Atmospheric carbon spatiotemporal neural network regression modeling method, system, medium and equipment

Abstract

The present invention discloses an atmospheric carbon spatiotemporal neural network regression modeling method, system, medium and equipment, which belong to the field of remote sensing atmospheric monitoring. The present invention first obtains the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required study area, and performs preliminary preprocessing and further preprocessing; constructs a geographic spatiotemporal weighted regression model to extract spatiotemporal features, wherein a spatial proximity convolutional neural network is used to extract spatial proximity representations, a temporal proximity neural network is used to extract temporal proximity representations, a spatiotemporal fusion proximity neural network is used to represent fusion, and a spatiotemporal weighted neural network is used to solve spatiotemporal weights, and finally a full coverage high spatiotemporal resolution atmospheric carbon concentration distribution is obtained by the geographic spatiotemporal weighted regression model and a visual map is generated. The method of the present invention has strong generalization ability and high data modeling accuracy, and can be applied to refined full coverage modeling of atmospheric XCO‑2 concentrations at different spatial scales and in different regions.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing atmospheric monitoring and big data, and in particular relates to an atmospheric carbon spatiotemporal neural network regression modeling method, system, medium and equipment. Background Art

[0002] In recent years, global warming has become increasingly destructive, requiring countries around the world to take urgent action to address climate change. Monitoring the carbon content in the atmosphere is of great significance for understanding the global carbon cycle, assessing climate change, and formulating emission reduction strategies.

[0003] Traditional atmospheric carbon monitoring methods mainly rely on ground monitoring stations and satellite remote sensing technology. Although these methods provide valuable data, they have limitations in spatial coverage, data update frequency and degree of refinement. In recent years, with the development of remote sensing technology and the improvement of data processing capabilities, the use of machine learning, especially deep learning methods, to process and analyze remote sensing data and improve the accuracy and efficiency of atmospheric carbon monitoring has become a research hotspot.

[0004] However, the complexity of atmospheric carbon monitoring requires the model to not only understand spatial information, but also capture the changing trends in time series data. In addition, the environmental characteristics of different geographical locations have different effects on atmospheric carbon content, which requires the model to have the ability to process high-dimensional data and understand geospatial information.

[0005] Carbon satellite monitoring data has obvious strip-like distribution characteristics, and the unevenness of spatiotemporal distribution is prominent. The existing spatiotemporal regression methods mainly rely on the multidimensional distance between samples to express spatiotemporal proximity, which is easily affected by the sample distribution and cannot fully and accurately express the proximity of geographic entities, which greatly limits the modeling accuracy of spatiotemporal relationships. Summary of the invention

[0006] The purpose of the present invention is to achieve full coverage modeling of atmospheric XCO2 refined spatiotemporal distribution based on geographic spatiotemporal convolutional neural network weighted regression, fully consider the complexity of data distribution spatiotemporal characteristics, improve the problem of uneven data distribution reducing modeling accuracy, so as to enhance the monitoring ability of atmospheric XCO2 concentration distribution changes and provide support for carbon reduction and emission reduction. And provide an atmospheric carbon spatiotemporal neural network regression modeling method, system, medium and equipment.

[0007] In order to achieve the above-mentioned invention object, the present invention specifically adopts the following technical solutions:

[0008] In a first aspect, the present invention provides an atmospheric carbon spatiotemporal neural network regression modeling method, comprising the following steps:

[0009] S1. Acquire the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required study area, and preliminarily preprocess the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required study area to obtain the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after the preliminary preprocessing;

[0010] S2, constructing a data set after further preprocessing the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after preliminary preprocessing;

[0011] S3. Input the data set into a trained geographic spatiotemporal weighted regression model. First, extract the spatial proximity representation of the data set by a spatial proximity convolutional neural network. Then, extract the temporal proximity representation of the data set by a temporal proximity neural network. Then, fuse the spatial proximity representation and the temporal proximity representation by a spatiotemporal fusion proximity neural network to obtain the spatiotemporal proximity representation. Input the spatiotemporal proximity representation into a spatiotemporal weighted neural network to calculate the spatiotemporal weight. Apply the least squares method to the data set to obtain a global regression coefficient. Multiply the global regression coefficient by the spatiotemporal weight to obtain a final regression coefficient. Use the atmospheric XCO2 influencing factor data after further preprocessing as an independent variable. Multiply the final regression coefficient by the independent variable to obtain an estimated result of the XCO2 concentration at each spatiotemporal coordinate. After visualization, a full coverage map of the atmospheric XCO2 concentration is obtained.

[0012] Based on the above scheme, each step can be implemented in the following preferred specific manner.

[0013] As a preferred embodiment of the above-mentioned first aspect, in step S1, the atmospheric XCO2 influencing factor data includes three categories, namely carbon source variables, carbon sink variables and environmental variables; wherein, the carbon source variables include land use type, night light index and surface temperature, the carbon sink variables include vegetation index, evapotranspiration and primary productivity, the vegetation index includes vegetation coverage, vegetation leaf area index and normalized vegetation index, and the environmental variables include ground elevation and surface roughness.

