GBDT model-based reduction method for aerosol pollution caused by high-score urban green land form

Abstract

The invention relates to a GBDT model-based method for reducing aerosol pollution by a high-resolution city green space form, and belongs to the technical field of aerosol reduction. The method comprises the following steps: S1, obtaining an aerosol observation network database and apparent reflectivity data; s2, adjusting the precision of the aerosol data to a preset value by using a machine learning method; s3, re-classifying the data in the ArcGIS, distinguishing pollution intensity areas, and outputting second data based on urban planning management unit grids; s4, acquiring a land utilization image map, processing the land utilization image map and acquiring green land information; s5, performing batch processing on the extracted green space information by using Fragstats software; s7, processing the morphological indexes, and performing green land morphological core index construction and rule model and threshold determination by applying a GBDT model; and S8, proposing a regulation and control strategy based on the green land form core index, the rule model and the threshold value. The method plays an important role in reducing aerosol pollution and improving urban air quality.

Description

Technical Field

[0001] The invention relates to the technical field of aerosol reduction, and in particular to a method for reducing aerosol pollution based on a high-resolution urban green space morphology of a GBDT model. Background Art

[0002] With the development of remote sensing technology and the invention of landscape pattern analysis software Fragstats, people can interpret green spaces such as urban parks, wetland lakes and forest parks through remote sensing technology, and obtain green landscape indexes and conduct correlation studies with single-type aerosols such as PM2.5 and PM10. However, there is currently no or a lack of regulatory strategies for reducing aerosols based on the construction of core indicators of green space morphology, regularity models and green space morphology thresholds. It is easy to cause high aerosol concentrations due to insufficient planning of green spaces, which affects people's quality of life. Summary of the invention

[0003] Based on this, it is necessary to provide a method for reducing aerosol pollution based on high-scoring urban green space morphology based on the GBDT model, in response to the current lack of or lack of regulatory strategies for reducing aerosols based on the construction of core indicators of green space morphology, regularity models and green space morphology thresholds.

[0004] A method for reducing aerosol pollution by high-resolution urban green space morphology based on the GBDT model includes the following steps:

[0005] S1. Obtain the aerosol observation network database and apparent reflectance data, and analyze and obtain the aerosol optical depth;

[0006] S2. After adjusting the accuracy of the aerosol data to a preset value using a machine learning method, the aerosol data is segmented using ARCGIS software to obtain a vector diagram of aerosol spatial distribution, and the first data is output;

[0007] S3. reclassify it in ArcGIS and distinguish the pollution intensity areas based on the urban planning management unit grid, and output the second data;

[0008] S4. Obtain a land use image map, process the land use image map and obtain green space information, process the green space information, segment the fishing net based on the urban planning management unit, and create a serial number for each fishing net in the management unit;

[0009] S5. batch-process the extracted green space information using Fragstats software, divide it into multiple morphological indicators, and obtain attribute data and vector data of the morphological indicators;

[0010] S6. Select potential morphological factors that have an impact on reducing aerosols, analyze them, and obtain green space morphological indicators;

[0011] S7. Process the morphological indicators and use the GBDT model to construct the core indicators of green space morphology, determine the regularity model and threshold;

[0012] S8. Propose control strategies based on core indicators, regularity models and thresholds of green space morphology.

[0013] The present application discloses a method for reducing aerosol pollution by high-resolution urban green space morphology based on the GBDT model. By using the aerosol observation network database and apparent reflectance data, the aerosol optical thickness can be accurately analyzed and obtained. This process ensures the accuracy and reliability of the data. The precision of the aerosol data is further adjusted by machine learning methods to ensure the scientificity and accuracy of the model analysis results. Using ARCGIS software to segment and vectorize the data can generate a clear aerosol spatial distribution map, which is of great significance for the subsequent distinction of pollution intensity. By reclassifying the aerosol spatial data in ArcGIS, it is possible to accurately distinguish areas with different pollution intensities, including severely polluted areas, heavily polluted areas, etc. This refined regional division helps urban managers better understand the pollution conditions in different regions and facilitates the adoption of targeted measures. Combining these data with the urban planning management unit grid can more effectively carry out urban environmental governance planning. The processing of land use image maps and the extraction of green space information help to obtain detailed distribution and morphological data of urban green spaces. Using Fragstats software to batch process green space morphology can divide green space into multiple morphological indicators and obtain corresponding attribute data and vector data. This step provides a solid data basis for the study of the relationship between green space morphology and aerosol reduction effect. By analyzing potential morphological factors, green space morphological indicators that have a significant impact on aerosol reduction are selected to lay the foundation for subsequent model construction. The GBDT model is used to conduct in-depth analysis of these morphological indicators, construct core indicators and regularity models of green space morphology, and determine the optimal threshold for aerosol reduction. This data and model-based analysis method can provide scientific and accurate control strategies to ensure the efficiency of urban green space planning in reducing aerosol pollution. Ultimately, the control strategies proposed based on core indicators, regularity models and thresholds will provide urban managers with effective green space planning and management suggestions, thereby playing an important role in reducing aerosol pollution and improving urban air quality.

[0014] In one embodiment, the step S1 comprises the following steps:

[0015] S11. Obtain the AERONET global ground-based aerosol observation network database for inversion verification of satellite remote sensing aerosols;

[0016] S12. Obtain OLI data from NASA’s official website;

[0017] S13. Extract Landsat8 apparent reflectance data using Google Earth Engine;

[0018] S14. Select the AERONET station measurement values ​​within the preset time before and after the satellite passes to match with the satellite data. If there are multiple sets of station measurement values ​​within the interval, linear interpolation is used to obtain the aerosol AOD at the current moment. The expression of the aerosol AOD at the current moment is as follows:

[0019]

[0020] Among them, τ is the AOD of aerosol at the time of imaging; ta and tb are the two time points closest to the imaging time; τa and τb are the AOD at the corresponding time; t is the imaging time of Landsat8 satellite;

[0021] S15. Estimation of aerosol optical depth using deep belief neural network.

[0022] By obtaining data from the AERONET global ground-based aerosol observation network database, the aerosol inversion results of satellite remote sensing can be verified. This process ensures the accuracy of remote sensing data and can effectively reduce the uncertainty caused by satellite data errors. Obtaining OLI data from NASA's official website and extracting Landsat8's apparent reflectance data through Google Earth Engine can provide multi-source remote sensing data for aerosol optical thickness analysis. The acquisition and integration of such multi-source data helps to establish a more comprehensive and accurate environmental model. By selecting AERONET station measurements within a preset time before and after the satellite transit to match with satellite data, the temporal consistency of aerosol optical thickness (AOD) estimation can be ensured. If multiple sets of station measurements appear, the linear interpolation method can be used to more accurately estimate the AOD at the current moment within the time interval. This precise matching and interpolation method improves the temporal and spatial resolution of AOD data and provides a reliable basis for subsequent analysis. Using deep belief neural network (DBN) to estimate aerosol optical thickness (AOD) can further improve the accuracy and robustness of the estimation through deep learning technology. Deep belief neural networks can capture complex nonlinear relationships and adapt to the diversity and complexity of data, thereby providing more accurate and stable AOD estimation results. Through these additional steps, the method further optimizes the estimation process of aerosol optical thickness in all aspects of data acquisition and processing. This not only improves the accuracy of the entire model, but also provides more scientific and more reliable data support for the formulation of aerosol pollution reduction strategies. At the same time, the use of advanced technical means such as deep learning also provides broader application prospects for future environmental monitoring and management. Among them, AERONET (Aerosol Robotic Network) is a ground-based aerosol remote sensing observation network jointly established by NASA and LOA-PHOTONS (CNRS). OLI data is Landsat 8 Operational Land Image. Landsat 8 satellite contains two sensors: OLI (Operational Land Imager) and TIRS (Thermal Infrared Sensor). Google Earth Engine is a cloud computing platform dedicated to processing satellite images and other earth observation data.

