Method for improving urban green land biological source emission based on high-resolution global mode

Abstract

The invention provides a method for improving urban green land biological source emission based on a high-resolution global mode, and provides a technical scheme for optimizing a biological source emission list by using high-resolution land coverage data aiming at the problem of urban green land emission deficiency in a current numerical model. The technologies comprise a weak supervision global land coverage mapping method, spectral information of images can be fused, and features extracted by a visual large model are used for drawing high-resolution land coverage images; the vegetation function type matching method is used for converting the high-resolution land coverage data into vegetation function types; and preparing a high-resolution numerical mode input file, converting the file into an input format suitable for a high-resolution earth system mode, carrying out numerical simulation, and estimating global biological source emission, especially urban green space emission.

Description

Technical Field

[0001] The present invention relates to the technical field of biogenic emission inventory preparation, and in particular to a method for improving biogenic emission of urban green space based on a high-resolution global model. Background Art

[0002] Air pollution is a major factor affecting human health and ecosystems. Typical air pollutants include ozone and fine particles. As an important precursor, one of the important sources of volatile organic compounds is biogenic, which accounts for 90% of the total global volatile organic compound emissions. Biogenic volatile organic compounds (BVOCs) mainly include isoprene, terpenes, etc., which have strong photochemical activity, can promote the formation of ozone and secondary organic aerosols, and induce air pollution.

[0003] Forests are the main source of BVOCs, which has been widely recognized in past studies. However, relatively little attention has been paid to BVOC emissions from vegetation in cities, which are crucial to air pollution. Take ozone as an example. The chemical reaction of its precursors requires the presence of volatile organic compounds and nitrogen oxides. BVOC emissions are large in virgin forests, but there is often a lack of nitrogen oxide sources (such as factories and motor vehicles); however, nitrogen oxide emissions are relatively large in cities, and BVOC emissions from urban vegetation can react with nitrogen oxides very effectively to produce ozone. An important reason for ignoring BVOC emissions in urban vegetation (or urban green space) in the past is that the spatial resolution of land cover type data is low, making it difficult to identify vegetation in cities. The production of land cover data relies on remote sensing data, such as the traditional and commonly used Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data with a spatial resolution of 500 meters. This resolution can identify large areas of vegetation such as forests, but the ability to identify vegetation in cities is very low. Urban vegetation is often scattered in different areas, such as streets and residential areas. Therefore, biogenic emissions based on MODIS satellite land cover data are basically unable to include emissions from urban green spaces.

[0004] To solve this problem, improving the spatial resolution of land cover is one of the important prerequisites. With the advancement of remote sensing technology, Landsat series (30 meters) and Sentinel series (10 meters) satellites provide higher spatial resolution. At present, from the perspective of spatial resolution, 10-meter high-resolution data products include ESA 10-meter land cover data (such as ESA World Cover 10m), Tsinghua University 10-meter land cover data (FROM-GLC10), etc. Although these data have a spatial resolution of up to 10 meters, trees are uniformly classified as one type (Trees). However, the emission factors of different vegetation types vary greatly. On the basis of improving the resolution, it is also necessary to identify different types of vegetation. A recent study has produced global 30-meter land cover data (GLC_FCS30-2020; global 30-m landcover classification with a fine classification system in 2020) based on ESA's Sentinel-2 satellite data, including 30 land cover types (trees are divided into several categories, such as broad-leaved forests, coniferous forests, etc.), but the identification of vegetation in cities is still relatively small (for example, taking Beijing as an example, Figure 2 c). Based on this, it is necessary to use the high spatial resolution (10 meters) Sentinel-2 data, combined with its own spectral feature extraction and visual models that can identify long-distance image dependencies (such as Vision Transformer), to produce 10-meter resolution, 30-category refined land cover type data, which will strongly support high-precision urban green space biogenic emission estimation and subsequent numerical simulation of air pollution.

