XCO2 spatial-temporal distribution high-resolution reconstruction method and related product

Abstract

The invention relates to the technical field of environmental monitoring and remote sensing, in particular to an XCO2 spatial-temporal distribution high-resolution reconstruction method and related products, and the method comprises the steps: firstly obtaining and fusing observation data of OCO-2, OCO-3 and GOSAT satellites to form a multi-source satellite XCO2 data set; then, environment covariables (such as meteorological data, altitude, vegetation index and the like) of XCO2 distribution are collected, and the data and an XCO2 data set are input into a WaveletResNet-LSTM model together for training; according to the model, multi-scale features are extracted through wavelet analysis, spatial features are extracted in combination with ResNet, time dynamic changes are captured through LSTM, finally, a high-precision XCO2 concentration predicted value is generated, and a space-time distribution diagram is constructed according to the high-precision XCO2 concentration predicted value.

Description

Technical Field

[0001] The present invention relates to the field of environmental monitoring and remote sensing technology, and in particular to a high-resolution reconstruction method for spatiotemporal distribution of XCO2 and related products. Background Art

[0002] In recent years, as the global climate change problem becomes increasingly serious, studies have found that carbon dioxide (CO2) is one of the main greenhouse gases, and its concentration changes directly affect the stability and sustainability of the global climate system. Therefore, how to accurately monitor and predict the temporal and spatial distribution of CO2 in the atmosphere has become a focus of attention. Traditional ground monitoring stations are difficult to fully reflect the changes in CO2 concentration over a large area due to their limited coverage and high equipment costs, especially in complex terrain areas.

[0003] Satellite remote sensing technology has gradually become an important means of atmospheric CO2 monitoring due to its wide coverage and high temporal and spatial resolution. Currently, there are many satellites, such as GOSAT (Greenhouse Gas Observation Satellite), OCO-2 and OCO-3 (Orbiting Carbon Observation Satellite), which can provide observation data of the average dry air mole fraction of atmospheric CO2 column (XCO2). However, these satellite data are limited by factors such as meteorological conditions, aerosols and surface reflectivity, and often contain a large amount of non-random missing data, making it difficult to obtain a full coverage XCO2 dataset for a specified area, and thus unable to obtain accurate XCO2 temporal and spatial distribution. Summary of the invention

[0004] The technical problem to be solved by the present invention is how to improve the reconstruction accuracy and resolution of the spatiotemporal distribution of XCO2 concentration. The purpose is to provide a high-resolution reconstruction method for the spatiotemporal distribution of XCO2 and related products, which realizes the use of deep learning models to fully mine the spatiotemporal information of satellite data and generate high-precision and high-resolution spatiotemporal distribution maps of XCO2 concentrations.

[0005] The present invention is achieved through the following technical solutions:

[0006] A high-resolution reconstruction method for the spatiotemporal distribution of XCO2, comprising:

[0007] Obtain OCO-2 satellite observation data and OCO-3 satellite observation data, and merge them to obtain the OCO dataset;

[0008] Obtain GOSAT satellite observation data, and fuse the GOSAT satellite observation data with the OCO dataset to obtain the XCO2 dataset; divide the XCO2 dataset into the XCO2 training dataset and the XCO2 verification dataset;

[0009] Determine the environmental covariates associated with the XCO2 distribution and obtain the environmental covariate data corresponding to the XCO2 dataset;

[0010] Build a Wavelet_ResNet-LSTM model and train it using the data in the XCO2 training dataset as the dependent variable and the environmental covariate data as the independent variable;

[0011] Obtain real-time environmental covariate data and input the real-time covariate data into the trained Wavelet_ResNet-LSTM model to obtain the corresponding XCO2 concentration prediction value;

[0012] According to the output XCO2 concentration prediction value, the XCO2 spatiotemporal distribution map is generated.

[0013] Specifically, the OCO-2 satellite observation data, OCO-3 satellite observation data, and GOSAT satellite observation data were spatially resampled by bilinear interpolation, and the original resolution was sampled to a 0.01°×0.01° grid.