[0014] As a preferred embodiment of the first aspect, in step S2, the specific process of further preprocessing is as follows:

[0015] S21, according to the required research area, the range of atmospheric XCO2 time series remote sensing monitoring data and the range of atmospheric XCO2 impact factor data after preliminary preprocessing are cut to obtain atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 impact factor data within the research area;

[0016] S22. Divide the required study area into a uniform grid according to a predetermined modeling spatiotemporal resolution scale, wherein each grid unit represents a local area in the required study area. Associate the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data within the required study area with the uniform grid according to the geographic spatiotemporal location information, and calculate the linear correlation between the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data within the required study area by using the Pearson correlation coefficient.

[0017] S23. The atmospheric XCO2 influencing factor data within the required study area are screened based on the linear correlation calculation results, and a data set is constructed using the screened atmospheric XCO2 influencing factor data and the atmospheric XCO2 time series remote sensing monitoring data within the study area.

[0018] As a preferred embodiment of the above-mentioned first aspect, in step S23, the specific process of screening the atmospheric XCO2 impact factor data within the required research area by the linear correlation calculation results is as follows: a linear correlation threshold is set in advance, and the impact factors in the atmospheric XCO2 impact factor data whose linear correlation calculation results are greater than or equal to the linear correlation threshold are retained to obtain the screened atmospheric XCO2 impact factor data.

[0019] As a preferred embodiment of the above-mentioned first aspect, the spatial proximity convolutional neural network includes a first input layer, a convolution block, a flattening layer, a first fully connected layer with a ReLU activation function, a second fully connected layer with a ReLU activation function and a first output layer, and each convolution block is composed of a convolution layer with He parameter initialization and a ReLU activation function, a batch normalization and a pooling layer cascaded in sequence; the temporal proximity neural network is composed of a second input layer, a third fully connected layer with a ReLU activation function, a fourth fully connected layer with a ReLU activation function and a second output layer cascaded in sequence; the spatiotemporal fusion proximity neural network is composed of a third input layer, a fifth fully connected layer with He parameter initialization, Dropout, PReLU activation function, a sixth fully connected layer with He parameter initialization, Dropout, PReLU activation function and a third output layer cascaded in sequence; the spatiotemporal weighted neural network is composed of a fourth input layer, a seventh fully connected layer and a fourth output layer cascaded in sequence.

[0020] Furthermore, in the spatial proximity convolutional neural network of step S3, the specific processing flow is as follows:

[0021] AS31, taking the geographic spatial position information as the input of the first input layer, taking the center point of each grid cell as the spatial sampling point, taking the center point of the grid cell located in the pth row and the qth column as the spatial position of the sample point to be estimated, calculating the spatial distances between the sample point to be estimated and the remaining spatial sampling points except the sample point to be estimated in sequence, and constructing a spatial distance matrix based on the calculation results of all spatial distances;

[0022] AS32, the spatial distance matrix is ​​sequentially passed through a plurality of convolution blocks, the flattening layer, the first fully connected layer, the second fully connected layer and the first output layer for feature extraction to obtain a spatial proximity representation.

[0023] Furthermore, in the time proximity neural network of step S3, the specific processing flow is as follows:

[0024] BS31. In the second input layer, a time field of a data record in the data set is used as a time sampling point, and the time field of the sample point to be estimated is subtracted from each time sampling point to obtain a time neighbor matrix; wherein the time neighbor matrix is ​​composed of time neighbor vectors, and each time neighbor vector is the result of subtracting the time field of the sample point to be estimated from a time sampling point;

[0025] BS32. Pass the temporal proximity matrix through the third fully connected layer, the fourth fully connected layer and the second output layer to obtain a temporal proximity representation.

[0026] Furthermore, the spatial proximity convolutional neural network includes three convolutional blocks, the convolution kernel size of the convolutional layer in each convolutional block is 5×5, the number of convolution kernels of the convolutional layers in the three convolutional blocks are 16, 16 and 32 respectively, the number of nodes in the first fully connected layer is 256, and the number of nodes in the second fully connected layer is 32; in the temporal proximity neural network, the number of nodes in the third fully connected layer is 64, and the number of nodes in the fourth fully connected layer is 32; in the spatiotemporal fusion proximity neural network, the number of nodes in the fifth fully connected layer is 32, and the number of nodes in the sixth fully connected layer is 16; in the spatiotemporal weighted neural network, the number of nodes in the seventh fully connected layer is 10.

[0027] In a second aspect, the present invention provides an atmospheric carbon spatiotemporal neural network regression modeling system, comprising:

[0028] The data acquisition module is used to acquire the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required research area, and to preliminarily preprocess the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required research area to obtain the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after the preliminary preprocessing;

[0029] The data processing module is used to construct a data set after further preprocessing of the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after preliminary preprocessing;

[0030] The result acquisition module is used to input the data set into a trained geographic spatiotemporal weighted regression model, first extract the spatial proximity representation of the data set by a spatial proximity convolutional neural network, then extract the temporal proximity representation of the data set by a temporal proximity neural network, and then fuse the spatial proximity representation and the temporal proximity representation by a spatiotemporal fusion proximity neural network to obtain the spatiotemporal proximity representation, input the spatiotemporal proximity representation into the spatiotemporal weighted neural network to solve the spatiotemporal weight, apply the least squares method to the data set to obtain the global regression coefficient, multiply the global regression coefficient by the spatiotemporal weight to obtain the final regression coefficient, use the atmospheric XCO2 influencing factor data after further preprocessing as an independent variable, multiply the final regression coefficient by the independent variable to obtain the estimated result of the XCO2 concentration at each spatiotemporal coordinate, and obtain a full coverage map of the atmospheric XCO2 concentration after visualization.