[0005] In addition to land cover, the application of high-resolution numerical models is a necessary means to improve the accuracy of biogenic emissions and their impact assessment on atmospheric pollution. On the one hand, biogenic emissions are closely related to extreme weather such as high temperature heat waves. High-resolution numerical models are expected to improve the simulation of extreme weather such as high temperature heat waves, better capture the process of heat waves promoting biogenic emissions, and then more reasonably simulate the impact of biogenic emissions on the concentration of atmospheric pollutants. In our previous work, we have achieved high-resolution Earth system (25 km atmosphere) simulation based on domestic multi-core heterogeneous supercomputers, and achieved improved ozone simulation (patent number: ZL 2023 1 1654525.2). This patent will improve the resolution of land cover data based on the previously optimized high-resolution Earth system model, and the high-precision land cover data can be efficiently applied in the high-resolution model after processing, so as to improve the accuracy of biogenic emissions, especially urban area emission simulation. Summary of the invention

[0006] In order to make up for the shortcomings of the existing technology, the present invention provides a method for improving the biogenic emissions of urban green spaces based on a high-resolution global model. Aiming at the problem of missing urban green space emissions in the current numerical model, a technical solution for optimizing the biogenic emissions inventory using high-resolution land cover data is proposed. These technologies include a weakly supervised global land cover mapping method, which can integrate the spectral information of the image itself and use the features extracted by the visual large model to draw high-resolution land cover images; a vegetation functional type (PFT) matching method, which is used to convert high-resolution land cover data into vegetation functional types; high-resolution numerical model input file preparation, which converts the above files into an input format suitable for high-resolution earth system models, and conducts numerical simulations to estimate global biogenic emissions, especially urban green space emissions. The present invention can effectively capture urban green space emissions and improve the accuracy of urban green space emission inventories and atmospheric pollution simulations.

[0007] The present invention is implemented by the following technical solution: a method for improving urban green space biogenic emissions based on a high-resolution global model, specifically comprising the following steps: Step S1: Data preprocessing: Six multispectral images from Sentinel-2 were selected, including three visible light and three infrared bands. After the data was selected, the infrared band data was preprocessed using the Google Earth Engine platform to interpolate the infrared band data to a 10-meter resolution, keeping the same resolution as the visible light band, thus obtaining the Sentinel image data with a 10-meter resolution; Step S2: pixel feature extraction: For each Sentinel-2 pixel’s six bands, we extract its own multispectral features and those of its surroundings. Next, we use Vision Transformer, which is pre-trained based on previous remote sensing data, to extract high-level features of the same image. This includes the following steps: Step S2-1, when extracting pixel features, use the sliding window method to focus on the extraction of spectral and texture details of the image; for each pixel, extract the 6-band spectral information of itself and the surrounding cross area, a total of 9 pixels; then connect the obtained spectral information to form a long sequence of 9 pixels × 6 bands with a length of 54; Step S2-2, using a pre-trained deep learning model for feature extraction, using the visual self-attention mechanism model encoder provided by Vision Transformer to extract high-level semantic features in remote sensing images; the encoder captures a wide range of image contexts and long-distance dependencies in the image; each image is cropped into multiple slices with overlapping parts, and then the inference calculation of the visual self-attention mechanism model encoder is performed on each slice to obtain a feature map with the same spatial size; finally, according to the above cropping mode, the feature maps of these slices are spliced ​​back to the original image size; Step S3: feature fusion: A feature fusion method is used to balance the local, low-level features from the pixel feature extraction branch with the global, high-level features extracted from the deep learning model to obtain the best clustering results. The specific steps include: merging the spectral data from the pixel feature extraction branch and the data from the deep learning model branch to form a feature vector with a length of 64, and normalizing this feature matrix to construct input features that can be directly used for the clustering algorithm; Step S4: Cluster calculation and 10-meter land cover data image production: After feature extraction, cluster analysis is performed based on the clustering algorithm to generate 300 clustering results. The relationship between the 30-meter GLC_FCS30-2020 resolution product and the clustering results is analyzed to find the corresponding land cover type for each cluster. Furthermore, based on the data identified as green space in the 10-meter resolution ESAWorldCover data, the previously generated 10-meter resolution land cover classification results are rechecked. If the classification results of some areas are not marked as green space in the generated 10-meter data, they are re-judged and reallocated to the green space category based on the spectral characteristics of the area and its surroundings. Automatic clustering and merging are completed to generate 10-meter high-resolution land cover data (WS-GLM1030-PSET; weakly-supervised 10-m 30-class global land cover mapping method with pixel spectral extraction for tree identification), which is stored in the labeled image file format TIFF. Step S5: Integration of 10-meter land cover data to 500-meter resolution The land cover data in TIFF format has a resolution of 10 meters. It is further integrated into multiple resolutions for model application as needed. The data is integrated into 500-meter grids. The integration method specifically includes the following steps: for each 500-meter grid, the 10-meter resolution grids contained in this grid are identified, and the percentages are recorded according to the land cover types to which these grids belong; Step S6: Matching of land cover data to vegetation functional types: The 500-meter land cover data is matched to the vegetation functional types required by the numerical model, which specifically includes the following steps: first, the 500-meter land cover data is read and matched to the 16 vegetation functional types used in the numerical simulation; Step S7: Comparison of vegetation functional types: The high-resolution vegetation functional type completed in step S6 is compared with the traditional MODIS and vegetation functional type data produced by GLC_FCS30-2020 and ESAWorldCover to evaluate the improvement of the newly produced high-resolution and refined land cover type data on vegetation, especially urban green space identification; Step S8: Interpolation of 500-meter vegetation functional types to numerical model resolution After completing the 500-meter vegetation functional type processing, these data are further interpolated into a 25-kilometer grid of spectral elements (SpectralElement), and the newly generated PFT data are converted into a NetCDF format compatible with the original input file using the NCL tool; Step S9: Create high-resolution earth system model SW-HRESM input file: A program was written based on Python and the climate data manipulation software CDO to extract the PFT data from the newly generated NetCDF file and replace the corresponding variables in the original SW-HRESM input file, ensuring that the variable name PCT_PFT, type double, dimension, and unit (%) were consistent with those in the original file; Step S10: Preparation of high-resolution biogenic emission inventory: Using the high-resolution Earth system model SW-HRESM and its land module's biogenic gas and aerosol emission estimation model MEGAN, the newly generated input file in step S9 is input into the model for a test run, and then SW-HRESM is run to generate a global high-resolution biogenic emission inventory with improved accuracy; Step S11, air pollution simulation and subsequent analysis: Based on the newly generated biogenic emission inventory, high-resolution Earth system model simulations are carried out, and then programs are written to assess the impact of global biogenic emissions, especially emissions from urban green spaces, on atmospheric pollutants.