[0014] Optionally, the method for obtaining the OCO dataset includes:

[0015] Align the observation time of OCO-2 satellite observation data and OCO-3 satellite observation data, and normalize the OCO-2 satellite observation data and OCO-3 satellite observation data;

[0016] The normalized OCO-2 satellite observation data and OCO-3 satellite observation data are converted to the same coordinate system and spatially matched;

[0017] If only OCO-2 satellite observation data or OCO-3 satellite observation data exists in the same geographic grid, the OCO-2 satellite observation data or OCO-3 satellite observation data will be used as the XCO2 concentration value of the geographic grid;

[0018] If both OCO-2 satellite observation data and OCO-3 satellite observation data exist in the same geographic grid, the weighted average of the OCO-2 satellite observation data and the OCO-3 satellite observation data is calculated, and the weighted average is used as the XCO2 concentration value of the geographic grid;

[0019] The XCO2 concentration values ​​for all geographic grids are obtained and combined into the OCO dataset.

[0020] Optionally, the method for obtaining the XC02 dataset includes:

[0021] Building a linear model OCO pred =α+β·GOSAT, where GOSAT is the GOSAT satellite observation data, OCO pred is the OCO prediction data, α and β are the linear model parameters estimated by the least squares method;

[0022] Align the observation time of GOSAT satellite observation data and OCC dataset; and transform GOSAT satellite observation data to the same coordinate system as OCO dataset;

[0023] OCO forecast data for all geographic grids are predicted using a linear model and GOSAT satellite observation data;

[0024] The OCO prediction data and the OCO dataset are weighted averaged to obtain the XCO2 dataset.

[0025] Specifically, the environmental covariate data include meteorological data, altitude data, normalized difference vegetation index, population density data, mountain utilization type data, and corresponding regional economic data.

[0026] Optionally, the constructed Wavelet_ResNet-LSTM model includes: an input layer, a Wavelet analysis layer, a ResNet network structure, an LSTM network structure, and a fully connected layer;

[0027] The methods for training the Wavelet_ResNet-LSTM model include:

[0028] Input the environmental covariate data and XCO2 training dataset into the input layer of the Wavelet_ResNet-LSTM model;

[0029] The input data is decomposed by wavelet through the Wavelet analysis layer to extract multi-scale features;

[0030] The ResNet network structure is used to perform convolution operations on multi-scale features to extract high-dimensional spatial features;

[0031] The high-dimensional spatial features are processed in time series through the LSTM network structure to extract temporal dynamic features;

[0032] The multi-scale features extracted by the Wavelet analysis layer, the high-dimensional spatial features extracted by the ResNet network structure, and the temporal dynamic features extracted by the LSTM network structure are fused through the fully connected layer, and the predicted XC02 concentration is output.

[0033] Specifically, the input signal f(t) is subjected to discrete wavelet transform to obtain the wavelet coefficient c corresponding to the input signal at scale j and position k: j，k , Among them, ψ j，k (t) is the wavelet basis function of scale j and position k;

[0034] Obtaining multi-scale features Where J is the maximum scale level of discrete wavelet transform;

[0035] Determine the residual function F(x, {W m})=σ(W2·σ(W1·x+b1)+b2), where x is the input of the residual block, W1 is the convolution kernel of the first layer convolution operation, W2 is the convolution kernel of the second layer convolution operation, b1 and b2 are biases, σ is a nonlinear activation function, and W m is the mth convolution kernel;

[0036] Determine the structure of the residual block in the ResNet network structure: y = F(x, {W m})+x+w;

[0037] Obtain the high-dimensional spatial feature y=σ(W2·σ(W1·x+b1)+b2)+x+w.

[0038] Specifically, construct the LSTM network structure. Among them, i t is the input gate, f t For the forget gate, t is the output gate, g t is the candidate state, c t For memory cells, h t is the hidden state, y t is the input of the high-dimensional spatial feature at each time step, W i , W f , W o and W g is the weight matrix, b i 、b f 、b o and b g is the bias, σ S is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, and ⊙ is the Hadamard product.

[0039] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the high-resolution reconstruction method of XCO2 spatiotemporal distribution is implemented.