[0031] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the atmospheric carbon spatiotemporal neural network regression modeling method as described in any of the schemes in the first aspect above is implemented.

[0032] In a fourth aspect, the present invention provides a computer electronic device, comprising a memory and a processor;

[0033] The memory is used to store computer programs;

[0034] The processor is used to implement the atmospheric carbon spatiotemporal neural network regression modeling method as described in any scheme of the first aspect when executing the computer program.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. By considering the global spatial position relationship of the required study area, the spatiotemporal proximity characteristics are expressed more accurately, the modeling accuracy of spatiotemporal non-stationary relationships is improved, and the interference of sample strip-like uneven distribution on modeling can be effectively weakened.

[0037] 2. The method of the present invention selects influencing factor data sets according to different research areas, and can be applied to multi-source satellite atmospheric XCO2 monitoring data in different time and space ranges and scales, and has strong portability.

[0038] 3. The full coverage results of atmospheric XCO2 concentration obtained by the method of the present invention can provide a certain reference for the distribution of atmospheric XCO2 concentration, the temporal changes of atmospheric XCO2 concentration, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of the process of the present invention;

[0040] Figure 2 This is a schematic diagram of the distribution of multi-source XCO2 remote sensing data of the present invention;

[0041] Figure 3 It is a structural schematic diagram of the geographic spatiotemporal weighted regression model of the present invention;

[0042] Figure 4 Schematic diagram of the training process of the geographic spatiotemporal weighted regression model of the present invention;

[0043] Figure 5 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0045] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0046] The following is an explanation of the definitions or concepts involved in the present invention: Figure 3As shown in the figure, the geographic spatiotemporal weighted regression model GTCNNWR contains four neural networks, namely the spatial proximity convolutional neural network SPCNN for extracting spatial proximity representation, the temporal proximity neural network TPNN for extracting temporal proximity representation, the spatiotemporal fusion proximity neural network STMPNN for spatiotemporal representation fusion, and the spatiotemporal weighted neural network STWNN for solving spatiotemporal weights. Among them, the spatial proximity convolutional neural network includes an input layer, a convolutional block (convolutional layer, pooling layer), a fully connected layer and an output layer, and the overall spatial information is extracted through the convolutional block and the fully connected layer; the temporal proximity neural network includes an input layer, a fully connected layer and an output layer, and the temporal information is extracted through the fully connected layer; the spatiotemporal fusion proximity neural network fuses the spatial proximity representation and the temporal proximity representation through the input layer, the fully connected layer and the output layer, thereby constructing a unified spatiotemporal proximity representation of the sample points; the spatiotemporal weighted neural network includes an input layer, a fully connected layer (He parameter initialization, a convolutional layer of the PReLU activation function, batch normalization) and an output layer to solve the spatiotemporal weights to construct a geographic spatiotemporal weighted regression model, and finally outputs the estimation result of the geographic spatiotemporal weighted regression model for the XCO2 concentration.

[0047] like Figure 1 As shown, in a preferred implementation of the present invention, the atmospheric carbon spatiotemporal neural network regression modeling method includes the following steps S1 to S3. The specific implementation process is described below.

[0048] The geographic spatiotemporal weighted regression model GTCNNWR provided in this embodiment realizes the modeling and mapping method of the refined spatiotemporal distribution of atmospheric carbon, and obtains the full coverage mapping of atmospheric XCO2 through regression analysis of the spatiotemporal distribution of atmospheric XCO2 and its influencing factors, thereby monitoring the distribution changes of atmospheric XCO2. Figure 1 As shown, an atmospheric carbon spatiotemporal neural network regression modeling method of the present invention comprises the following steps:

[0049] 1. Data acquisition and preliminary preprocessing

[0050] S1. Acquire the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required study area, and preliminarily preprocess the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required study area to obtain the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after the preliminary preprocessing.

[0051] It should be noted that in step S1, the atmospheric XCO2 influencing factor data includes three categories, namely carbon source variables, carbon sink variables and environmental variables; among which, carbon source variables include land use type, night light index and surface temperature, carbon sink variables include vegetation index, evapotranspiration and primary productivity, vegetation index includes vegetation cover, vegetation leaf area index and normalized vegetation index, and environmental variables include ground elevation and surface roughness.

[0052] It should be noted that in step S1 of the present invention, the acquired atmospheric XCO2 time series remote sensing monitoring data includes observation data of multi-source satellites such as OCO-2 and OCO-3, and the XCO2 observation data of OCO-2 and OCO-3 satellites are orbital data of a 16-day period, with a spatial resolution of 2.25km×1.29km. In this embodiment, when the atmospheric XCO2 time series remote sensing monitoring data of the required study area is preliminarily preprocessed, the atmospheric XCO2 time series remote sensing monitoring data with reliable quality (the quality mark is zero) is first screened out, and after the quality screening is completed, standardization processing and multi-source data fusion are performed, and finally the atmospheric XCO2 time series remote sensing monitoring data after preliminary preprocessing is obtained.