[0008] As a preferred solution, in step S1, 6 spectral images are selected from Sentinel-2, with a total data of 87TB; the visible light is red, green and blue, with a resolution of 10 meters; the resolution of the three infrared images is 10 meters to 20 meters.

[0009] As a preferred solution, the data processing software in step S9 also includes NCO.

[0010] Due to the adoption of the above technical solutions, the present invention has the following beneficial effects compared with the prior art: a weakly supervised global land cover mapping method is proposed, which integrates the spectral information of the Sentinel-2 (satellite data) image itself and the image features extracted by the Vision Transformer (ViT), and produces a set of high-resolution land cover data (WS-GLM1030-PSET), which effectively improves the accuracy of land cover identification and reduces the pressure of data storage and processing. By using high-resolution land cover data, PFT is improved, the simulation of BVOC emissions in the global model SW-HRESM is optimized, and the biogenic emissions of urban green spaces are more accurately estimated. At the same time, by improving the BVOC emissions of urban green spaces, the model can also better simulate the generation of ozone. It is of great significance to formulate effective ozone pollution control strategies and improve urban air quality.

[0011] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 Schematic diagram of the feature extraction process based on pixel neighborhood and visual Transformer Figure 2 Land cover mapping based on Sentinel images, MODIS, GLC_FCS30-2020, and WS-GLM1030-PSET: Taking Beijing (Grid No. 190) as an example. DETAILED DESCRIPTION

[0013] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0014] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0015] Combine the following Figure 1 to Figure 2The method for improving the biogenic emissions of urban green spaces based on a high-resolution global model according to an embodiment of the present invention is specifically described. A weakly supervised global land cover mapping method is proposed, which improves the accuracy of land cover recognition by fusing the spectral information of Sentinel-2 images and the high-level features extracted by the Vision Transformer. By utilizing the details of multispectral data and the global semantic information of the deep learning model, the edge detection performance of land cover mapping is improved, enabling the model to generate more accurate boundaries of objects, thereby reducing the burden of data storage and processing and improving the recognition effect, and producing global 10-meter land cover data (WS-GLM1030-PSET).

[0016] The 10-meter land cover data WS-GLM1030-PSET was upscaled to 500 meters for comparative evaluation with the traditional commonly used MODIS satellite data.

