Method for analyzing influence of vegetation coverage on aerosol concentration through LAI data

Abstract

The invention relates to a method for analyzing the influence of vegetation coverage on aerosol concentration through LAI data, and belongs to the field of environmental climate protection. The invention discloses a method for analyzing the influence of vegetation coverage on aerosol concentration by LAI data. The method comprises the following steps: S1, acquiring data and producing a long-time-sequence, high-consistency and time-space continuous LAI product; s2, MODIS AOD aerosol data are obtained, and data processing is carried out on the MODIS AOD aerosol data and LAI in the same period; and S3, researching the relationship between the LAI and the aerosol section by adopting a segmented quantile regression method. The invention discloses a method for analyzing the influence of vegetation coverage on aerosol concentration through LAI data, and the method for analyzing the influence of vegetation coverage on aerosol concentration through a high-resolution LAI product can accurately quantify the growth state and coverage degree of vegetation, further deeply understand whether the vegetation affects the aerosol concentration in the atmosphere, and improve the quality of the aerosol concentration. And the complex mutual relation between the ecological system and the atmospheric environment is disclosed.

Description

Technical Field

[0001] The invention relates to the field of environmental climate protection, and in particular to a method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data. Background Art

[0002] At present, the research on vegetation coverage and particulate matter pollution based on remote sensing inversion is mainly focused on two-dimensional green volume. Although some scholars have studied the quantitative relationship between vegetation coverage and AOD based on remote sensing inversion of particulate matter concentration, there is a lack of high-resolution LAI products at home and abroad, and in-depth research on leaf area index and aerosol pollution represented by AOD is rarely involved. Summary of the invention

[0003] Based on this, it is necessary to study the relationship between vegetation coverage and aerosols in view of the lack of high-resolution LAI products, and to provide a method for LAI data analysis of the impact of vegetation coverage on aerosol concentration.

[0004] A method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data, characterized in that the method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data comprises the following steps:

[0005] S1. Acquire data and produce long-term, highly consistent, and spatially continuous LAI products;

[0006] S2, obtain MODIS AOD aerosol data and process it together with the LAI of the same period;

[0007] S3. The segmented quantile regression method was used to study the relationship between LAI and aerosol segments.

[0008] The present application discloses a method for analyzing the effect of vegetation coverage on aerosol concentration using LAI data. Currently, long-term vegetation observation data is particularly important for monitoring and simulating the response and feedback of vegetation systems to climate change. The leaf area index (LAI) is identified as a key climate variable by the Global Climate Observing System. Current LAI products have uncertainty and it is difficult to capture LAI changes caused by changes in land cover types. Applications in fields such as agriculture and hydrology are in urgent need of LAI products with higher spatial resolution. Therefore, the present application uses high-resolution LAI data to evaluate and analyze the effect of vegetation coverage on aerosol concentration in the atmosphere. First, a large amount of data is obtained from various platforms to produce long-term, highly consistent, and spatiotemporally continuous LAI products. Through data for many years, the changing trends of vegetation growth, development, aging, and other changes throughout the life cycle can be captured. The long time series and high consistency of the data make the LAI values ​​obtained at different time points comparable, avoiding misjudgment of vegetation change trends due to data differences, and can greatly increase the depth and breadth of LAI product applications. The MODIS sensor is used to collect AOD aerosol data, and the high-resolution LAI data of the same period are used to process the two together. Based on the reliability of the high-resolution LAI data, by analyzing the difference in AOD aerosols between vegetation-covered areas and non-vegetation-covered areas, it is possible to intuitively infer whether the concentration of aerosols is related to the degree of vegetation coverage. Finally, the segmented quantile regression method is used to analyze and study the LAI data and aerosols. Since the relationship between LAI and aerosols may not be a simple linear relationship, the segmented quantile regression method can capture this nonlinear feature, more accurately describe the complex relationship between the two, better adapt to the characteristics of the data, reduce errors and uncertainties, and provide more accurate parameter estimates and model predictions. The method of using LAI data to analyze the effect of vegetation coverage on aerosol concentration can accurately quantify the growth state and coverage of vegetation, and then deeply understand whether vegetation affects the concentration of aerosols in the atmosphere, revealing the complex relationship between ecosystems and the atmospheric environment.