[0053] In this embodiment, if Figure 2 As shown in the figure, the atmospheric XCO2 time series remote sensing monitoring data from multiple source satellites including the Orbiting Carbon Observatory-2 (OCO-2), the Orbiting Carbon Observatory-3 (OCO-3), and the Greenhouse Gas Observation Satellite GOAST are obtained. After data fusion, the temporal and spatial coverage of the data can be effectively improved, and more complete atmospheric XCO2 time series remote sensing monitoring data can be obtained. Figure 2 In the equation, XCO2 represents the column-averaged dry-air mixing ratio of atmospheric carbon dioxide.

[0054] 2. Further data preprocessing and dataset construction

[0055] S2. After further preprocessing, the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data are constructed into a data set.

[0056] It should be noted that in step S2 of the present invention, the specific process of further pretreatment is as follows:

[0057] S21. According to the required research area, the range of atmospheric XCO2 time series remote sensing monitoring data and the range of atmospheric XCO2 influencing factor data after preliminary preprocessing are trimmed to obtain atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data within the research area.

[0058] In step S21 of this embodiment, the time-space position matching is first performed to ensure that the atmospheric XCO2 time series remote sensing monitoring data and the atmospheric XCO2 influencing factor data are in the same time-space coordinate system, and then the atmospheric XCO2 time series remote sensing monitoring data and the atmospheric XCO2 influencing factor data are cropped to retain the data within the required research area.

[0059] S22. The required study area is divided into a uniform grid according to a predetermined modeling spatiotemporal resolution scale, and each grid unit represents a local area in the required study area. The atmospheric XCO2 time series remote sensing monitoring data and the atmospheric XCO2 influencing factor data within the required study area are respectively associated with the uniform grid according to the geographic spatiotemporal location information. The linear correlation between the atmospheric XCO2 time series remote sensing monitoring data and the atmospheric XCO2 influencing factor data within the required study area is calculated by the Pearson correlation coefficient.

[0060] In step S22 of this embodiment, the modeling spatiotemporal resolution scale is determined according to the spatiotemporal resolution of the data. The modeling spatiotemporal resolution scale can be generally set between the maximum value of the spatial resolution grid length and the minimum value of the spatial resolution of the atmospheric XCO2 time series remote sensing monitoring data and the atmospheric XCO2 influencing factor data. The appropriate modeling spatiotemporal resolution scale can also be selected as needed, and then the required research scope is divided into uniform grids according to the determined modeling spatiotemporal resolution scale, and the above two types of data are respectively associated with the uniform grid according to the geographic spatiotemporal location information. Then, the linear correlation between each type of influencing factor in the atmospheric XCO2 influencing factor data and the atmospheric XCO2 time series remote sensing monitoring data is calculated using the Pearson correlation coefficient as a reference for the screening of atmospheric XCO2 influencing factor data.

[0061] The calculation of the Pearson correlation coefficient belongs to the prior art. In this embodiment, it can be calculated in the following way:

[0062]

[0063]

[0064] in, represents the Pearson correlation coefficient, which conforms to the t-2 degree of freedom Distribution, by consulting The distribution table can test whether the Pearson correlation between the atmospheric XCO2 influencing factor data and the atmospheric XCO2 time series remote sensing monitoring data is significant; Indicates the spatial and temporal coordinates of each type of influencing factor in the atmospheric XCO2 influencing factor data The data value at Represents the atmospheric XCO2 time series remote sensing monitoring data in time and space coordinates The data value at It represents the average value of each type of influencing factor in the atmospheric XCO2 influencing factor data; It represents the average value of atmospheric XCO2 time series remote sensing monitoring data; Represents the number of space-time coordinates.

[0065] S23. The atmospheric XCO2 influencing factor data within the required study area are screened based on the linear correlation calculation results, and a data set is constructed using the screened atmospheric XCO2 influencing factor data and the atmospheric XCO2 time series remote sensing monitoring data within the study area.

[0066] In step S23 of this embodiment, the specific process of screening the atmospheric XCO2 impact factor data within the required research area by the linear correlation calculation results is as follows: a linear correlation threshold is set in advance, and the impact factors in the atmospheric XCO2 impact factor data whose linear correlation calculation results are greater than or equal to the linear correlation threshold are retained to obtain the screened atmospheric XCO2 impact factor data.

[0067] In step S23 of this embodiment, due to the different required study areas, the relationship between each type of influencing factor in the atmospheric XCO2 influencing factor data and the XCO2 concentration may be different. It is necessary to select the influencing factor index for the required study area to construct a data set suitable for the study area, which includes the screened atmospheric XCO2 influencing factor data and the atmospheric XCO2 time series remote sensing monitoring data within the study area.