[0017] The 500-meter land cover data was converted into 25-kilometer spectral elements, and PFT was calculated based on the land cover data and provided as input files to the high-resolution earth system model (SW-HRESM). The high-resolution earth system model is able to estimate the biogenic emissions of urban green spaces, especially urban green space emissions.

[0018] like Figure 1 to Figure 2 As shown, the present invention proposes a method for improving biogenic emissions from urban green spaces based on a high-resolution global model, which specifically includes the following steps: Step S1: Data preprocessing: Selecting 6 multispectral images from Sentinel-2 (total data of 87TB), including three visible light (red, green, blue; RGB; resolution of 10 meters) and three infrared bands (resolution of 10 meters-20 meters), can effectively improve the recognition accuracy and reduce the pressure of data storage and processing. After selecting the data, the Google Earth Engine (GEE) platform is used for preprocessing, and the infrared band data is interpolated to a resolution of 10 meters, which is consistent with the resolution of the visible light band, so as to obtain the Sentinel image data with a resolution of 10 meters; Step S2: pixel feature extraction: For each Sentinel-2 pixel’s six bands, we extract multispectral features of itself and its surroundings. These features with detailed textures can help the model generate accurate boundaries of objects. Secondly, we use Vision Transformer, which is pre-trained based on previous remote sensing data, to extract high-level features of the same image. This deep learning model based on Vision Transformer can capture long-range dependencies in images. It helps the model obtain coarse-grained semantic information from images. The specific steps include: Step S2-1, the process of feature extraction is as follows Figure 1 As shown. When extracting pixel features, the sliding window method is used to focus on the extraction of spectral and texture details of the image; multispectral information can enrich feature characterization and improve the accuracy and robustness of the model. For each pixel, the 6-band spectral information of itself and the surrounding cross area, a total of 9 pixels, is extracted; then the obtained spectral information is connected to form a long sequence of 9 pixels × 6 bands with a length of 54; this sliding window design can improve the performance of edge detection and enable the model to produce clear and sharp edges.

[0019] Step S2-2, using a pre-trained deep learning model for feature extraction, using the visual self-attention mechanism model encoder provided by Vision Transformer to extract high-level semantic features in remote sensing images; the encoder captures a wide range of image contexts and long-distance dependencies in the image; each image is cropped into multiple slices with overlapping parts to meet the memory requirements of the encoder calculation and the continuity of the edges between slices; then based on the visual self-attention mechanism model encoder, each slice is processed into a feature map with the same spatial size; finally, according to the above cropping mode, the feature maps of these slices are spliced ​​back to the original image size; Step S3: feature fusion: A feature fusion method is used to balance the local, low-level features from the pixel feature extraction branch with the global, high-level features extracted from the deep learning model to obtain the best clustering results. The specific steps include: merging the spectral data from the pixel feature extraction branch and the data from the deep learning model branch to form a feature vector with a length of 64, and normalizing this feature matrix to construct input features that can be directly used for the clustering algorithm; Step S4: Cluster calculation and 10-meter land cover data image production: After feature extraction, cluster analysis is performed based on the clustering algorithm (K-means) to generate 300 clustering results. The relationship between the slightly lower resolution product (30 meters, GLC_FCS30-2020) and the clustering results is analyzed to find the corresponding land cover type for each cluster; and based on the data predicted as green space in the 10-meter resolution ESAWorldCover data, the previously generated 10-meter resolution land cover classification results are rechecked. If the classification results of some areas are not marked as green space in the generated 10-meter data, the area is re-judged and reallocated to the green space category based on the spectral characteristics of the area and its surroundings, and the automatic clustering and merging are completed to generate 10-meter high-resolution land cover data (WS-GLM1030-PSET), which is stored in the labeled image file format TIFF; Beijing is selected as an example to compare the high-resolution land cover data generated by Sentinel-2 and this technology, and MODIS and GLC_FCS30-2020 are used as controls, such as Figure 2 As shown, this technology product improves the identification of urban green spaces.

