data, raw satellite formaldehyde data and various other data are gridded at 1-degree resolution. The training set consists of data from 2018 to 2021, with 2022 serving as the validation set. During training and testing, 50% of the satellite-observed formaldehyde data points are randomly selected and masked to serve as target gaps for model learning. For the first 100 epochs, the model is pre-trained at 1-degree resolution, ultimately reducing the error between the model's predictions and observations on the test set to a low level. Subsequently, the satellite data and various auxiliary data are increased to a high resolution of 0.1 degrees. The model is trained on this high-resolution data for another 50 epochs to capture more detailed features, ultimately achieving seamless global atmospheric formaldehyde analysis.

[0065] The prediction performance of the model was evaluated on a global test set in 2022. The results showed that the model achieved good results in analyzing the spatial distribution of atmospheric formaldehyde. 2 The value is 0.58. The deviation between the model's predictions and the actual measured values ​​is significantly reduced. The following is the temporal and spatial interpolation effect of the model on the global formaldehyde distribution in a certain prediction. Figure 4 The original formaldehyde tropospheric column concentration distribution map obtained by inversion of the China Environmental Monitoring Satellite on July 18, 2023, is obtained through the weighted analysis and completion method of this embodiment. Figure 5 Global coverage results for the day. Figure 6 Verification of the effectiveness of filling the gap in formaldehyde tropospheric column concentrations showed a good correlation between the original observed concentrations and the analyzed concentrations, but the reconstruction results achieved significantly smoother spatial coverage. The observations from the first half of 2023 were reconstructed and verified against the ground-based MAX-DOAS instrument at the Beijing Institute of Atmospheric Physics. The left side shows the verification results of the original satellite data, while the right side shows the verification results after the analytical completion. The analytical completion results achieved better consistency.

[0066] It can be seen that the neural operator model of this embodiment well captures the concentration differences of formaldehyde between land and ocean, industrial areas and forest areas, and accurately reconstructs the missing parts of the original data.

[0067] This example demonstrates the effectiveness and accuracy of the proposed method for filling global satellite formaldehyde data. By leveraging prior knowledge such as auxiliary data and emission inventories, the model accurately learns the temporal and spatial distribution patterns of atmospheric formaldehyde concentrations. This method avoids reliance on supervised labels and can flexibly address large gaps in long-term data series, providing strong support for environmental monitoring and atmospheric chemistry research. In the future, this method could be extended to monitor other important atmospheric components and environmental factors.

[0068] like Figure 8 As shown, the embodiment also provides a domestic satellite full-time and space continuous atmospheric formaldehyde analysis device, including a data acquisition and processing module, a model construction and training module, and an analysis and completion module, wherein the data acquisition and processing module is used to obtain formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained by domestic satellite observation inversion, and match and align these data to construct the earth's space-time information data; the model construction and training module is used to partially randomly mask the formaldehyde concentration data in the earth's space-time information data and input it into a neural operator model designed according to the data type, and the random mask is calculated by the neural operator to complete the formaldehyde concentration data, and semi-supervised learning is performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters; the analysis and completion module is used to use the parameter-optimized neural operator model to analyze and complete the observed formaldehyde concentration data to obtain the completed formaldehyde concentration data.

[0069] It should be noted that the above-mentioned embodiments provide a full-time and space-time continuous atmospheric formaldehyde analysis and completion device. The division of the above-mentioned functional modules is used as an example to illustrate the full-time and space-time continuous atmospheric formaldehyde analysis and completion. The above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the terminal or server is divided into different functional modules to complete all or part of the functions described above. Furthermore, the full-time and space-time continuous atmospheric formaldehyde analysis and completion device provided in the above-mentioned embodiments and the full-time and space-time continuous atmospheric formaldehyde analysis and completion method embodiment are based on the same concept. The specific implementation process is detailed in the full-time and space-time continuous atmospheric formaldehyde analysis and completion method embodiment, and will not be repeated here.

[0070] Based on the same inventive concept, an embodiment further provides a computing device including a memory and one or more processors. The memory stores executable code. When the one or more processors execute the executable code, the computing device is used to implement the above-mentioned method for analyzing and completing atmospheric formaldehyde in a continuous manner over a domestic satellite in all time and space. The method specifically includes the following steps:

[0071] S1, obtain formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained by inversion of domestic satellite observations, and align these data to construct Earth's spatiotemporal information data;

[0072] S2: The formaldehyde concentration data in the Earth's spatiotemporal information data is partially randomly masked and then input into a neural operator model designed according to the data type. The neural operator model calculates and completes the formaldehyde concentration data with the random mask. Semi-supervised learning is performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters.

[0073] S3, using the parameter-optimized neural operator model to parse and complete the observed formaldehyde concentration data to obtain the completed formaldehyde concentration data.