[0040] A computer program product comprises a computer program / instruction, which, when executed by a processor, implements the above-mentioned high-resolution reconstruction method of XCO2 spatiotemporal distribution.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] The present invention proposes a high-resolution reconstruction method for the spatiotemporal distribution of XCO2 based on a Wavelet_ResNet-LSTM model. Firstly, the observation data of OCO-2, OCO-3 and GOSAT satellites are acquired and integrated to form a multi-source satellite XCO2 data set. Then, environmental covariates (such as meteorological data, altitude, vegetation index, etc.) of the XCO2 distribution are collected, and these data are input into the Wavelet_ResNet-LSTM model together with the XCO2 data set for training. The model extracts multi-scale features through wavelet analysis, extracts spatial features in combination with ResNet, and captures temporal dynamic changes through LSTM, finally generating a high-precision XCO2 concentration prediction value, and constructing a spatiotemporal distribution map based on the value.

[0043] The present invention significantly improves the reconstruction accuracy and resolution of the spatiotemporal distribution of XCO2. By introducing the Wavelet_ResNet-LSTM model, extracting multi-scale features through Wavelet analysis, combining ResNet to extract spatial features and LSTM to process time series data, the spatiotemporal fusion accuracy of multi-source satellite data is effectively improved, which can effectively cope with the complexity of high-dimensional spatial features and temporal dynamic changes, and reduce the impact of data missing on the reconstruction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings illustrate exemplary embodiments of the present invention and, together with the description thereof, are used to explain the principles of the present invention. These drawings are included to provide a further understanding of the present invention, and the accompanying drawings are included in and constitute a part of this specification and do not constitute a limitation of the embodiments of the present invention.

[0045] Figure 1 It is a schematic diagram of a process of a high-resolution reconstruction method of XCO2 spatiotemporal distribution according to the present invention.

[0046] Figure 2 It is a schematic diagram of the satellite data processing and fusion process according to the present invention.

[0047] Figure 3 is a graph of the model training and validation loss curves according to the present invention.

[0048] Figure 4 It is a scatter plot comparing the predicted value and the true value on the validation set according to the present invention.

[0049] Figure 5 It is a comparative verification diagram of the traditional model according to the present invention.

[0050] Figure 6 This is a graph showing changes in seasonal average XCO2 concentration in Sichuan Province from 2015 to 2022 according to the present invention.

[0051] Figure 7 This is a graph showing changes in monthly average XCO2 concentration in Sichuan Province from 2015 to 2022 according to the present invention.

[0052] Figure 8 This is a graph showing changes in the annual average XCO2 concentration in Sichuan Province from 2015 to 2022 according to the present invention.

[0053] Fig. 9 This is the average spatial distribution of XCO2 in Sichuan from 2015 to 2022 according to the present invention. DETAILED DESCRIPTION

[0054] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and implementation methods. It is understood that the specific implementation methods described herein are only used to explain the relevant content, rather than to limit the present invention.

[0055] It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings.

[0056] In the absence of conflict, the embodiments and features of the embodiments of the present invention may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0057] Most existing technologies use interpolation methods, traditional statistical methods or machine learning-based algorithms (such as xgboost, random forest, etc.) to fill in and fuse these satellite data. However, these methods have significant limitations in processing the high-dimensional spatial characteristics and temporal dynamic changes of multi-source satellite data, and it is difficult to fully utilize the spatiotemporal information of multi-source data, resulting in low accuracy and resolution of XCO2 concentration reconstruction.

[0058] Therefore, there is an urgent need for a new method that can effectively fuse multi-source satellite data and fully exploit their spatiotemporal characteristics to improve the accuracy and spatiotemporal resolution of XCO2 concentration reconstruction.

[0059] Embodiment 1

[0060] like Figure 1 As shown, a high-resolution reconstruction method for the spatiotemporal distribution of XCO2 is provided, comprising:

[0061] The OCO-2 and OCO-3 satellite observation data were obtained and merged to obtain the OCO dataset; the OCO-2 and OCO-3 satellite observation data are CO2 column average dry air mole fraction (XCO2) data obtained by the orbiting carbon observation satellite launched by the National Aeronautics and Space Administration (NASA).