[0023] In one embodiment, the step S15 comprises the following steps:

[0024] S151. Input the Landsat8 apparent reflectance data into the DBN model, use the RBM model to perform unsupervised pre-training on the collected samples layer by layer, extract the necessary features through unsupervised pre-training and pass them to the next layer of RBM model;

[0025] S152. Generate initial weights using the result of step S151 and calculate AOD; then calculate the root mean square error between the AOD estimated by the model and the actual site data of AERONET, return the error to the BP layer and perform repeated fine-tuning until the DBN model achieves satisfactory performance;

[0026] S153. The DBN model was cross-validated before use. When the accuracy was met, it was used to estimate the AOD value in areas without ground monitoring stations, thereby further obtaining the aerosol optical thickness distribution of the Landsat8 satellite.

[0027] S154. A ten-fold cross-validation technique was used to test the fitting and estimation capabilities of the DBN model, and a sample-based cross-validation was used to create an independent validation dataset.

[0028] The Landsat8 apparent reflectance data is input into the DBN model, and the restricted Boltzmann machine (RBM) model is used for layer-by-layer unsupervised pre-training, which can effectively extract the key features in the data. This layer-by-layer feature extraction method can deeply explore the intrinsic structure of the data, improve the expressiveness of the model when processing complex remote sensing data, and lay a solid foundation for the subsequent AOD estimation. After generating the initial weights and calculating the AOD based on step S151, the error of the model can be quantified by calculating the root mean square error (RMSE) between the estimated AOD and the actual AERONET data. This error is returned to the back propagation (BP) layer and fine-tuned repeatedly to continuously optimize the performance of the model. This process ensures that the DBN model can achieve the best accuracy and reliability in AOD estimation, reduce errors, and improve the accuracy of the final results. After the DBN model reaches satisfactory accuracy, it is used to estimate AOD in areas without ground monitoring stations, further expanding the coverage of aerosol optical thickness distribution. This step is particularly helpful in filling data gaps and providing scientific and reliable AOD estimation results for areas lacking observation data, thereby more comprehensively depicting the aerosol pollution status of the entire city. The DBN model is tested by using the ten-fold cross-validation technique, which can effectively evaluate the model's fitting and estimation capabilities. By creating an independent validation data set, the model is ensured to have efficient and stable performance in practical applications. The ten-fold cross-validation can avoid overfitting of the model and ensure its generalization ability on new data, making the AOD estimation results more robust and reliable. Through these steps, the entire DBN model achieves high accuracy and reliability in the process of estimating aerosol optical thickness. These technical measures not only improve the adaptability of the model in complex environments, but also provide more comprehensive and accurate aerosol pollution data support for urban managers, helping to formulate more effective environmental monitoring and governance strategies. Among them, the RBM model is a restricted Boltzmann machine (RBM), which is a randomly generated neural network that can learn probability distributions through input data sets. The DBN model is a deep belief network (DeepBelief Networks), a generative graph model composed of multiple layers of restricted Boltzmann machines (RBM) stacked together.

[0029] In one of the embodiments, the spatial resolution of the OLI data is 30m; the Landsat8 apparent reflectance data includes: TOA reflectance data, radiation spectrum, observation geometry data, NDVI data and quality control band data; and the preset time is 30min.

[0030] In one embodiment, the TOA reflectivity data is TOA reflectivity data in the B1 to B7 bands, the wavelength range of the radiation spectrum is from the wavelength corresponding to visible light to the wavelength corresponding to near infrared, and the observation geometry data is the solar altitude angle and relative azimuth angle.

[0031] In one embodiment, the step S6 comprises the following steps:

[0032] S61. Select potential morphological factors that affect aerosol reduction, connect the AOD data of aerosols and the morphological indicators of green spaces to the planning management unit grid of ArcGIS, obtain one-to-one corresponding attribute data, import them into ArcGIS for visualization, observe their spatial distribution, and then export the attribute table;

[0033] S62. Use the Spearman bivariate correlation analysis of SPSS software to preliminarily screen the green space morphological data that affect aerosol pollution, retain the green space morphological indicators with large and significant correlation coefficients with aerosols, and eliminate the insignificant morphological indicators;

[0034] S63. Perform pairwise Spearman correlation analysis on the green space morphology indicators preliminarily screened above. If the correlation coefficients between the selected green space morphology indicators are small or insignificant, the obviously significant indicators are eliminated and the green space morphology indicators are screened out for the second time.

[0035] By connecting the AOD data of aerosols and green space morphology indicators to the planning and management unit grid of ArcGIS, effective integration of spatial data can be achieved. After visualizing the attribute data, the spatial distribution relationship between green space morphology and aerosol pollution in different regions can be intuitively observed. This refined data processing method provides reliable spatial data support for further analysis and helps researchers identify potential influencing factors through visualization. Using the Spearman bivariate correlation analysis of SPSS software, green space morphology indicators that are highly correlated with aerosol pollution can be preliminarily screened out. By retaining indicators with large and significant correlation coefficients with aerosols and eliminating insignificant morphological indicators, the scope of the study is effectively narrowed, allowing subsequent analysis to focus more on key variables. This process improves the efficiency of the analysis and ensures that the model only contains green space morphology factors that have a significant impact on pollution reduction. Further comparisons of the preliminarily screened green space morphology indicators through Spearman correlation analysis can effectively exclude those indicators that are significantly correlated with each other. After eliminating the significantly correlated indicators, the independence of the green space morphology indicators finally screened out is ensured. This step greatly reduces the impact of multicollinearity on the model, improves the accuracy and reliability of the analysis results, and thus enhances the predictive ability of the model in practical applications. Through these steps, the method can accurately identify the green space morphological factors that have the greatest impact on aerosol pollution reduction. This not only improves the prediction accuracy of the model, but also provides a scientific decision-making basis for urban green space planning and environmental governance. The refined data processing and screening process ensures the simplicity and effectiveness of the final model, which helps to better manage and optimize green space morphology in complex urban environments, thereby achieving more effective aerosol pollution control.