[0074] The computing device provided in the embodiment, in addition to the processor and memory, also includes hardware required for other services such as internal bus, network interface, memory, etc. at the hardware level. The memory is a non-volatile memory, and the processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method described in S1-S3 above. Of course, in addition to software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0075] Based on the same inventive concept, an embodiment further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the above-mentioned domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method is implemented, specifically comprising the following steps:

[0076] S1, obtain formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained by inversion of domestic satellite observations, and align these data to construct Earth's spatiotemporal information data;

[0077] S2: The formaldehyde concentration data in the Earth's spatiotemporal information data is partially randomly masked and then input into a neural operator model designed according to the data type. The neural operator model calculates and completes the formaldehyde concentration data with the random mask. Semi-supervised learning is performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters.

[0078] S3, using the parameter-optimized neural operator model to parse and complete the observed formaldehyde concentration data to obtain the completed formaldehyde concentration data.

[0079] In the embodiment, computer-readable media includes permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data.

[0080] In short, in the global formaldehyde monitoring results of domestic satellites, due to the limitations of sensor layout and cloud obstruction problems, formaldehyde concentration data at certain locations may not be available. In order to solve this problem, the above-mentioned analytical completion scheme proposed in the present invention uses a neural operator model to analyze and model the existing satellite global formaldehyde data. The neural operator can capture the trend and change law of formaldehyde concentration at unknown locations based on the periodic characteristics of existing data. Then, a model is established in combination with the emission inventory and auxiliary data related to formaldehyde to fill in the missing data and obtain a global formaldehyde gap-free distribution result. Experimental results show that the method proposed in the present invention shows high performance in filling accuracy and stability, and can effectively provide complete data on global formaldehyde column concentrations. The application of the present invention is helpful for research and decision-making in fields such as environmental protection and health risk assessment.

[0081] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method, characterized in that: The following steps are involved: Obtain formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained from domestic satellite observation inversion, and align these data to construct Earth's spatiotemporal information data. Related substance emission data include formaldehyde emission data and emission data of other related species that are interrelated with the chemical transformation process of formaldehyde in the atmosphere; human geography data include population, leaf area index, and topographic data; The formaldehyde concentration data in the earth's spatiotemporal information data is partially randomly masked and then input into a neural operator model designed according to the data type. The random mask is calculated by the neural operator model to complete the formaldehyde concentration data, and semi-supervised learning is performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters. The neural operator model includes an encoder, a neural operator layer, and a decoder connected in sequence, wherein the neural operator layer adopts a spherical Fourier neural operator, an orthogonal neural operator, a Transolver, or a universal neural operator transformer; When performing semi-supervised learning on the neural operator, the Earth's spatiotemporal information data is coarsely sampled at a coarse resolution grid. The coarsely sampled sample data is used to learn the neural operator model. At the same time, some random mask areas are filled with constants, and the null value areas are skipped when calculating the reconstruction error. The parameter-optimized neural operator model is used to analyze and complete the observed formaldehyde concentration data to obtain the completed formaldehyde concentration data.

2. The domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method according to claim 1, characterized in that, The meteorological data includes temperature, humidity, pressure, wind speed, and potential.

3. The domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method according to claim 1 is characterized in that, The reconstruction error uses mean square error, L2 norm or absolute error.

4. The domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method according to claim 1 is characterized in that, When performing semi-supervised learning on the neural operator, after using the coarsely sampled sample data to learn the neural operator model, the earth's spatiotemporal information data is finely sampled with a fine-resolution grid, and the neural operator is learned using the finely sampled sample data.

5. A domestic satellite full-time and space continuous atmospheric formaldehyde analysis and supplement device, characterized in that: include: A data acquisition and processing module is used to acquire formaldehyde concentration data, meteorological data, related substance emission data, and human geography data obtained from domestic satellite observations and inversion, and to align these data to construct Earth's spatiotemporal information data. The related substance emission data includes formaldehyde emission data and emission data of other related species that are interrelated with the chemical transformation process of formaldehyde in the atmosphere. The human geography data includes population, leaf area index, and topographic data. A model construction and training module is used to perform partial random masking on the formaldehyde concentration data in the earth's spatiotemporal information data and input it into a neural operator model designed according to the data type. The random mask is used to complete the formaldehyde concentration data through neural operator calculation, and semi-supervised learning is performed based on the reconstruction error between the completed formaldehyde concentration data and the observed formaldehyde concentration data to optimize the neural operator parameters. The neural operator model includes an encoder, a neural operator layer, and a decoder connected in sequence, wherein the neural operator layer adopts a spherical Fourier neural operator, an orthogonal neural operator, a transolver, or a universal neural operator transformer; When performing semi-supervised learning on the neural operator, the Earth's spatiotemporal information data is coarsely sampled at a coarse resolution grid. The coarsely sampled sample data is used to learn the neural operator model. At the same time, some random mask areas are filled with constants, and the null value areas are skipped when calculating the reconstruction error. The parsing and completion module is used to parse and complete the observed formaldehyde concentration data using a parameter-optimized neural operator model to obtain the completed formaldehyde concentration data.

6. A computing device comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the one or more processors execute the executable code, they are used to implement the domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that A program is stored thereon, and when the program is executed by the processor, the domestic satellite full-time and space continuous atmospheric formaldehyde analysis and completion method described in any one of claims 1-4 is implemented.

Images