[0068] 3. Use the divided uniform grid to express the spatial characteristics of the entire study area

[0069] S3. Input the data set into the trained geographic spatiotemporal weighted regression model GTCNNWR. First, the spatial proximity representation of the data set is extracted by the spatial proximity convolutional neural network SPCNN. Then, the temporal proximity representation of the data set is extracted by the temporal proximity neural network TPNN. Then, the spatial proximity representation and the temporal proximity representation are fused by the spatiotemporal fusion proximity neural network STMPNN to obtain the spatiotemporal proximity representation. The spatiotemporal proximity representation is input into the spatiotemporal weighted neural network STWNN to solve the spatiotemporal weights. The data set is applied with the least squares method to obtain the global regression coefficient. , the global regression coefficient With time and space weight Multiply to get the final regression coefficient , used to characterize the spatiotemporal relationship between XCO2 and various influencing factors, and the atmospheric XCO2 influencing factor data after further preprocessing was used as the independent variable, and the final regression coefficient was compared with the independent variable Multiply them together to get the estimated result of XCO2 concentration at each time and space coordinate , and after visualization, a full coverage map of atmospheric XCO2 concentration is obtained.

[0070] Among them, the spatial proximity convolutional neural network SPCNN includes a first input layer, a convolution block, a flattening layer, a first fully connected layer with a ReLU activation function, a second fully connected layer with a ReLU activation function and a first output layer. Each convolution block is composed of a convolution layer with He parameter initialization and a ReLU activation function, a batch normalization and a pooling layer cascaded in sequence; the temporal proximity neural network TPNN is composed of a second input layer, a third fully connected layer with a ReLU activation function, a fourth fully connected layer with a ReLU activation function and a second output layer cascaded in sequence; the spatiotemporal fusion proximity neural network STMPNN is composed of a third input layer, a fifth fully connected layer with He parameter initialization, Dropout, PReLU activation function, a sixth fully connected layer with He parameter initialization, Dropout, PReLU activation function and a third output layer cascaded in sequence; the spatiotemporal weighted neural network is composed of a fourth input layer, a seventh fully connected layer and a fourth output layer cascaded in sequence.

[0071] It should be noted that in the spatial proximity convolutional neural network SPCNN in step S3 of the present invention, the specific processing flow is as follows:

[0072] AS31, taking the geographic spatial location information as the input of the first input layer, taking the center point of each grid cell as the spatial sampling point, and taking the center point of each grid cell as the spatial sampling point. Line The center point of the grid cell of the column As the spatial position of the sample point to be estimated, the spatial distances between the sample point to be estimated and the remaining spatial sampling points except the sample point to be estimated are calculated in sequence, and a spatial distance matrix is ​​constructed based on the calculation results of all spatial distances.

[0073] In this embodiment, the spatial distance matrix The specific form is as follows:

[0074]

[0075] Each element in the spatial distance matrix represents the spatial distance between the sample point to be estimated and the remaining spatial sampling points except the sample point to be estimated.

[0076] Furthermore, the sample points to be estimated and located in Line The center point of the grid cell of the column The spatial distance is recorded as , its mathematical expression is as follows:

[0077]

[0078] in, Represents the sample points to be estimated The spatial position coordinates of .

[0079] AS32, the spatial distance matrix is ​​sequentially subjected to several convolution blocks, the flattening layer, the first fully connected layer, the second fully connected layer and the first output layer for feature extraction to obtain a spatial proximity representation .

[0080] It should be noted that in step S3 of this embodiment, the spatial proximity convolutional neural network SPCNN has a multi-layer structure, specifically including three convolution blocks, a flattening layer, two fully connected layers, an input layer and an output layer. The number of convolution blocks here can be selected according to actual needs and is not limited in the present invention.

[0081] It should be noted that the temporal proximity neural network TPNN in step S3 of the present invention is used to extract the temporal characteristics of the temporal proximity vector (TPV), and uses the temporal proximity vector of the entire research time range as a reference, so as to construct the temporal proximity representation of the sample points.

[0082] In the temporal proximity neural network TPNN, the specific processing flow is as follows:

[0083] BS31. In the second input layer, the time field of a data record in the data set is taken as a time sampling point, and the time field of the sample point to be estimated is subtracted from each time sampling point to obtain a time neighbor matrix; wherein the time neighbor matrix is ​​composed of time neighbor vectors, and each time neighbor vector is the result of subtracting the time field of the sample point to be estimated from a time sampling point.

[0084] BS32, the time proximity matrix After the third fully connected layer, the fourth fully connected layer and the second output layer, the temporal proximity representation is obtained .in, Indicates time proximity vector, correspondingly, The time neighbor vectors are the time fields of the sample points to be estimated. With The result of subtracting the time sampling points.

[0085] It should be noted that in step S3 of this embodiment, the temporal proximity neural network TPNN has a 4-layer structure, specifically including two fully connected layers, an input layer and an output layer. The number of fully connected layers here can be selected according to actual needs and is not limited in the present invention.

[0086] It should be noted that in the spatiotemporal weighted neural network of step S3 of the present invention, the specific processing flow is as follows: the spatiotemporal proximity representation After passing through the fourth input layer, the seventh fully connected layer and the fourth output layer, the spatiotemporal weights are obtained. .in, Represents the space-time coordinates The longitude, latitude and time location of the location.