[0020] Step S5: Integration of 10-meter land cover data to 500-meter resolution The resolution of land cover data in TIFF format is 10 meters. It is further integrated into multiple resolutions for model application as needed. In order to compare with the commonly used MODIS satellite data, we integrate the data into 500-meter grids. The integration method specifically includes the following steps: For each 500-meter grid, clarify the 10-meter resolution grids contained in this grid, and record them in percentage form according to the land cover type to which these grids belong; for example, there are 2,500 10-meter grids in a 500-meter grid, of which 250 are deciduous broad-leaved forests, which means that the proportion of deciduous broad-leaved forests recorded in this grid is 10%. All data are comma-separated value (CSV) files.

[0021] Step S6: Matching of land cover data to vegetation functional types: The 500-meter land cover data is matched to the vegetation functional type required by the numerical model, which specifically includes the following steps: first, the 500-meter land cover data is read and matched to the 16 vegetation functional types used in the numerical simulation; for example, among the 16 commonly used vegetation functional types, temperate deciduous broadleaved forest (Broadleaf deciduous temperate trees) is one of the types, and Open deciduous broadleaved forest and Closed deciduousbroadleaved forest in the data are matched according to the latitude band; Step S7: Comparison of vegetation functional types: The high-resolution vegetation functional type completed in step S6 is compared with the traditional MODIS and vegetation functional type data produced by GLC_FCS30-2020 and ESAWorldCover to evaluate the degree of improvement of vegetation, especially urban green space, by the newly produced high-resolution and refined land cover type data. Taking the urban area of ​​Beijing as an example, the WS-GLM1030-PSET land cover image can capture more details of urban green spaces. In MODIS data, due to the resolution limitation of 500 meters, many scattered or small-scale urban green spaces are difficult to be accurately identified, which leads to an underestimation of the area and distribution of urban green spaces.

[0022] Step S8: Interpolation of 500-meter vegetation functional types to numerical model resolution (25 km) After completing the 500-meter vegetation functional type processing, these data were further interpolated into a 25-km grid of spectral elements, and the newly generated PFT data were converted into a NetCDF format compatible with the original input file using the NCL tool; Step S9: Create high-resolution Earth system model (SW-HRESM) input file: Write a program based on Python and climate data operation software CDO to extract PFT data from the newly generated NetCDF file and replace the corresponding variables (variable name PCT_PFT) of the original SW-HRESM input file, ensuring that the variable name, type (double), dimension and unit (%) are consistent with the original file; Step S10: Preparation of high-resolution biogenic emission inventory: Using the high-resolution Earth system model SW-HRESM and its land module's biogenic gas and aerosol emission estimation model (MEGAN), input the newly generated input file in step S9 into the model for a test run to ensure that the MEGAN model can correctly read and use the new data. Then run SW-HRESM to generate a global high-resolution biogenic emission inventory with improved accuracy, especially a global urban green space volatile organic compound emission inventory; Step S11, air pollution simulation and subsequent analysis: Based on the newly generated biogenic emission inventory, high-resolution Earth system model simulations are carried out, and then programs are written to assess the impact of global biogenic emissions, especially urban green space emissions, on air pollutants (such as ozone and fine particulate matter). For example, NCAR Command Language (NCL) is used to interpolate model results (convert spectral element grids into latitude and longitude grids), extract atmospheric pollutants, and draw time series graphs and spatial distribution maps.