[0009] In one embodiment, the specific steps of S1 are as follows:

[0010] S11. Data collection and preprocessing;

[0011] S12. Data purification, optimization and strict screening, including element purity screening, outlier cleaning and thin cloud removal;

[0012] S13, due to the registration errors caused by the projection differences and spatial resolution differences between MODIS and Landsat, it is proposed to select homogeneous samples by expanding the spatial neighborhood window to improve the uncertainty caused by the spatial registration errors;

[0013] S14. Train a random forest model based on clean samples to produce a high-resolution LAI reference dataset covering a preset time and prediction range;

[0014] S15. Using the reference dataset of S14 and the PKU GIMMS NDVI product, a BP neural network model that takes into account the rapid expansion characteristics of forests is constructed, and the neural network model is used to produce long-term LAI products for the preset area.

[0015] S16: Patching, verifying and releasing the long-time LAI products produced by S15.

[0016] Through the collection and preprocessing of data, the first step of the study of LAI data and aerosols is achieved. The various data collected provide basic materials for subsequent analysis and modeling. Without sufficient data, it is impossible to deeply understand the relationship between LAI and aerosols and their impact on the environment and climate. Since there may be some values ​​in the data that are significantly deviated from the normal range, these outliers may be caused by measurement errors, emergencies or interference in the data collection process. Therefore, the collected data should be purified and optimized. For example, the use of element purity screening, outlier cleaning, thin cloud removal and other links can identify and correct these errors and errors, reduce the error level in the data, improve the accuracy of LAI data, and make the data closer to the actual situation. Due to the registration errors caused by the projection differences and spatial resolution differences between MODIS and Landsat, the method of expanding the spatial neighborhood window is used to select homogeneous samples to improve the uncertainty caused by spatial registration errors and improve the accuracy of data processing. Because the optimized data set can better represent the target group or phenomenon, the random forest model trained based on this data has stronger generalization ability, can perform well on new and unseen data, and can produce high-resolution LAI reference data sets covering the preset time and prediction range. By combining these reference data sets with the PKU GIMMS NDVI product, it is possible to construct a BP neural network model that takes into account the expression of rapid forest expansion characteristics. Based on the BP neural network model, Chinese long-term products can be produced, and after repairing and verifying the product, it can be released for use, providing strong support for ecological protection and sustainable development, and contributing to global ecological protection.

[0017] In one embodiment, the specific steps of S11 are as follows:

[0018] S111. Download the Landsat reflectivity data and MODIS LAI data required for the project via the GEE platform;

[0019] S112, performing preprocessing steps including quality control, sample purification and outlier cleaning on the data of S111;

[0020] S113, obtain remote sensing data, including GLASS GLC land cover and PKU GIMMS NDVI;

[0021] S114, performing pre-processing steps such as projection conversion, time synthesis and space aggregation on the remote sensing data of S113.

[0022] By downloading Landsat reflectivity data and MODIS LAI data on the Google Earth Engine (GEE) platform and performing preprocessing steps such as quality control, sample purification, and outlier cleaning, critical, correct, and useful data are provided for the production of LAI products. Since Landsat reflectivity data can be used to calculate vegetation indices, quantitative assessments of vegetation growth, coverage, biomass, etc. are provided, this is very helpful for studying ecosystem productivity, seasonal changes in vegetation, ecosystem health, and the impact of climate change on vegetation. MODIS LAI data provides large-scale, continuous LAI information that can reflect the growth status and health of vegetation. By analyzing LAI data, we can understand the growth stage, growth rate, biomass accumulation, etc. of vegetation, and understand the long-term dynamic trend of vegetation. These data provide great help for the production of LAI products. In addition to the above data, remote sensing data including GLASS GLC land cover and PKUGIMMS NDVI were also obtained, and these remote sensing data were preprocessed by projection conversion, time synthesis and spatial aggregation. Remote sensing data can quickly monitor the distribution and coverage of vegetation over a large area, provide data support for vegetation resource inventory and management, and greatly improve the completeness of the collected data.

[0023] In one embodiment, the specific steps of “training a random forest model based on clean samples” in S14 are as follows:

[0024] S141. Evaluate the spatial and temporal representativeness of random forest model training samples and supplement samples for special periods and special areas;

[0025] S142. Introduce MODIS saturated LAI pixels to train the random forest model and test the effect of sample balance on model accuracy to improve the accuracy of the random forest model.