[0062] Obtain GOSAT satellite observation data, and fuse the GOSAT satellite observation data with the OCO dataset to obtain the XCO2 dataset; divide the XCO2 dataset into the XCO2 training dataset and the XCO2 validation dataset; GOSAT satellite observation data is obtained by the Greenhouse Gas Observation Satellite (GOSAT) launched by the Japan Aerospace Exploration Agency (JAXA). By fusing GOSAT data with the previously generated OCO dataset, a more comprehensive XCO2 dataset can be formed. The training set is used to train the model, while the validation set is used to evaluate the performance of the model to ensure its accuracy on new data.

[0063] Determine the environmental covariates associated with the XCO2 distribution and obtain the environmental covariate data corresponding to the XCO2 dataset;

[0064] A Wavelet_ResNet-LSTM model was constructed, and the data in the XCO2 training dataset was used as the dependent variable and the environmental covariate data as the independent variable to train the Wavelet_ResNet-LSTM model; environmental covariates refer to other factors that may affect the distribution of XCO2, such as meteorological data (temperature, humidity, etc.), altitude, vegetation index (NDVI), etc. These data are obtained through ground observations or other satellites and associated with the XCO2 dataset as input features of the model.

[0065] Real-time environmental covariate data are obtained and input into the trained Wavelet_ResNet-LSTM model to obtain the corresponding XCO2 concentration prediction value; the Wavelet_ResNet-LSTM model is a deep learning model that combines wavelet analysis (Wavelet), residual network (ResNet) and long short-term memory network (LSTM). Wavelet analysis is used to extract multi-scale features, ResNet is used to capture spatial features, and LSTM processes temporal dynamic features. The XCO2 value in the XCO2 training data set is used as the dependent variable (i.e., the prediction target of the model), and the environmental covariate data is used as the independent variable (i.e., the input feature of the model). These data are input into the model for training to capture the spatiotemporal variation of XCO2 concentration.

[0066] According to the output XCO2 concentration prediction value, the XCO2 spatiotemporal distribution map is generated.

[0067] Embodiment 2

[0068] In practice, satellite observation data are usually collected at different resolutions. In order to ensure data consistency, these data need to be resampled, that is, the data is converted from the original resolution to a uniform resolution. In this embodiment, the target resolution is set to a 0.01°×0.01° grid, which is equivalent to a resolution of approximately 1 km×1 km on the surface. This high-resolution setting can provide finer spatial details, which helps to improve the accuracy of XCO2 concentration prediction.

[0069] The OCO-2 satellite observation data, OCO-3 satellite observation data, and GOSAT satellite observation data are spatially resampled through bilinear interpolation. The purpose of spatial resampling is to unify all satellite data to a finer spatial resolution.

[0070] Bilinear interpolation is to interpolate data points on a two-dimensional plane. This method estimates the value of the point to be interpolated by taking a weighted average of the values ​​of four adjacent known points. Its core principle is to perform linear interpolation in the horizontal and vertical directions respectively, and finally obtain the interpolation result of the target point.

[0071] Embodiment 3

[0072] like Figure 3 As shown, the method of obtaining the OCO dataset includes:

[0073] Align the observation time of OCO-2 satellite observation data and OCO-3 satellite observation data, and normalize the OCO-2 satellite observation data and OCO-3 satellite observation data; aligning the observation time means matching the observation data of OCO-2 and OCO-3 satellites in the same time period. Since the observation time of satellites may be different, it is necessary to adjust their data to the same time point to ensure the time consistency of the data. Normalization refers to the standardization of the observation data of OCO-2 and OCO-3 satellites, generally converting the data values ​​to a standard range or scale to eliminate the possible deviations or scale differences between different satellites.

[0074] The normalized OCO-2 satellite observation data and OCO-3 satellite observation data are converted to the same coordinate system and spatially matched; the coordinate system conversion is to ensure the consistency of the spatial positions of different satellite data, and spatial matching is to locate these data into a unified geographic grid system.