[0023] In the description of the present invention, the term "plurality" refers to two or more than two. Unless otherwise clearly defined, the orientation or positional relationship indicated by the terms "upper" and "lower" is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention; the terms "connection", "installation", "fixation", etc. should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0024] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0025] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for improving biogenic emissions from urban green spaces based on a high-resolution global model, characterized in that: The specific steps include: Step S1: Data preprocessing: Six multispectral images from Sentinel-2 were selected, including three visible light and three infrared bands. After the data was selected, the infrared band data was preprocessed using the Google Earth Engine platform to interpolate the infrared band data to a 10-meter resolution, keeping the same resolution as the visible light band, thus obtaining the Sentinel image data with a 10-meter resolution; Step S2: pixel feature extraction: For each Sentinel-2 pixel’s six bands, we extract multispectral features of itself and its surroundings. Secondly, we use Vision Transformer, which is pre-trained based on previous remote sensing data, to extract high-level features of the same image. The specific steps include: Step S2-1, when extracting pixel features, use the sliding window method to focus on the extraction of spectral and texture details of the image; for each pixel, extract the 6-band spectral information of itself and the surrounding cross area, a total of 9 pixels; then connect the obtained spectral information to form a long sequence of 9 pixels × 6 bands with a length of 54; Step S2-2, using a pre-trained deep learning model for feature extraction, using the visual self-attention mechanism model encoder provided by Vision Transformer to extract high-level semantic features in remote sensing images; the encoder captures a wide range of image contexts and long-range dependencies in the image; each image is cropped into multiple slices with overlapping parts, and then each slice is processed into a feature map with the same spatial size based on the visual self-attention mechanism model encoder; finally, according to the above cropping mode, the feature maps of these slices are spliced ​​back to the original image size; Step S3: feature fusion: A feature fusion method is used to balance the local, low-level features from the pixel feature extraction branch with the global, high-level features extracted from the deep learning model to obtain the best clustering results. The specific steps include: merging the spectral data from the pixel feature extraction branch and the data from the deep learning model branch to form a feature vector with a length of 64, and normalizing this feature matrix to construct input features that can be directly used for the clustering algorithm; Step S4: Cluster calculation and 10-meter land cover data image production: After completing feature extraction, cluster analysis is performed based on the clustering algorithm to generate 300 clustering results. The relationship between the 30-meter resolution product, GLC_FCS30-2020, and the clustering results is analyzed to find the corresponding land cover type for each cluster. Furthermore, based on the data predicted as green space in the 10-meter resolution ESAWorldCover data, the previously generated 10-meter resolution land cover classification results are rechecked. If the classification results of some areas are not marked as green space in the generated 10-meter data, the spectral characteristics of the area and its surroundings are re-judged and reallocated to the category of green space. Automatic clustering and merging are completed to generate 10-meter high-resolution land cover data WS-GLM1030-PSET, which is stored in the labeled image file format TIFF. Step S5: Integration of 10-meter land cover data to 500-meter resolution The land cover data in TIFF format has a resolution of 10 meters. It is further integrated into multiple resolutions for model application as needed. The data is integrated into 500-meter grids. The integration method specifically includes the following steps: for each 500-meter grid, the 10-meter resolution grids contained in this grid are identified, and the percentages are recorded according to the land cover types to which these grids belong; Step S6: Matching of land cover data to vegetation functional types: The 500-meter land cover data is matched to the vegetation functional types required by the numerical model, which specifically includes the following steps: first, the 500-meter land cover data is read and matched to the 16 vegetation functional types used in the numerical simulation; Step S7: Comparison of vegetation functional types: Compare the high-resolution vegetation functional type data completed in step S6 with the traditional MODIS and vegetation functional type data produced by GLC_FCS30-2020 and ESAWorldCover to evaluate the degree of improvement of vegetation, especially urban green space, by the newly produced high-resolution and refined land cover type data; Step S8: Interpolation of 500-meter vegetation functional types to numerical model resolution After completing the 500-meter vegetation functional type processing, these data were further interpolated into a 25-km grid of spectral elements, and the newly generated PFT data were converted into a NetCDF format compatible with the original input file using the NCL tool; Step S9: Create high-resolution earth system model SW-HRESM input file: Write a program based on Python and climate data manipulation software CDO to extract PFT data from the newly generated NetCDF file and replace the corresponding variables in the original SW-HRESM input file, ensuring that the variable name PCT_PFT, type double, dimension and unit % are consistent with the original file; Step S10: Preparation of high-resolution biogenic emission inventory: Using the high-resolution Earth system model SW-HRESM and its land module's biogenic gas and aerosol emission estimation model MEGAN, the newly generated input file in step S9 is input into the MEGAN model for a test run, and then SW-HRESM is run to generate a global high-resolution biogenic emission inventory with improved accuracy; Step S11, air pollution simulation and subsequent analysis: Based on the newly generated biogenic emission inventory, high-resolution Earth system model simulations are carried out, and then programs are written to assess the impact of global biogenic emissions, especially emissions from urban green spaces, on atmospheric pollutants.

2. A method for improving urban green space biogenic emissions based on a high-resolution global model according to claim 1, characterized in that In step S1, 6 spectral images are selected from the 13 spectral images of Sentinel-2, and the total data of the 6 spectral images reaches 87TB; the visible light is red, green and blue, with a resolution of 10 meters; the resolution of the three infrared is 10 meters-20 meters.

3. A method for improving urban green space biogenic emissions based on a high-resolution global model according to claim 1, characterized in that ,The data processing software in step S9 also includes NCO.

Images