[0087] It should be noted that when training the geographic spatiotemporal weighted regression model, Figure 4 As shown, it is necessary to combine the current data set suitable for the study area to complete the training. The dependent variable of any spatial factor of the geographic spatiotemporal weighted regression model (GTCNNWR) is the XCO2 concentration value measured by the remote sensing satellite. , and its corresponding space-time coordinates The estimated results at for:

[0088]

[0089] in, Representing space-time coordinates The independent variable at Representing space-time coordinates The final regression coefficient at ; Represented in space-time coordinates Estimation results of XCO2 concentration using the geographic spatiotemporal weighted regression model.

[0090] Calculate space-time coordinates XCO2 concentration values ​​measured by remote sensing satellites The corresponding estimation results The mean square error (MSE) between is used as the loss function to update the parameters of the geographic spatiotemporal weighted regression model:

[0091]

[0092] The present invention will now use a specific example to demonstrate the application effect of the atmospheric carbon spatiotemporal neural network regression modeling method described in S1 to S3 in the above embodiments on a specific data set, so as to facilitate understanding of the essence of the present invention. Example

[0093] The specific implementation process of the atmospheric carbon spatiotemporal neural network regression modeling method used in this embodiment is as described above and will not be repeated here.

[0094] In this embodiment, the above method is implemented using the public atmospheric carbon satellite OCO-2 as the source data. First, the original OCO-2 data is obtained from the NASA official website, with a time span from January 2018 to December 2022, and the corresponding influencing factor data is collected to form a data set. The records with missing attribute values ​​are cleaned, and a new integer time element column is created for each table, and the time field is a year-month combination value such as 2020-1. Then the data within the administrative scope of Shanghai is screened out. The screened data is further sparsely sampled, and 50*40*20 center points and data within their specific ranges are found in three dimensions of longitude, latitude, and time. The average value of the data within the specific range of each center point is calculated and assigned to the parameter value of the center point. Taking into account factors such as the lack of original data, 3746 representative data are finally obtained to form a data set of feature samples for subsequent training and testing of geographic spatiotemporal weighted regression models.

[0095] This embodiment adopts a ten-fold cross-validation strategy to train the geographic spatiotemporal weighted regression model. Before the geographic spatiotemporal weighted regression model is trained, 90% of the data set is randomly divided into a cross-validation set, 10% is divided into a test set, and the cross-validation set is evenly divided into 10 groups. In each round of cross-validation, 9 groups in the cross-validation set are taken as training sets, and the remaining 1 group is used as a validation set. After completing 10 rounds of cross-validation, each group in the cross-validation set will generate a result as a validation set. Before training the geographic spatiotemporal weighted regression model, the time distance matrix and space distance matrix of the training set, validation set, and test set are calculated respectively, and normalized. The normalized time distance matrix and space distance matrix are used as the input of the geographic spatiotemporal weighted regression model.

[0096] When training the geographic spatiotemporal weighted regression model, the initial learning rate is set to 0.01 and the optimizer is Adam. The spatial proximity convolutional neural network in the geographic spatiotemporal weighted regression model includes three convolution blocks (where the convolution kernel size of each convolution layer is 5×5, the number of convolution kernels is 16, 16 and 32, and the activation function is ReLU), a flattening layer and two fully connected layers (the number of nodes is 256 and 32, and the activation function is ReLU). The temporal proximity neural network includes two fully connected layers (the number of nodes is 64 and 32, and the activation function is ReLU). After splicing the outputs of the two, they pass through the spatiotemporal fusion proximity neural network, which contains two fully connected layers (the number of nodes is 32 and 16, and the activation function is PReLU and Dropout regularization). The spatiotemporal weighted neural network contains a fully connected layer (the number of nodes is 10) to obtain the final weight prediction value. The geographic spatiotemporal weighted regression model was trained using the aforementioned data set until convergence, and then the performance of the geographic spatiotemporal weighted regression model was tested using the test set, and the optimal result was selected to obtain the weight relationship between atmospheric XCO2 and influencing factors.

[0097] In this embodiment, in order to evaluate the performance of the geographic spatiotemporal weighted regression model obtained by the above training, indicators such as determination coefficient (R2), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are selected to evaluate the geographic spatiotemporal weighted regression model. The results are shown in Table 1. The R2 of the geographic spatiotemporal weighted regression model on the training set, validation set, and test set is higher than 0.85, indicating that the geographic spatiotemporal weighted regression model of the present invention has good accuracy.

[0098] Table 1 Results of geographic spatiotemporal weighted regression model

[0099]

[0100] In summary, the atmospheric carbon spatiotemporal neural network regression modeling method of the present invention constructs a geographic spatiotemporal weighted regression model by designing a unified characterization algorithm for spatiotemporal proximity, which can effectively handle the temporal information of atmospheric XCO2 in the time dimension and the uneven sampling problem in the spatial dimension. Among them, the sampling problem in the spatial dimension is to divide the target research area into grids, calculate the distance between each test point and the grid, so as to greatly reduce the impact of the distribution of sampling points on the results, and can be adaptively used for full coverage distribution modeling of multi-source satellite atmospheric XCO2 monitoring data in different spatiotemporal ranges and scales. Therefore, the method of the present invention has high data modeling accuracy and strong generalization ability, and is of great significance for the spatiotemporal big data mining and analysis of atmospheric carbon emissions.