[0026] By evaluating the temporal and spatial representativeness of the training samples of the random forest model, it can be determined whether the training samples can fully and accurately reflect the characteristics and changes of the research objects in different time and space, avoid inaccurate prediction results, and supplement samples in special periods and special areas, so that the model can better capture the patterns and laws in various situations and improve the accuracy of the model. Reasonable sample selection and supplementation can reduce the computational cost of the model, improve the training and prediction efficiency of the model, and samples with good temporal and spatial representativeness can make the model more stable and reduce the risk of overfitting and underfitting. By introducing MODIS saturated LAI pixels to train the random forest model, the diversity of training data is increased, and the prediction accuracy of the model in high vegetation coverage areas is improved. The impact of sample balance on model accuracy cannot be ignored. When dealing with unbalanced data sets, random forests have better robustness than other models due to their integration characteristics based on multiple decision trees. However, when the category distribution of the data set is extremely unbalanced, even random forests may tend to the majority class, thereby ignoring the characteristics of the minority class. Therefore, the use of sample balancing technology can help the model better learn the characteristics of all categories and improve the prediction accuracy of the model for the minority class, thereby achieving the purpose of improving the accuracy of the random forest model.

[0027] In one embodiment, the specific steps of S16 are as follows:

[0028] S161. Develop a long short-term memory neural network algorithm to perform spatiotemporal data patching on long-term LAI products in a preset area, and use a variety of product verification methods to systematically evaluate the uncertainty of the produced LAI products;

[0029] S162. Release the long-term, highly consistent spatiotemporal continuous LAI products and their quality control documents in the preset area after product verification and quality assessment, and release the 30m Landsat LAI and 10m Sentinel LAI inversion models in the preset area online on the GEE platform.

[0030] By developing a long short-term memory neural network algorithm, it is possible to predict the LAI value in the future time period based on the spatiotemporal correlation of the data according to the existing LAI time series data, and it is possible to perform spatiotemporal repair on the long-time series LAI products in the preset area. The development of a long short-term memory neural network algorithm can smooth the LAI time series, reduce data fluctuations caused by sensor noise or environmental factors, improve data quality, and provide support for further data analysis and ecological research. A variety of product verification methods are then used to systematically evaluate the uncertainty of the produced LAI products, which can identify errors and uncertainties in LAI products, provide a basis for further optimizing algorithms and improving data processing, and a comprehensive assessment of uncertainty helps to enhance users' trust in LAI products and ensure the reliability of data in scientific research and practical applications. Release long-term, highly consistent spatiotemporal continuous LAI products and their quality control files within the preset area after product verification and quality assessment. At the same time, release the 30m Landsat LAI and 10m Sentinel LAI inversion models within the preset area online on the GEE platform. By using the powerful computing power and user interface of GEE, users around the world can easily access and process LAI data. These data and tools can not only support the research on aerosols, but also provide help in many fields such as ecology, agriculture, environmental science, climate change, etc. At the same time, release the inversion model code to the GEE platform to encourage other researchers to use, modify and expand these models to promote scientific collaboration and knowledge sharing.

[0031] In one embodiment, the specific steps of S2 are as follows:

[0032] S21. Obtain MODIS AOD aerosol data, use ENVI software to crop and calibrate, use ENVI modeler and LUTS table to invert aerosol AOD, and use machine learning methods to improve accuracy;

[0033] S22. Select LAI data of the same period as aerosol data, and use ArcGIS software to perform fishing net segmentation on LAI and AOD data based on urban planning management units;

[0034] S23. Count the attribute data of the urban planning management unit of each LAI and AOD data, match the data of the same management unit in pairs, and then export them to Excel and save them as a file in CSV format.

[0035] By acquiring MODIS AOD aerosol data and using ENVI software to remove unnecessary areas after cropping and correction, the data set is made more targeted and refined. Cropping is used to extract data in the study area, while correction ensures the accuracy of the data's geographic positioning, providing a reliable basis for subsequent analysis. ENVI modeler and LUTS table are then used to invert aerosol AOD, which can more accurately simulate the optical properties of aerosols under different conditions, thereby improving the accuracy of inverted AOD and reusing it. Finally, machine learning methods are used to further improve the accuracy of MODIS AOD aerosol data. Based on the processed MODIS AOD aerosol data, we obtain the LAI data of the same period. Based on the urban planning management unit, ArcGIS software is used to perform fishing net segmentation on the LAI and AOD data respectively. These large-area data sets can be divided into smaller units that are easier to manage and process, which is convenient for users to perform spatial statistics and analysis. Then, the attribute data of the urban planning management unit of each LAI and AOD data is counted, and the data of the same management unit are matched two by two, which can more intuitively observe the impact of LAI data on AOD aerosol data. Finally, these data are exported to Excel and saved as CSV format files, which can be easily transferred and used between multiple operating systems and different programs, facilitating data sharing between different users or systems.