[0075] When processing this data, you may encounter two situations:

[0076] Single data source: If only OCO-2 satellite observation data or OCO-3 satellite observation data exist in the same geographic grid, the OCO-2 satellite observation data or OCO-3 satellite observation data will be used as the XCO2 concentration value of the geographic grid;

[0077] Multiple data sources: If both OCO-2 and OCO-3 satellite observation data exist in the same geographic grid, the weighted average of the OCO-2 and OCO-3 satellite observation data is calculated and used as the XCO2 concentration value of the geographic grid.

[0078] The XCO2 concentration values ​​for all geographic grids are obtained and combined into the OCO dataset.

[0079] The methods for obtaining the XCO2 dataset include:

[0080] Building a linear model OCO pred =α+β·GOSAT, where GOSAT is the GOSAT satellite observation data, OCO pred is the OCO prediction data, α and β are the linear model parameters estimated by the least squares method;

[0081] Similarly, the observation time of the GOSAT satellite observation data and the OCC dataset is aligned; and the GOSAT satellite observation data is converted to the same coordinate system as the OCO dataset.

[0082] OCO forecast data for all geographic grids are predicted using a linear model and GOSAT satellite observation data;

[0083] The OCO prediction data and the OCO dataset are weighted averaged to obtain the XCO2 dataset.

[0084] Embodiment 3

[0085] The environmental covariate data include meteorological data, altitude data, normalized difference vegetation index, population density data, mountain utilization type data, and corresponding regional economic data.

[0086] Meteorological data: Key meteorological elements such as temperature, humidity, precipitation, atmospheric pressure, relative humidity, sunshine duration, and wind speed are obtained from the ERA5 dataset of the European Centre for Medium-Range Weather Forecasts (ECMWF).

[0087] Elevation data: Uses high-precision terrain data provided by NASA's Shuttle Radar Topography Mission (SRTM).

[0088] Normalized Difference Vegetation Index (NDVI): The vegetation coverage is obtained using the MOD13Q1 and MYD13Q1 datasets generated by NASA's Terra and Aqua satellites.

[0089] Population density data: Get detailed population distribution information from the World Grid Socio-Economic Population Data and Applications Centre (SEDAC).

[0090] Land use type: land use type data provided by the European Space Agency (ESA), including cultivated land, forest land, garden land, mountain land, city and water body, etc.

[0091] Corresponding regional economic data: collect the gross domestic product (GDP), population and carbon sink data of the corresponding region, which come from the statistical yearbook and NASA's global net primary productivity (NPP) data products respectively.

[0092] Embodiment 4

[0093] Combining wavelet analysis, ResNet structure in convolutional neural network (CNN) and long short-term memory network (LSTM), the constructed Wavelet_ResNet-LSTM model includes: input layer, Wavelet analysis layer, ResNet network structure, LSTM network structure and fully connected layer;

[0094] The methods for training the Wavelet_ResNet-LSTM model include:

[0095] The environmental covariate data and the XCO2 training data set are input to the input layer of the Wavelet_ResNet-LSTM model; the input layer receives input data, including environmental covariate data and the XCO2 training data set. Environmental covariate data refers to various factors that affect the distribution of XCO2, such as temperature, humidity, altitude, etc., while the XCO2 training data set is the known XCO2 concentration value.

[0096] The input data is decomposed by wavelet through the Wavelet analysis layer to extract multi-scale features; wavelet decomposition decomposes the input signal into sub-signals of different scales to extract the multi-scale features of the data.

[0097] The ResNet network structure performs convolution operations on multi-scale features to extract high-dimensional spatial features; convolution operations are used in convolutional neural networks (CNNs) to extract high-dimensional spatial features in image processing or spatial data processing. ResNet (residual network) alleviates the problem of gradient vanishing in deep networks by introducing residual blocks.

[0098] The LSTM network structure is used to process high-dimensional spatial features in time series and extract temporal dynamic features. LSTM is a recurrent neural network (RNN) used to process time series data and can capture temporal dynamic features in the data. LSTM can retain long-term dependent information and filter out irrelevant information through its special gating mechanism (input gate, forget gate, output gate), making it suitable for analyzing and predicting trends and periodic changes in time series data.