[0101] It should also be noted that the atmospheric carbon spatiotemporal neural network regression modeling method in the above embodiment can essentially be executed by a computer program or module. Therefore, similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides an atmospheric carbon spatiotemporal neural network regression modeling system corresponding to the atmospheric carbon spatiotemporal neural network regression modeling method provided in the above embodiment, such as Figure 5 As shown, it includes:

[0102] The data acquisition module is used to acquire the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required research area, and to preliminarily preprocess the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required research area to obtain the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after the preliminary preprocessing;

[0103] The data processing module is used to construct a data set after further preprocessing of the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after preliminary preprocessing;

[0104] The result acquisition module is used to input the data set into a trained geographic spatiotemporal weighted regression model, first extract the spatial proximity representation of the data set by a spatial proximity convolutional neural network, then extract the temporal proximity representation of the data set by a temporal proximity neural network, and then fuse the spatial proximity representation and the temporal proximity representation by a spatiotemporal fusion proximity neural network to obtain the spatiotemporal proximity representation, input the spatiotemporal proximity representation into the spatiotemporal weighted neural network to solve the spatiotemporal weight, apply the least squares method to the data set to obtain the global regression coefficient, multiply the global regression coefficient by the spatiotemporal weight to obtain the final regression coefficient, use the atmospheric XCO2 influencing factor data after further preprocessing as an independent variable, multiply the final regression coefficient by the independent variable to obtain the estimated result of the XCO2 concentration at each spatiotemporal coordinate, and obtain a full coverage map of the atmospheric XCO2 concentration after visualization.

[0105] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the atmospheric carbon spatiotemporal neural network regression modeling method provided in the above embodiment, which includes a memory and a processor;

[0106] The memory is used to store computer programs;

[0107] The processor is used to implement the atmospheric carbon spatiotemporal neural network regression modeling method in the above embodiment when executing the computer program.

[0108] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0109] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the atmospheric carbon spatiotemporal neural network regression modeling method provided in the above embodiment, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the atmospheric carbon spatiotemporal neural network regression modeling method in the above embodiment can be implemented.

[0110] It is understandable that the above storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0111] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0112] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0113] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for atmospheric carbon spatiotemporal neural network regression modeling, characterized in that: The following steps are involved: S1. Acquire the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required study area, and preliminarily preprocess the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required study area to obtain the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after the preliminary preprocessing; S2, constructing a data set after further preprocessing the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after preliminary preprocessing; S3, input the data set into a trained geographic spatiotemporal weighted regression model, first extract the spatial proximity representation of the data set by a spatial proximity convolutional neural network, then extract the temporal proximity representation of the data set by a temporal proximity neural network, and then fuse the spatial proximity representation and the temporal proximity representation by a spatiotemporal fusion proximity neural network to obtain a spatiotemporal proximity representation, input the spatiotemporal proximity representation into a spatiotemporal weighted neural network to solve the spatiotemporal weight, apply the least squares method to the data set to obtain a global regression coefficient, multiply the global regression coefficient by the spatiotemporal weight to obtain a final regression coefficient, use the atmospheric XCO2 influencing factor data after further preprocessing as an independent variable, multiply the final regression coefficient by the independent variable, obtain an estimated result of the XCO2 concentration at each spatiotemporal coordinate, and obtain a full coverage map of the atmospheric XCO2 concentration after visualization; In step S2, the specific process of further preprocessing is as follows: S21, according to the required research area, the range of atmospheric XCO2 time series remote sensing monitoring data and the range of atmospheric XCO2 impact factor data after preliminary preprocessing are cut to obtain atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 impact factor data within the research area; S22. Divide the required study area into a uniform grid according to a predetermined modeling spatiotemporal resolution scale, wherein each grid unit represents a local area in the required study area. Associate the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data within the required study area with the uniform grid according to the geographic spatiotemporal location information, and calculate the linear correlation between the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data within the required study area by using the Pearson correlation coefficient. S23. The atmospheric XCO2 influencing factor data within the required study area are screened based on the linear correlation calculation results, and a data set is constructed using the screened atmospheric XCO2 influencing factor data and the atmospheric XCO2 time series remote sensing monitoring data within the study area.

2. The atmospheric carbon spatiotemporal neural network regression modeling method according to claim 1, characterized in that: In step S1, the atmospheric XCO2 influencing factor data includes three categories, namely carbon source variables, carbon sink variables and environmental variables; wherein, carbon source variables include land use type, night light index and surface temperature, carbon sink variables include vegetation index, evapotranspiration and primary productivity, vegetation index includes vegetation coverage, vegetation leaf area index and normalized vegetation index, and environmental variables include ground elevation and surface roughness.