[0036] In one embodiment, the step S4 comprises the following steps:

[0037] S41. Download 0.5m pixel high-resolution land use imagery from Google Earth in slices;

[0038] S42. Obtaining Google Earth digital map data and ground truth sample data for verifying the extracted green space information;

[0039] S43. Use ENVI software to preprocess the original remote sensing image data, including atmosphere removal, radiation calibration, and geometric correction;

[0040] S44. Selecting spectral information, texture information, and morphological information features to establish a green space extraction model;

[0041] S45. Training the green space extraction model using ground truth sample data;

[0042] S46. Insert the results of ENVI area of ​​interest, ENVI feature markers, and ENVI traditional classification method into ENVIDeepLearning Module, and insert the green space marker data obtained from Open Street Maps and ArcGIS Pro platform into ENVIDeep Learning Module to obtain training samples;

[0043] S47. When the above training samples cannot meet the requirements, a new green space extraction model is created, or deep processing is performed on the existing green space extraction model;

[0044] S48. Extracting the required green space vector from the image through the green space extraction model;

[0045] S49. Post-processing the extracted green space vector results, including vector result checking and editing, vector data splicing and clipping, and attribute assignment;

[0046] S410. The 0.5m high-precision green space vector data obtained above is segmented into fishing nets based on urban planning management units, and a new Arabic numeral serial number of each management unit fishing net is created in the attribute table.

[0047] Downloading high-resolution land use images with a pixel size of 0.5 m from Google Earth ensures that the green space information is rich in details and accurate. High-resolution data provides a solid foundation for the subsequent extraction and analysis of green space information. By obtaining Google Earth digital map data and ground truth sample data for verification, it can be ensured that the extracted green space information is highly consistent with the actual situation. This verification process improves the reliability of the data and makes the green space extraction results more credible. The original remote sensing image data is preprocessed by ENVI software, such as atmospheric removal, radiation calibration and geometric correction, to ensure the quality and consistency of the image data. These processing steps eliminate errors and noise in the image and enhance the accuracy of subsequent analysis. Selecting spectral information, texture information and morphological information features to establish a green space extraction model can comprehensively consider multi-dimensional information, making the model more robust and adaptable. This lays the foundation for the accurate extraction of green space information. The green space extraction model is trained using ground truth sample data so that the model can better reflect the actual green space distribution. In the case of insufficient training samples, by creating a new green space extraction model or deep processing the existing model, it can flexibly respond to the needs of different data sets and ensure the versatility and efficiency of the model. Through the above green space extraction model, the green space vector information is accurately extracted from the image. This step ensures the high accuracy of the green space information and provides detailed basic data for subsequent urban planning and environmental analysis. The extracted green space vector results are post-processed, including inspection, editing, splicing and clipping, to ensure the integrity and consistency of the data. By segmenting these high-precision green space vector data based on urban planning management units and creating a serial number for each management unit, the efficiency of data organization and management is further improved, providing an accurate tool for urban green space management. These steps ensure the accuracy and reliability of the extraction and processing process from remote sensing data to the final green space vector data. The high-precision green space data generated in the end not only helps to accurately analyze the effect of green space morphology on the reduction of aerosol pollution, but also provides detailed spatial information support for urban planners, thereby promoting more scientific and effective urban green space management and environmental governance. Google Earth is a virtual earth software developed by Google, which arranges satellite photos, aerial photography and GIS on a three-dimensional model of the earth.

[0048] In one embodiment, the step S7 comprises the following steps:

[0049] S71. The green space morphological indicators were randomly divided into training set and test set in a ratio of 8:2, and the GBM package of R language was used to fit the model;

[0050] S72. Use the GBDT model to construct core indicators of green space morphology, determine regularity models and thresholds.

[0051] The green space morphology index is randomly divided into a training set and a test set at a ratio of 8:2. This segmentation method ensures that the model can fully learn the data features during the training process, while retaining a part of the data for model verification. This method can effectively evaluate the generalization ability of the model and ensure that the model can still show good prediction results on unseen data. Using the GBM package of R language for model fitting, the relationship model between green space morphology index and aerosol pollution reduction effect can be constructed through the powerful function of gradient boosting machine (GBM). The GBM model has advantages in processing complex data structures and capturing nonlinear relationships, making the model fitting process more accurate and efficient. The use of the GBDT model to construct the core indicators of green space morphology can identify the key factors that have the greatest impact on aerosol pollution reduction. The GBDT model can capture complex variable relationships through the combination of multiple decision trees, thereby more accurately reflecting the aerosol reduction effect of green space morphology. Determining the regularity model and threshold can provide clear guidance for actual urban green space management. For example, after finding the thresholds of certain key green space morphology indicators, the green space planning and layout can be adjusted according to these thresholds to maximize its aerosol pollution reduction effect. This data-based decision support helps to formulate more scientific and accurate environmental governance strategies. Through these steps, the entire model has achieved higher accuracy and operability in the analysis of the effect of green space morphology on aerosol pollution reduction. The reasonable segmentation of data, the application of the GBDT model, and the identification of key indicators and thresholds make this method not only have strong predictive capabilities, but also provide feasible operational guidelines for actual urban planning and environmental governance, helping cities to use green space more effectively to improve air quality.

[0052] The parameters of the GBDT model and their adjustment methods are as follows:

[0053] Set the learning rate: The learning rate determines the contribution of each tree to the final prediction. A smaller learning rate usually means more trees are needed to achieve the same performance, but it can reduce the risk of overfitting. Therefore, it is necessary to control the weight of each step of the fitting process to improve the accuracy of the model;

[0054] Number of trees (N_estimators): The number of trees determines the total number of weak learners in the model. Increasing the number of trees usually improves the performance of the model, but may also lead to overfitting;

[0055] Max Depth: The depth of the tree limits the maximum number of layers in the tree. Deeper trees can capture more complex patterns, but are also more prone to overfitting.

[0056] Min Samples Split: This parameter determines the minimum number of samples required to split an internal node. Increasing this value can reduce the complexity of the tree and reduce the risk of overfitting;

[0057] Min Samples Leaf: This parameter determines the minimum number of samples required for a leaf node. Increasing this value can prevent the model from overfitting in areas with less data;

[0058] Subsample: This parameter determines the number of features randomly selected when building each tree. Using feature subsets can increase the generalization ability of the model;

[0059] Loss Function: Depending on the regression task, it is crucial to choose an appropriate loss function;

[0060] Regularization parameters (such as L1 and L2 regularization): Regularization can help control the complexity of the model and reduce overfitting;

[0061] Early Stopping: During the cross-validation process, if the model does not improve in multiple consecutive iterations, stop training to avoid overfitting;

[0062] Feature importance evaluation: By evaluating the importance of features, we can understand which features contribute most to the model's prediction, which helps with feature selection and model interpretation;

[0063] Parallelization settings: Depending on the available computing resources, adjust the parallelization settings of the model to speed up training.