[0099] The multi-scale features extracted by the Wavelet analysis layer, the high-dimensional spatial features extracted by the ResNet network structure, and the temporal dynamic features extracted by the LSTM network structure are fused through the fully connected layer, and the predicted XCO2 concentration is output. The fully connected layer combines all the extracted features and generates the final output, which is the predicted XCO2 concentration.

[0100] Embodiment 5

[0101] A specific method is provided for the fourth embodiment.

[0102] Perform discrete wavelet transform on the input signal f(t) to obtain the wavelet coefficient c corresponding to the input signal at scale j and position k j,k , Among them, ψ j，k (t) is the wavelet basis function of scale j and position k.

[0103] Obtaining multi-scale features Where J is the maximum scale level of discrete wavelet transform;

[0104] Determine the residual function F(x, {W m})=σ(W2·σ(W1·x+b1)+b2), where x is the input of the residual block, W1 is the convolution kernel of the first layer convolution operation, W2 is the convolution kernel of the second layer convolution operation, b1 and b2 are biases, σ is a nonlinear activation function, and W m is the mth convolution kernel;

[0105] Determine the structure of the residual block in the ResNet network structure: y = F(x, {W m})+x+w;

[0106] Obtain the high-dimensional spatial feature y=σ(W2·σ(W1·x+b1)+b2)+x+w.

[0107] Specifically, construct the LSTM network structure. Among them, i t is the input gate, f t For the forget gate, t is the output gate, g tis the candidate state, c t For memory cells, h t is the hidden state, y t is the input of the high-dimensional spatial feature at each time step, W i , W f , W o and W g is the weight matrix, b i 、b f 、b o and b g is the bias, σ S is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, and ⊙ is the Hadamard product.

[0108] Provide the specific parameters of a Wavelet_ResNet-LSTM model.

[0109] ResNet layers: N = 50;

[0110] Number of residual blocks: M = 16;

[0111] Convolution kernel size: 3×3;

[0112] Wavelet basis function: Daubechies 4 (db4);

[0113] Number of wavelet decomposition layers: 3;

[0114] Number of LSTM layers: K = 2;

[0115] Number of LSTM hidden units: hidden_units = 256;

[0116] Learning rate: learning_rate=0.001;

[0117] Number of training rounds: epochs = 100;

[0118] Batch size: batch_size=64.

[0119] Embodiment 6

[0120] This embodiment takes the data of Sichuan Province from 2015 to 2022 as an example to provide a specific implementation case.

[0121] like Figure 3 , Figure 4 and Figure 5 As shown in the figure, the Wavelet_ResNet-LSTM model established by the present invention is verified. 2The values ​​reached 0.97 and the root mean square error (RMSE) was 1.40ppm, which were significantly better than the traditional xgboost and random forest models, showing higher prediction accuracy and stability.

[0122] like Figure 6 , Figure 7 , Figure 8 and Fig. 9 As shown in the data, during the period of 2015-2022, the average XCO2 concentration in Sichuan Province showed a continuous upward trend, rising from 398.9ppm in 2015 to 416.6ppm in 2022, with an average annual growth rate of 2.21ppm / year. Although the XCO2 concentration is increasing overall, its growth rate has slowed down in the later period, which may be related to the transformation of economic development mode, the adjustment of industrial structure and the implementation of emission reduction policies.

[0123] The XCO2 concentration in Sichuan Province shows significant seasonal changes. The XCO2 concentration is highest in spring (March, April, and May), decreases significantly in summer (June, July, and August), increases month by month in autumn (September, October, and November), and is relatively stable in winter (December, January, and February). In the transition period of spring and autumn, the XCO2 concentration shows an upward and downward trend, respectively. In summer, which is the peak period of photosynthesis, the concentration decreases significantly, while in winter, plants basically stop photosynthesis and the concentration changes little.