3. The atmospheric carbon spatiotemporal neural network regression modeling method according to claim 1, characterized in that: In step S23, the specific process of screening the atmospheric XCO2 impact factor data within the required research area by the linear correlation calculation results is as follows: a linear correlation threshold is set in advance, and the impact factors in the atmospheric XCO2 impact factor data whose linear correlation calculation results are greater than or equal to the linear correlation threshold are retained to obtain the screened atmospheric XCO2 impact factor data.

4. The atmospheric carbon spatiotemporal neural network regression modeling method according to claim 1, characterized in that: The spatial proximity convolutional neural network includes a first input layer, a convolution block, a flattening layer, a first fully connected layer with a ReLU activation function, a second fully connected layer with a ReLU activation function, and a first output layer. Each convolution block is composed of a convolution layer with He parameter initialization and a ReLU activation function, a batch normalization, and a pooling layer cascaded in sequence; the temporal proximity neural network is composed of a second input layer, a third fully connected layer with a ReLU activation function, a fourth fully connected layer with a ReLU activation function, and a second output layer cascaded in sequence; the spatiotemporal fusion proximity neural network is composed of a third input layer, a fifth fully connected layer with He parameter initialization, Dropout, PReLU activation function, a sixth fully connected layer with He parameter initialization, Dropout, PReLU activation function, and a third output layer cascaded in sequence; the spatiotemporal weighted neural network is composed of a fourth input layer, a seventh fully connected layer, and a fourth output layer cascaded in sequence.

5. The atmospheric carbon spatiotemporal neural network regression modeling method according to claim 4, characterized in that: In the spatial proximity convolutional neural network of step S3, the specific processing flow is as follows: AS31, taking the geographic spatial position information as the input of the first input layer, taking the center point of each grid cell as the spatial sampling point, taking the center point of the grid cell located in the pth row and the qth column as the spatial position of the sample point to be estimated, calculating the spatial distances between the sample point to be estimated and the remaining spatial sampling points except the sample point to be estimated in sequence, and constructing a spatial distance matrix based on the calculation results of all spatial distances; AS32, the spatial distance matrix is ​​sequentially passed through a plurality of convolution blocks, the flattening layer, the first fully connected layer, the second fully connected layer and the first output layer for feature extraction to obtain a spatial proximity representation.

6. The atmospheric carbon spatiotemporal neural network regression modeling method according to claim 4, characterized in that: In the temporal proximity neural network of step S3, the specific processing flow is as follows: BS31. In the second input layer, a time field of a data record in the data set is used as a time sampling point, and the time field of the sample point to be estimated is subtracted from each time sampling point to obtain a time neighbor matrix; wherein the time neighbor matrix is ​​composed of time neighbor vectors, and each time neighbor vector is the result of subtracting the time field of the sample point to be estimated from a time sampling point; BS32. Pass the temporal proximity matrix through the third fully connected layer, the fourth fully connected layer and the second output layer to obtain a temporal proximity representation.

7. An atmospheric carbon spatiotemporal neural network regression modeling system, characterized in that: include: The data acquisition module is used to acquire the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required research area, and to preliminarily preprocess the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data of the required research area to obtain the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after the preliminary preprocessing; The data processing module is used to construct a data set after further preprocessing of the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data after preliminary preprocessing; A result acquisition module is used to input the data set into a trained geographic spatiotemporal weighted regression model, first extract the spatial proximity representation of the data set by a spatial proximity convolutional neural network, then extract the temporal proximity representation of the data set by a temporal proximity neural network, and then fuse the spatial proximity representation and the temporal proximity representation by a spatiotemporal fusion proximity neural network to obtain the spatiotemporal proximity representation, input the spatiotemporal proximity representation into the spatiotemporal weighted neural network to solve the spatiotemporal weight, apply the least squares method to the data set to obtain a global regression coefficient, multiply the global regression coefficient by the spatiotemporal weight to obtain a final regression coefficient, use the atmospheric XCO2 influencing factor data after further preprocessing as an independent variable, multiply the final regression coefficient by the independent variable, obtain the estimated result of the XCO2 concentration at each spatiotemporal coordinate, and obtain a full coverage map of the atmospheric XCO2 concentration after visualization; In the data processing module, the specific process of further preprocessing is as follows: S21, according to the required research area, the range of atmospheric XCO2 time series remote sensing monitoring data and the range of atmospheric XCO2 impact factor data after preliminary preprocessing are cut to obtain atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 impact factor data within the research area; S22. Divide the required study area into a uniform grid according to a predetermined modeling spatiotemporal resolution scale, wherein each grid unit represents a local area in the required study area. Associate the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data within the required study area with the uniform grid according to the geographic spatiotemporal location information, and calculate the linear correlation between the atmospheric XCO2 time series remote sensing monitoring data and atmospheric XCO2 influencing factor data within the required study area by using the Pearson correlation coefficient. S23. The atmospheric XCO2 influencing factor data within the required study area are screened based on the linear correlation calculation results, and a data set is constructed using the screened atmospheric XCO2 influencing factor data and the atmospheric XCO2 time series remote sensing monitoring data within the study area.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the atmospheric carbon spatiotemporal neural network regression modeling method according to any one of claims 1 to 7 is implemented.

9. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the atmospheric carbon spatiotemporal neural network regression modeling method as described in any one of claims 1 to 7 when executing the computer program.

Images