[0064] In one embodiment, the step S72 comprises the following steps:

[0065] S721. Based on the GBDT model, we obtain the core indicators of green space morphology for reducing aerosols, reveal the regularity model and threshold effect, and use it as a morphology evaluation tool to form a model tool for green space morphology to reduce aerosol pollution;

[0066] S722. Based on the threshold of green space morphology obtained by GBDT model and combined with geographically weighted regression (GWR) method, a threshold-based regulation strategy is proposed;

[0067] S723. Combined with the permanent population data, a distribution map of polluted residential areas is generated. Based on the correlation between the core morphological indicators and aerosol concentrations of different pollution zones, core control indicators, key control indicators, and general control indicators are proposed for each type of management unit. Combined with the planning, a hierarchical control strategy for the spatial morphology of different types of pollution units is proposed;

[0068] S724. Four least squares models were constructed with the core spatial morphology index as the explanatory variable and the aerosol of each management unit as the dependent variable. The residuals of each standard unit were counted. The spatial distribution of the residuals in the simulation results was used to find potential control variables that should be included in the model, so as to further control and optimize the spatial morphology of special functional areas and reduce the impact of aerosol pollution on residents.

[0069] S725. Based on the management unit, several administrative areas with extremely low green coverage and extremely high green coverage were selected, and the differences in the regulation strategies before and after optimization of these two types of sample sites were compared to further optimize the core indicators of green space morphology.

[0070] Based on the core indicators of green space morphology obtained by the GBDT model, and by revealing the regular model and threshold effect, a special morphology evaluation tool was created. This tool can not only accurately evaluate the effect of green space morphology on aerosol pollution reduction, but also provide a scientific basis for urban planners to help them better design and manage green space layout, thereby effectively reducing aerosol pollution. Combined with the geographically weighted regression (GWR) method, the green space morphology thresholds obtained based on the GBDT model were further analyzed, and control strategies based on these thresholds were proposed. The GWR method can take into account spatial heterogeneity, making the control strategy more locally adaptable. This threshold-based control strategy can guide urban managers to implement targeted green space planning in different regions to maximize its environmental benefits. By combining the distribution map of polluted residential areas generated by permanent population data, the correlation between the core morphology indicators and aerosol concentrations in different pollution zones was analyzed. Based on these analysis results, core control indicators, key control indicators and general control indicators were proposed for each type of management unit. Combined with urban planning, these hierarchical control strategies can effectively reduce the impact of aerosol pollution on residential areas and improve the quality of life of residents. By constructing a least squares model and analyzing the spatial distribution of the model residuals, potential control variables that should be included in the model are identified. This residual analysis method helps to discover the deficiencies in the existing model and optimize the spatial morphology design of special functional areas by adjusting the model to include more control variables, further reducing the impact of aerosol pollution on residents. By comparing the differences between administrative districts with extremely low and extremely high green coverage before and after the implementation of the regulation strategy, the actual effect of the regulation strategy can be evaluated. This comparison not only verifies the effectiveness of the regulation strategy, but also provides a practical basis for further optimizing the core indicators of green space morphology. Ultimately, by continuously optimizing the core indicators, the role of urban green space in reducing aerosol pollution can be better realized. These steps further improve the efficiency and effectiveness of urban green space morphology in reducing aerosol pollution through scientific analysis and strategy formulation. Refined regulation strategies enable urban management to deal with pollution problems in different regions in a more targeted manner, thereby providing residents with a healthier living environment.

[0071] In one of the embodiments, the preset value is 30m; the specific steps of outputting the first data are: statistical attribute table data, exporting to EXCEL, and saving as CSV format data; the specific steps of outputting the second data are: extracting the reclassified new values ​​into attribute fields of different grids, and exporting to EXCEL tables, and storing in CSV format; the multiple types of indicators are area, density, boundary length, shape, agglomeration, connectivity, and core area indicators; the pollution intensity areas include severely polluted areas, heavily polluted areas, moderately polluted areas, lightly polluted areas, low polluted areas, and extremely low polluted areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a flowchart of the method of reducing aerosol pollution by high-resolution urban green space morphology based on the GBDT model;

[0073] Figure 2 is a specific flow chart of step S1;

[0074] Figure 3 is a specific flow chart of step S15;

[0075] Figure 4 is a specific flow chart of step S6;

[0076] Figure 5 is a specific flow chart of step S4;

[0077] Figure 6 is a specific flow chart of step S7;

[0078] Figure 7 Schematic diagram of the specific process of step S72. DETAILED DESCRIPTION

[0079] 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.

[0080] 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.

[0081] like Figure 1 As shown, this embodiment discloses a method for reducing aerosol pollution based on high-resolution urban green space morphology of the GBDT model, comprising the following steps:

[0082] S1. Obtain the aerosol observation network database and apparent reflectance data, and analyze and obtain the aerosol optical depth;

[0083] S2. After adjusting the accuracy of the aerosol data to a preset value using a machine learning method, the aerosol data is segmented using ARCGIS software to obtain a vector diagram of aerosol spatial distribution, and the first data is output;

[0084] S3. reclassify it in ArcGIS and distinguish the pollution intensity areas based on the urban planning management unit grid, and output the second data;

[0085] S4. Obtain a land use image map, process the land use image map and obtain green space information, process the green space information, segment the fishing net based on the urban planning management unit, and create a serial number for each fishing net in the management unit;

[0086] S5. batch-process the extracted green space information using Fragstats software, divide it into multiple morphological indicators, and obtain attribute data and vector data of the morphological indicators;

[0087] S6. Select potential morphological factors that have an impact on reducing aerosols, analyze them, and obtain green space morphological indicators;

[0088] S7. Process the morphological indicators and use the GBDT model to construct the core indicators of green space morphology, determine the regularity model and threshold;

[0089] S8. Propose control strategies based on core indicators, regularity models and thresholds of green space morphology.

[0090] The present application discloses a method for reducing aerosol pollution by high-resolution urban green space morphology based on the GBDT model. By using the aerosol observation network database and apparent reflectance data, the aerosol optical thickness can be accurately analyzed and obtained. This process ensures the accuracy and reliability of the data. The precision of the aerosol data is further adjusted by machine learning methods to ensure the scientificity and accuracy of the model analysis results. Using ARCGIS software to segment and vectorize the data can generate a clear aerosol spatial distribution map, which is of great significance for the subsequent distinction of pollution intensity. By reclassifying the aerosol spatial data in ArcGIS, it is possible to accurately distinguish areas with different pollution intensities, including severely polluted areas, heavily polluted areas, etc. This refined regional division helps urban managers better understand the pollution conditions in different regions and facilitates the adoption of targeted measures. Combining these data with the urban planning management unit grid can more effectively carry out urban environmental governance planning. The processing of land use image maps and the extraction of green space information help to obtain detailed distribution and morphological data of urban green spaces. Using Fragstats software to batch process green space morphology can divide green space into multiple morphological indicators and obtain corresponding attribute data and vector data. This step provides a solid data basis for the study of the relationship between green space morphology and aerosol reduction effect. By analyzing potential morphological factors, green space morphological indicators that have a significant impact on aerosol reduction are selected to lay the foundation for subsequent model construction. The GBDT model is used to conduct in-depth analysis of these morphological indicators, construct core indicators and regularity models of green space morphology, and determine the optimal threshold for aerosol reduction. This data and model-based analysis method can provide scientific and accurate control strategies to ensure the efficiency of urban green space planning in reducing aerosol pollution. Ultimately, the control strategies proposed based on core indicators, regularity models and thresholds will provide urban managers with effective green space planning and management suggestions, thereby playing an important role in reducing aerosol pollution and improving urban air quality.