[0124] Due to the diversity of geographical features and climatic conditions in Sichuan Province, there are significant differences in the distribution of XCO2 concentrations in different regions. Vegetation in high-altitude areas has a stronger photosynthesis capacity during the summer growing season and has a stronger absorption effect on CO2 in the atmosphere, which significantly affects the spatial heterogeneity of XCO2 in the region.

[0125] The spatial distribution maps and spatial distribution data obtained provide a solid data foundation for the spatial heterogeneity analysis and temporal dynamic monitoring of carbon emissions, which helps local governments formulate scientific and reasonable carbon reduction strategies and policies. For example, intensify afforestation and greening measures in spring and autumn to enhance the ability of plants to absorb CO2, or implement more stringent emission control measures in winter to reduce anthropogenic CO2 emissions. Accurate high-resolution XCO2 spatiotemporal distribution results can be used to verify local emission inventories and provide quantitative data support for local emission reduction and carbon reduction strategies.

[0126] Embodiment 6

[0127] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the high-resolution reconstruction method of XCO2 spatiotemporal distribution is implemented.

[0128] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer readable instruction data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storage technology, CD-ROM, DVD or other optical storage, cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above. The above-mentioned system memory and mass storage devices can be collectively referred to as memory.

[0129] A computer program product comprises a computer program / instruction, which, when executed by a processor, implements the above-mentioned high-resolution reconstruction method of XCO2 spatiotemporal distribution.

[0130] A computer program product includes a computer program or set of instructions for performing specific tasks or implementing specific functions. These programs or instructions are designed to be executed by a processor to implement a series of predefined steps or operations. The program product may be stored in various forms of computer storage media, such as memory, hard disk, solid-state drive, optical disk or other forms of digital storage devices. It may exist in the form of compiled binary code or in the form of scripts or bytecodes that can be executed by an interpreter. The program product uses carefully designed algorithms and logical instructions to enable the processor to process data in a specific order and manner to complete various functions such as data analysis, user interaction, device control, etc.

[0131] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments / methods or examples described in this specification and the features of the different embodiments / methods or examples, unless they are contradictory.

[0132] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0133] It should be understood by those skilled in the art that the above embodiments are only for the purpose of clearly illustrating the present invention, and are not intended to limit the scope of the present invention. For those skilled in the art, other changes or modifications may be made based on the above invention, and these changes or modifications are still within the scope of the present invention.

Claims

1. A high-resolution reconstruction method for the spatiotemporal distribution of XCO2, characterized in that: include: Obtain OCO-2 satellite observation data and OCO-3 satellite observation data, and merge them to obtain the OCO dataset; Obtain GOSAT satellite observation data and fuse the GOSAT satellite observation data with the OCO dataset to obtain the XCO2 dataset; Divide the XCO2 dataset into the XCO2 training dataset and the XCO2 validation dataset; Determine the environmental covariates associated with the XCO2 distribution and obtain the environmental covariate data corresponding to the XCO2 dataset; Build a Wavelet_ResNet-LSTM model and train it using the data in the XCO2 training dataset as the dependent variable and the environmental covariate data as the independent variable; Obtain real-time environmental covariate data and input the real-time covariate data into the trained Wavelet_ResNet-LSTM model to obtain the corresponding XCO2 concentration prediction value; According to the output XCO2 concentration prediction value, the XCO2 spatiotemporal distribution map is generated.

2. A high-resolution reconstruction method of XCO2 spatiotemporal distribution according to claim 1, characterized in that: The OCO-2 satellite observation data, OCO-3 satellite observation data, and GOSAT satellite observation data were spatially resampled by bilinear interpolation, and the original resolution was sampled to a 0.01°×0.01° grid.