[0091] like Figure 2 As shown, in addition to the features of the above embodiment, this embodiment further defines that: the step S1 includes the following steps:

[0092] S11. Obtain the AERONET global ground-based aerosol observation network database for inversion verification of satellite remote sensing aerosols;

[0093] S12. Obtain OLI data from NASA’s official website;

[0094] S13. Extract Landsat8 apparent reflectance data using Google Earth Engine;

[0095] S14. Select the AERONET station measurement values ​​within the preset time before and after the satellite passes to match with the satellite data. If there are multiple sets of station measurement values ​​within the interval, linear interpolation is used to obtain the aerosol AOD at the current moment. The expression of the aerosol AOD at the current moment is as follows:

[0096]

[0097] Among them, τ is the AOD of aerosol at the time of imaging; ta and tb are the two time points closest to the imaging time; τa and τb are the AOD at the corresponding time; t is the imaging time of Landsat8 satellite;

[0098] S15. Estimation of aerosol optical depth using deep belief neural network.

[0099] By obtaining data from the AERONET global ground-based aerosol observation network database, the aerosol inversion results of satellite remote sensing can be verified. This process ensures the accuracy of remote sensing data and can effectively reduce the uncertainty caused by satellite data errors. Obtaining OLI data from NASA's official website and extracting Landsat8's apparent reflectance data through Google Earth Engine can provide multi-source remote sensing data for aerosol optical thickness analysis. The acquisition and integration of such multi-source data helps to establish a more comprehensive and accurate environmental model. By selecting AERONET station measurements within a preset time before and after the satellite transit to match with satellite data, the temporal consistency of aerosol optical thickness (AOD) estimation can be ensured. If multiple sets of station measurements appear, the linear interpolation method can be used to more accurately estimate the AOD at the current moment within the time interval. This precise matching and interpolation method improves the temporal and spatial resolution of AOD data and provides a reliable basis for subsequent analysis. Using deep belief neural network (DBN) to estimate aerosol optical thickness (AOD) can further improve the accuracy and robustness of the estimation through deep learning technology. Deep belief neural networks can capture complex nonlinear relationships and adapt to the diversity and complexity of data, thereby providing more accurate and stable AOD estimation results. Through these additional steps, the method further optimizes the estimation process of aerosol optical thickness in all aspects of data acquisition and processing. This not only improves the accuracy of the entire model, but also provides more scientific and more reliable data support for the formulation of aerosol pollution reduction strategies. At the same time, the use of advanced technical means such as deep learning also provides broader application prospects for future environmental monitoring and management. Among them, AERONET (Aerosol Robotic Network) is a ground-based aerosol remote sensing observation network jointly established by NASA and LOA-PHOTONS (CNRS). OLI data is Landsat 8 Operational Land Image. Landsat 8 satellite contains two sensors: OLI (Operational Land Imager) and TIRS (Thermal Infrared Sensor). Google Earth Engine is a cloud computing platform dedicated to processing satellite images and other earth observation data.

[0100] like Figure 3 As shown, in addition to the features of the above embodiment, this embodiment further defines: the step S15 includes the following steps:

[0101] S151. Input the Landsat8 apparent reflectance data into the DBN model, use the RBM model to perform unsupervised pre-training on the collected samples layer by layer, extract the necessary features through unsupervised pre-training and pass them to the next layer of RBM model;

[0102] S152. Generate initial weights using the result of step S151 and calculate AOD; then calculate the root mean square error between the AOD estimated by the model and the actual site data of AERONET, return the error to the BP layer and perform repeated fine-tuning until the DBN model achieves satisfactory performance;

[0103] S153. The DBN model was cross-validated before use. When the accuracy was met, it was used to estimate the AOD value in areas without ground monitoring stations, thereby further obtaining the aerosol optical thickness distribution of the Landsat8 satellite.

[0104] S154. A ten-fold cross-validation technique was used to test the fitting and estimation capabilities of the DBN model, and a sample-based cross-validation was used to create an independent validation dataset.

[0105] The Landsat8 apparent reflectance data is input into the DBN model, and the restricted Boltzmann machine (RBM) model is used for layer-by-layer unsupervised pre-training, which can effectively extract the key features in the data. This layer-by-layer feature extraction method can deeply explore the intrinsic structure of the data, improve the expressiveness of the model when processing complex remote sensing data, and lay a solid foundation for the subsequent AOD estimation. After generating the initial weights and calculating the AOD based on step S151, the error of the model can be quantified by calculating the root mean square error (RMSE) between the estimated AOD and the actual AERONET data. This error is returned to the back propagation (BP) layer and fine-tuned repeatedly to continuously optimize the performance of the model. This process ensures that the DBN model can achieve the best accuracy and reliability in AOD estimation, reduce errors, and improve the accuracy of the final results. After the DBN model reaches satisfactory accuracy, it is used to estimate AOD in areas without ground monitoring stations, further expanding the coverage of aerosol optical thickness distribution. This step is particularly helpful in filling data gaps and providing scientific and reliable AOD estimation results for areas lacking observation data, thereby more comprehensively depicting the aerosol pollution status of the entire city. The DBN model is tested by using the ten-fold cross-validation technique, which can effectively evaluate the model's fitting and estimation capabilities. By creating an independent validation data set, the model is ensured to have efficient and stable performance in practical applications. The ten-fold cross-validation can avoid overfitting of the model and ensure its generalization ability on new data, making the AOD estimation results more robust and reliable. Through these steps, the entire DBN model achieves high accuracy and reliability in the process of estimating aerosol optical thickness. These technical measures not only improve the adaptability of the model in complex environments, but also provide more comprehensive and accurate aerosol pollution data support for urban managers, helping to formulate more effective environmental monitoring and governance strategies. Among them, the RBM model is a restricted Boltzmann machine (RBM), which is a randomly generated neural network that can learn probability distributions through input data sets. The DBN model is a deep belief network (DeepBelief Networks), a generative graph model composed of multiple layers of restricted Boltzmann machines (RBM) stacked together.

[0106] In addition to the features of the above embodiments, this embodiment further defines: the spatial resolution of the OLI data is 30m; the Landsat8 apparent reflectance data includes: TOA reflectance data, radiation spectrum, observation geometry data, NDVI data and quality control band data; the preset time is 30min.

[0107] In addition to the features of the above embodiments, this embodiment further defines: the TOA reflectivity data is the TOA reflectivity data of the B1 to B7 bands, the wavelength range of the radiation spectrum is the wavelength corresponding to visible light to the wavelength corresponding to near infrared, and the observation geometry data is the solar altitude angle and the relative azimuth angle.

[0108] like Figure 4 As shown, in addition to the features of the above embodiment, this embodiment further defines that: the step S6 includes the following steps:

[0109] S61. Select potential morphological factors that affect aerosol reduction, connect the AOD data of aerosols and the morphological indicators of green spaces to the planning management unit grid of ArcGIS, obtain one-to-one corresponding attribute data, import them into ArcGIS for visualization, observe their spatial distribution, and then export the attribute table;

[0110] S62. Use the Spearman bivariate correlation analysis of SPSS software to preliminarily screen the green space morphological data that affect aerosol pollution, retain the green space morphological indicators with large and significant correlation coefficients with aerosols, and eliminate the insignificant morphological indicators;

[0111] S63. Perform pairwise Spearman correlation analysis on the green space morphology indicators preliminarily screened above. If the correlation coefficients between the selected green space morphology indicators are small or insignificant, the obviously significant indicators are eliminated and the green space morphology indicators are screened out for the second time.

[0112] By connecting the AOD data of aerosols and green space morphology indicators to the planning and management unit grid of ArcGIS, effective integration of spatial data can be achieved. After visualizing the attribute data, the spatial distribution relationship between green space morphology and aerosol pollution in different regions can be intuitively observed. This refined data processing method provides reliable spatial data support for further analysis and helps researchers identify potential influencing factors through visualization. Using the Spearman bivariate correlation analysis of SPSS software, green space morphology indicators that are highly correlated with aerosol pollution can be preliminarily screened out. By retaining indicators with large and significant correlation coefficients with aerosols and eliminating insignificant morphological indicators, the scope of the study is effectively narrowed, allowing subsequent analysis to focus more on key variables. This process improves the efficiency of the analysis and ensures that the model only contains green space morphology factors that have a significant impact on pollution reduction. Further comparisons of the preliminarily screened green space morphology indicators through Spearman correlation analysis can effectively exclude those indicators that are significantly correlated with each other. After eliminating the significantly correlated indicators, the independence of the green space morphology indicators finally screened out is ensured. This step greatly reduces the impact of multicollinearity on the model, improves the accuracy and reliability of the analysis results, and thus enhances the predictive ability of the model in practical applications. Through these steps, the method can accurately identify the green space morphological factors that have the greatest impact on aerosol pollution reduction. This not only improves the prediction accuracy of the model, but also provides a scientific decision-making basis for urban green space planning and environmental governance. The refined data processing and screening process ensures the simplicity and effectiveness of the final model, which helps to better manage and optimize green space morphology in complex urban environments, thereby achieving more effective aerosol pollution control.

[0113] like Figure 5 As shown, in addition to the features of the above embodiment, this embodiment further defines that: the step S4 includes the following steps:

[0114] S41. Download 0.5m pixel high-resolution land use imagery from Google Earth in slices;

[0115] S42. Obtaining Google Earth digital map data and ground truth sample data for verifying the extracted green space information;

[0116] S43. Use ENVI software to preprocess the original remote sensing image data, including atmosphere removal, radiation calibration, and geometric correction;

[0117] S44. Selecting spectral information, texture information, and morphological information features to establish a green space extraction model;

[0118] S45. Training the green space extraction model using ground truth sample data;

[0119] S46. Insert the results of ENVI area of ​​interest, ENVI feature markers, and ENVI traditional classification method into ENVIDeepLearning Module, and insert the green space marker data obtained from Open Street Maps and ArcGIS Pro platform into ENVIDeep Learning Module to obtain training samples;

[0120] S47. When the above training samples cannot meet the requirements, a new green space extraction model is created, or deep processing is performed on the existing green space extraction model;

[0121] S48. Extracting the required green space vector from the image through the green space extraction model;

[0122] S49. Post-processing the extracted green space vector results, including vector result checking and editing, vector data splicing and clipping, and attribute assignment;

[0123] S410. The 0.5m high-precision green space vector data obtained above is segmented into fishing nets based on urban planning management units, and a new Arabic numeral serial number of each management unit fishing net is created in the attribute table.

[0124] Downloading high-resolution land use images with a pixel size of 0.5 m from Google Earth ensures that the green space information is rich in details and accurate. High-resolution data provides a solid foundation for the subsequent extraction and analysis of green space information. By obtaining Google Earth digital map data and ground truth sample data for verification, it can be ensured that the extracted green space information is highly consistent with the actual situation. This verification process improves the reliability of the data and makes the green space extraction results more credible. The original remote sensing image data is preprocessed by ENVI software, such as atmospheric removal, radiation calibration and geometric correction, to ensure the quality and consistency of the image data. These processing steps eliminate errors and noise in the image and enhance the accuracy of subsequent analysis. Selecting spectral information, texture information and morphological information features to establish a green space extraction model can comprehensively consider multi-dimensional information, making the model more robust and adaptable. This lays the foundation for the accurate extraction of green space information. The green space extraction model is trained using ground truth sample data so that the model can better reflect the actual green space distribution. In the case of insufficient training samples, by creating a new green space extraction model or deep processing the existing model, it can flexibly respond to the needs of different data sets and ensure the versatility and efficiency of the model. Through the above green space extraction model, the green space vector information is accurately extracted from the image. This step ensures the high accuracy of the green space information and provides detailed basic data for subsequent urban planning and environmental analysis. The extracted green space vector results are post-processed, including inspection, editing, splicing and clipping, to ensure the integrity and consistency of the data. By segmenting these high-precision green space vector data based on urban planning management units and creating a serial number for each management unit, the efficiency of data organization and management is further improved, providing an accurate tool for urban green space management. These steps ensure the accuracy and reliability of the extraction and processing process from remote sensing data to the final green space vector data. The high-precision green space data generated in the end not only helps to accurately analyze the effect of green space morphology on the reduction of aerosol pollution, but also provides detailed spatial information support for urban planners, thereby promoting more scientific and effective urban green space management and environmental governance. Google Earth is a virtual earth software developed by Google, which arranges satellite photos, aerial photography and GIS on a three-dimensional model of the earth.

[0125] like Figure 6 As shown, in addition to the features of the above embodiment, this embodiment further defines that: the step S7 includes the following steps:

[0126] S71. The green space morphological indicators were randomly divided into training set and test set in a ratio of 8:2, and the GBM package of R language was used to fit the model;

[0127] S72. Use the GBDT model to construct core indicators of green space morphology, determine regularity models and thresholds.

[0128] The green space morphology index is randomly divided into a training set and a test set at a ratio of 8:2. This segmentation method ensures that the model can fully learn the data features during the training process, while retaining a part of the data for model verification. This method can effectively evaluate the generalization ability of the model and ensure that the model can still show good prediction results on unseen data. Using the GBM package of R language for model fitting, the relationship model between green space morphology index and aerosol pollution reduction effect can be constructed through the powerful function of gradient boosting machine (GBM). The GBM model has advantages in processing complex data structures and capturing nonlinear relationships, making the model fitting process more accurate and efficient. The use of the GBDT model to construct the core indicators of green space morphology can identify the key factors that have the greatest impact on aerosol pollution reduction. The GBDT model can capture complex variable relationships through the combination of multiple decision trees, thereby more accurately reflecting the aerosol reduction effect of green space morphology. Determining the regularity model and threshold can provide clear guidance for actual urban green space management. For example, after finding the thresholds of certain key green space morphology indicators, the green space planning and layout can be adjusted according to these thresholds to maximize its aerosol pollution reduction effect. This data-based decision support helps to formulate more scientific and accurate environmental governance strategies. Through these steps, the entire model has achieved higher accuracy and operability in the analysis of the effect of green space morphology on aerosol pollution reduction. The reasonable segmentation of data, the application of the GBDT model, and the identification of key indicators and thresholds make this method not only have strong predictive capabilities, but also provide feasible operational guidelines for actual urban planning and environmental governance, helping cities to use green space more effectively to improve air quality.