3. The method for high-resolution reconstruction of spatiotemporal distribution of XCO2 according to claim 1, characterized in that: Methods for obtaining the OCO dataset include: Align the observation time of OCO-2 satellite observation data and OCO-3 satellite observation data, and normalize the OCO-2 satellite observation data and OCO-3 satellite observation data; The normalized OCO-2 satellite observation data and OCO-3 satellite observation data are converted to the same coordinate system and spatially matched; If only OCO-2 satellite observation data or OCO-3 satellite observation data exists in the same geographic grid, the OCO-2 satellite observation data or OCO-3 satellite observation data will be used as the XCO2 concentration value of the geographic grid; If both OCO-2 satellite observation data and OCO-3 satellite observation data exist in the same geographic grid, the weighted average of the OCO-2 satellite observation data and the OCO-3 satellite observation data is calculated, and the weighted average is used as the XCO2 concentration value of the geographic grid; The XCO2 concentration values ​​for all geographic grids are obtained and combined into the OCO dataset.

4. The method for high-resolution reconstruction of spatiotemporal distribution of XCO2 according to claim 1, characterized in that: The methods for obtaining the XCO2 dataset include: Building a linear model OCO pred =α+β·GOSAT, where GOSAT is the GOSAT satellite observation data, OCO pred is the OCO prediction data, α and β are the linear model parameters estimated by the least squares method; Align the observation time of GOSAT satellite observation data and OCC dataset; and transform GOSAT satellite observation data to the same coordinate system as OCO dataset; OCO forecast data for all geographic grids are predicted using a linear model and GOSAT satellite observation data; The OCO prediction data and the OCO dataset are weighted averaged to obtain the XCO2 dataset.

5. The method for high-resolution reconstruction of spatiotemporal distribution of XCO2 according to claim 1, characterized in that: The environmental covariate data include meteorological data, altitude data, normalized difference vegetation index, population density data, mountain utilization type data, and corresponding regional economic data.

6. The method for high-resolution reconstruction of spatiotemporal distribution of XCO2 according to claim 1, characterized in that: The constructed Wavelet_ResNet-LSTM model includes: input layer, Wavelet analysis layer, ResNet network structure, LSTM network structure and fully connected layer; Methods for training the Wavelet_ResNet-LSTM model include: Input the environmental covariate data and XCO2 training dataset into the input layer of the Wavelet_ResNet-LSTM model; The input data is decomposed by wavelet analysis layer to extract multi-scale features; The ResNet network structure is used to perform convolution operations on multi-scale features to extract high-dimensional spatial features; The high-dimensional spatial features are processed in time series through the LSTM network structure to extract temporal dynamic features; The multi-scale features extracted by the Wavelet analysis layer, the high-dimensional spatial features extracted by the ResNet network structure, and the temporal dynamic features extracted by the LSTM network structure are fused through the fully connected layer, and the predicted XCO2 concentration is output.

7. A high-resolution reconstruction method of XCO2 spatiotemporal distribution according to claim 6, characterized in that: Perform discrete wavelet transform on the input signal f(t) to obtain the wavelet coefficient c corresponding to the input signal at scale j and position k j，k , Among them, ψ j,k (t) is the wavelet basis function of scale j and position k; Obtaining multi-scale features Where J is the maximum scale level of discrete wavelet transform; Determine the residual function F(x, {W m })=σ(W2·σ(W1·x+b1)+b2), where x is the input of the residual block, W1 is the convolution kernel of the first layer convolution operation, W2 is the convolution kernel of the second layer convolution operation, b1 and b2 are biases, σ is a nonlinear activation function, and W m is the mth convolution kernel; Determine the structure of the residual block in the ResNet network structure: y = F(x, {W m })+x+w; Obtain the high-dimensional spatial feature y=σ(W2·σ(W1·x+b1)+b2)+x+w.

8. The method for high-resolution reconstruction of spatiotemporal distribution of XCO2 according to claim 6, characterized in that: Construct LSTM network structure, Among them, i t is the input gate, f t For the forget gate, t is the output gate, h t is the candidate state, c t For memory cells, h t is the hidden state, y t is the input of the high-dimensional spatial feature at each time step, W i , W f , W o and W g is the weight matrix, b i , b f , b o and b g is the bias, σ S is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, and ⊙ is the Hadamard product.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for high-resolution reconstruction of the spatiotemporal distribution of XCO2 according to any one of claims 1 to 8 is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the method for high-resolution reconstruction of the spatiotemporal distribution of XCO2 as claimed in any one of claims 1 to 8 is implemented.

Images