[0129] The parameters of the GBDT model and their adjustment methods are as follows:

[0130] Set the learning rate: The learning rate determines the contribution of each tree to the final prediction. A smaller learning rate usually means more trees are needed to achieve the same performance, but it can reduce the risk of overfitting. Therefore, it is necessary to control the weight of each step of the fitting process to improve the accuracy of the model;

[0131] Number of trees (N_estimators): The number of trees determines the total number of weak learners in the model. Increasing the number of trees usually improves the performance of the model, but may also lead to overfitting;

[0132] Max Depth: The depth of the tree limits the maximum number of layers in the tree. Deeper trees can capture more complex patterns, but are also more prone to overfitting.

[0133] Min Samples Split: This parameter determines the minimum number of samples required to split an internal node. Increasing this value can reduce the complexity of the tree and reduce the risk of overfitting;

[0134] Min Samples Leaf: This parameter determines the minimum number of samples required for a leaf node. Increasing this value can prevent the model from overfitting in areas with less data;

[0135] Subsample: This parameter determines the number of features randomly selected when building each tree. Using feature subsets can increase the generalization ability of the model;

[0136] Loss Function: Depending on the regression task, it is crucial to choose an appropriate loss function;

[0137] Regularization parameters (such as L1 and L2 regularization): Regularization can help control the complexity of the model and reduce overfitting;

[0138] Early Stopping: During the cross-validation process, if the model does not improve in multiple consecutive iterations, stop training to avoid overfitting;

[0139] Feature importance evaluation: By evaluating the importance of features, we can understand which features contribute most to the model's prediction, which helps with feature selection and model interpretation;

[0140] Parallelization settings: Depending on the available computing resources, adjust the parallelization settings of the model to speed up training.

[0141] like Figure 7 As shown, in addition to the features of the above embodiment, this embodiment further defines that: the step S72 includes the following steps:

[0142] S721. Based on the GBDT model, we obtain the core indicators of green space morphology for reducing aerosols, reveal the regularity model and threshold effect, and use it as a morphology evaluation tool to form a model tool for green space morphology to reduce aerosol pollution;

[0143] S722. Based on the threshold of green space morphology obtained by GBDT model and combined with geographically weighted regression (GWR) method, a threshold-based regulation strategy is proposed;

[0144] S723. Combined with the permanent population data, a distribution map of polluted residential areas is generated. Based on the correlation between the core morphological indicators and aerosol concentrations of different pollution zones, core control indicators, key control indicators, and general control indicators are proposed for each type of management unit. Combined with the planning, a hierarchical control strategy for the spatial morphology of different types of pollution units is proposed;

[0145] S724. Four least squares models were constructed with the core spatial morphology index as the explanatory variable and the aerosol of each management unit as the dependent variable. The residuals of each standard unit were counted. The spatial distribution of the residuals in the simulation results was used to find potential control variables that should be included in the model, so as to further control and optimize the spatial morphology of special functional areas and reduce the impact of aerosol pollution on residents.

[0146] S725. Based on the management unit, several administrative areas with extremely low green coverage and extremely high green coverage were selected, and the differences in the regulation strategies before and after optimization of these two types of sample sites were compared to further optimize the core indicators of green space morphology.

[0147] Based on the core indicators of green space morphology obtained by the GBDT model, and by revealing the regular model and threshold effect, a special morphology evaluation tool was created. This tool can not only accurately evaluate the effect of green space morphology on aerosol pollution reduction, but also provide a scientific basis for urban planners to help them better design and manage green space layout, thereby effectively reducing aerosol pollution. Combined with the geographically weighted regression (GWR) method, the green space morphology thresholds obtained based on the GBDT model were further analyzed, and control strategies based on these thresholds were proposed. The GWR method can take into account spatial heterogeneity, making the control strategy more locally adaptable. This threshold-based control strategy can guide urban managers to implement targeted green space planning in different regions to maximize its environmental benefits. By combining the distribution map of polluted residential areas generated by permanent population data, the correlation between the core morphology indicators and aerosol concentrations in different pollution zones was analyzed. Based on these analysis results, core control indicators, key control indicators and general control indicators were proposed for each type of management unit. Combined with urban planning, these hierarchical control strategies can effectively reduce the impact of aerosol pollution on residential areas and improve the quality of life of residents. By constructing a least squares model and analyzing the spatial distribution of the model residuals, potential control variables that should be included in the model are identified. This residual analysis method helps to discover the deficiencies in the existing model and optimize the spatial morphology design of special functional areas by adjusting the model to include more control variables, further reducing the impact of aerosol pollution on residents. By comparing the differences between administrative districts with extremely low and extremely high green coverage before and after the implementation of the regulation strategy, the actual effect of the regulation strategy can be evaluated. This comparison not only verifies the effectiveness of the regulation strategy, but also provides a practical basis for further optimizing the core indicators of green space morphology. Ultimately, by continuously optimizing the core indicators, the role of urban green space in reducing aerosol pollution can be better realized. These steps further improve the efficiency and effectiveness of urban green space morphology in reducing aerosol pollution through scientific analysis and strategy formulation. Refined regulation strategies enable urban management to deal with pollution problems in different regions in a more targeted manner, thereby providing residents with a healthier living environment.

[0148] In addition to the features of the above embodiments, this embodiment further defines: the preset value is 30m; the specific steps of outputting the first data are: statistical attribute table data, exporting to EXCEL, and saving as CSV format data; the specific steps of outputting the second data are: extracting the reclassified new values ​​into the attribute fields of different grids, and exporting to EXCEL tables, and storing in CSV format; the multiple types of indicators are area, density, boundary length, shape, agglomeration, connectivity, and core area indicators; the pollution intensity areas include severe pollution areas, heavy pollution areas, moderate pollution areas, light pollution areas, low pollution areas, and extremely low pollution areas.

[0149] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model, characterized in that: The method comprises the following steps: S1. Obtain the aerosol observation network database and apparent reflectance data, and analyze and obtain the aerosol optical depth; S2. After adjusting the accuracy of the aerosol data to a preset value using a machine learning method, the aerosol data is segmented using ARCGIS software to obtain a vector diagram of aerosol spatial distribution, and the first data is output; S3. reclassify it in ArcGIS and distinguish the pollution intensity areas based on the urban planning management unit grid, and output the second data; S4. Obtain a land use image map, process the land use image map and obtain green space information, process the green space information, segment the fishing net based on the urban planning management unit, and create a serial number for each fishing net in the management unit; S5. batch-process the extracted green space information using Fragstats software, divide it into multiple morphological indicators, and obtain attribute data and vector data of the morphological indicators; S6. Select potential morphological factors that have an impact on reducing aerosols, analyze them, and obtain green space morphological indicators; S7. Process the morphological indicators and use the GBDT model to construct the core indicators of green space morphology, determine the regularity model and threshold; S8. Propose control strategies based on core indicators, regularity models and thresholds of green space morphology.

2. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 1 is characterized in that: The step S1 comprises the following steps: S11. Obtain the AERONET global ground-based aerosol observation network database for inversion verification of satellite remote sensing aerosols; S12. Obtain OLI data from NASA’s official website; S13. Extract Landsat8 apparent reflectance data using Google Earth Engine; S14. Select the AERONET station measurement values ​​within the preset time before and after the satellite passes to match with the satellite data. If there are multiple sets of station measurement values ​​within the interval, linear interpolation is used to obtain the aerosol AOD at the current moment. The expression of the aerosol AOD at the current moment is as follows: Among them, τ is the AOD of aerosol at the time of imaging; ta and tb are the two time points closest to the imaging time; τa and τb are the AOD at the corresponding time; t is the imaging time of Landsat8 satellite; S15. Estimation of aerosol optical depth using deep belief neural network.

3. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 2 is characterized in that: The step S15 comprises the following steps: S151. Input the Landsat8 apparent reflectance data into the DBN model, use the RBM model to perform unsupervised pre-training on the collected samples layer by layer, extract the necessary features through unsupervised pre-training and pass them to the next layer of RBM model; S152. Generate initial weights using the result of step S151 and calculate AOD; then calculate the root mean square error between the AOD estimated by the model and the actual site data of AERONET, return the error to the BP layer and perform repeated fine-tuning until the DBN model achieves satisfactory performance; S153. The DBN model was cross-validated before use. When the accuracy was met, it was used to estimate the AOD value in areas without ground monitoring stations, thereby further obtaining the aerosol optical thickness distribution of the Landsat8 satellite. S154. A ten-fold cross-validation technique was used to test the fitting and estimation capabilities of the DBN model, and a sample-based cross-validation was used to create an independent validation dataset.

4. The method for reducing aerosol pollution by high-resolution urban green space morphology based on the GBDT model according to claim 2 is characterized in that: The spatial resolution of the OLI data is 30m; The Landsat8 apparent reflectance data includes: TOA reflectance data, radiation spectrum, observation geometry data, NDVI data and quality control band data; The preset time is 30 minutes.

5. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 4 is characterized in that: The TOA reflectivity data are TOA reflectivity data in the B1 to B7 bands, the wavelength range of the radiation spectrum is from the wavelength corresponding to visible light to the wavelength corresponding to near infrared, and the observation geometry data are the solar altitude angle and the relative azimuth angle.

6. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 2 is characterized in that: The step S6 comprises the following steps: S61. Select potential morphological factors that affect aerosol reduction, connect the AOD data of aerosols and the morphological indicators of green spaces to the planning management unit grid of ArcGIS, obtain one-to-one corresponding attribute data, import them into ArcGIS for visualization, observe their spatial distribution, and then export the attribute table; S62. Use the Spearman bivariate correlation analysis of SPSS software to preliminarily screen the green space morphological data that affect aerosol pollution, retain the green space morphological indicators with large and significant correlation coefficients with aerosols, and eliminate the insignificant morphological indicators; S63. Perform pairwise Spearman correlation analysis on the green space morphology indicators preliminarily screened above. If the correlation coefficients between the selected green space morphology indicators are small or insignificant, the obviously significant indicators are eliminated and the green space morphology indicators are screened out for the second time.

7. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 1 is characterized in that: The step S4 comprises the following steps: S41. Download 0.5m pixel high-resolution land use imagery from Google Earth in slices; S42. Obtaining Google Earth digital map data and ground truth sample data for verifying the extracted green space information; S43. Use ENVI software to preprocess the original remote sensing image data, including atmosphere removal, radiation calibration, and geometric correction; S44. Selecting spectral information, texture information, and morphological information features to establish a green space extraction model; S45. Training the green space extraction model using ground truth sample data; S46. Insert the results of ENVI area of ​​interest, ENVI feature markers, and ENVI traditional classification method into ENVIDeepLearning Module, and insert the green space marker data obtained from Open Street Maps and ArcGIS Pro platform into ENVIDeep Learning Module to obtain training samples; S47. When the above training samples cannot meet the requirements, a new green space extraction model is created, or deep processing is performed on the existing green space extraction model; S48. Extracting the required green space vector from the image through the green space extraction model; S49. Post-processing the extracted green space vector results, including vector result checking and editing, vector data splicing and clipping, and attribute assignment; S410. The 0.5m high-precision green space vector data obtained above is segmented into fishing nets based on urban planning management units, and a new Arabic numeral serial number of each management unit fishing net is created in the attribute table.

8. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 1 is characterized in that: The step S7 comprises the following steps: S71. The green space morphological indicators were randomly divided into training set and test set in a ratio of 8:2, and the GBM package of R language was used to fit the model; S72. Use the GBDT model to construct core indicators of green space morphology, determine regularity models and thresholds.

9. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 8 is characterized in that: The step S72 includes the following steps: S721. Based on the GBDT model, we obtain the core indicators of green space morphology for reducing aerosols, reveal the regularity model and threshold effect, and use it as a morphology evaluation tool to form a model tool for green space morphology to reduce aerosol pollution; S722. Based on the threshold of green space morphology obtained by GBDT model and combined with geographically weighted regression (GWR) method, a threshold-based regulation strategy is proposed; S723. Combined with the permanent population data, a distribution map of polluted residential areas is generated. Based on the correlation between the core morphological indicators and aerosol concentrations of different pollution zones, core control indicators, key control indicators, and general control indicators are proposed for each type of management unit. Combined with the planning, a hierarchical control strategy for the spatial morphology of different types of pollution units is proposed; S724. Four least squares models were constructed with the core spatial morphology index as the explanatory variable and the aerosol of each management unit as the dependent variable. The residuals of each standard unit were counted. The spatial distribution of the residuals in the simulation results was used to find potential control variables that should be included in the model, so as to further control and optimize the spatial morphology of special functional areas and reduce the impact of aerosol pollution on residents. S725. Based on the management unit, several administrative areas with extremely low green coverage and extremely high green coverage were selected, and the differences in the regulation strategies before and after optimization of these two types of sample sites were compared to further optimize the core indicators of green space morphology.

10. The method for reducing aerosol pollution based on high-resolution urban green space morphology of GBDT model according to claim 1 is characterized in that: The preset value is 30m; The specific steps of outputting the first data are: counting the attribute table data, exporting to EXCEL, and saving as CSV format data; The specific steps of outputting the second data are: extracting the reclassified new values ​​into the attribute fields of different grids, exporting them to an EXCEL table, and storing them in a CSV format; The multiple types of indicators are area, density, boundary length, shape, agglomeration, connectivity, and core area indicators; The pollution intensity areas include severe pollution areas, heavy pollution areas, moderate pollution areas, light pollution areas, low pollution areas, and extremely low pollution areas.

Images