[0036] In one embodiment, the specific steps of S3 are as follows:

[0037] S31. confirm the number of segments in the segmented quantile regression method and screen the segments;

[0038] S32, segmenting the LAI, and establishing a quantile regression model for each segment based on the segmented quantile regression method, selecting an appropriate quantile and estimating parameters using the quantile regression method;

[0039] S33, extracting a boundary line around the “scattered point cloud”, and determining whether the boundary line belongs to an upper boundary or a lower boundary type according to the shape of the scattered point cloud;

[0040] S34. In order to eliminate the influence of outliers, the 95% quantile of each part was obtained using MATLAB program, and 10%, 25%, 50%, 75%, and 90% were tried as boundary points, and Origin 8 software was used to fit these boundary points to obtain the constraint lines;

[0041] S35, identifying the type of the constraint line of S34 according to the fitted R2 and analyzing, studying and utilizing it;

[0042] S36. Obtain a quantile regression model based on the constraint line type of S35, and estimate and evaluate the model.

[0043] By segmenting the LAI and confirming the number of segments in the segmented quantile regression method, the nonlinear relationship and complex patterns in the data can be better captured. Reasonable selection of the number of segments can avoid the overfitting problem caused by the model being too complex. At the same time, the number of segments also affects the explanatory power and prediction accuracy of the quantile regression model. A quantile regression model is established for each LAI segment, and the appropriate quantile is selected and the parameters are estimated using the quantile regression method. First, quantile regression provides a robust method to estimate the central tendency and dispersion of the data. Secondly, different quantiles can help reveal the influencing factors of LAI data changes under different environmental conditions. By comparing the regression results of different quantiles, the robustness of the model for different data subsets can be evaluated. In the LAI data segment, the boundary line around the "scatter cloud" is extracted. According to the shape of the scatter cloud, it is determined whether the boundary line belongs to the upper bound or lower bound type, which is convenient for better confirmation of the mathematical equation and provides assistance for establishing the quantile regression model. The MATLAB program was used to obtain the 95% quantile of each part to eliminate the influence of outliers, and then the Origin 8 software was used to fit these boundary points to obtain the constraint lines. The constraint lines were analyzed and studied, and a quantile regression model was established. Finally, the model was estimated and evaluated, and the reasons why LAI data affected the reduction of aerosol concentration were found.

[0044] In one embodiment, the specific steps of S31 are as follows:

[0045] S311. Preset the number of LAI segments;

[0046] S312. Data exploration: Use box plot analysis to find outliers or natural groupings of data, or use histograms and cumulative distribution functions to observe the distribution characteristics of data;

[0047] S313. Determine the number of segments based on current knowledge and experience;

[0048] S314, Model selection criteria: Use statistical criteria to assist in determining the number of segments;

[0049] S315, cross validation: cross validation is performed on different numbers of segments, and the number of segments with the best prediction performance is selected;

[0050] S316, trade-off between model complexity and interpretability: Increasing the number of segments can increase the complexity of the model and may result in a better fit, but it may also make the model too complex to interpret. It is necessary to strike a balance between the complexity and interpretability of the model;

[0051] S317, Stepwise Regression: Determine the optimal number of segments by stepwise regression method, and evaluate the performance change of the model by gradually adding or removing segments;

[0052] S318, Model diagnosis: Diagnose models with different numbers of segments to check whether there are violations of model assumptions, including the distribution and heteroscedasticity of residuals.

[0053] Segmenting the LAI data is a very important step. Segmentation helps to analyze LAI data at different scales, from local to global, and better understand the scale effect. Segmentation has the following steps and methods. First, the number of segments is preset based on existing knowledge and experience. Secondly, the number of segments can be confirmed through data exploration, model selection criteria and cross-validation. In the process of segmentation, it should be noted that the number of segments should not be too many. Although too many segments can increase the complexity of the model and may obtain a better fit, it may also cause the model to be too complex and difficult to explain. It is necessary to balance the complexity and interpretability of the model. The optimal number of segments can also be confirmed by stepwise regression. Finally, the models with different numbers of segments are diagnosed to check whether there are violations of model assumptions, including the distribution and heteroscedasticity of residuals, to ensure that the selected model is suitable for the specific segmentation of the data, and to check whether the model can reasonably capture the characteristics of the data.

[0054] In one embodiment, the specific steps of S35 are as follows:

[0055] S351. Study the shape and characteristics of the constraint line, including convex wave type, hump type or exponential type, analyze the specific constraint relationship of aerosol reduction in different LAI segments through characteristics, use it as an evaluation tool for LAI aerosol reduction, and extract the threshold when a threshold exists;

[0056] S352. Identify and analyze the main sections that affect the constraint line, and understand how these sections affect the aerosol interaction and constraint effect;

[0057] S353. Use the constraint line method to simulate the potential changes in aerosols and predict the response of aerosol concentrations under different segment scenarios;

[0058] S354. Evaluate the scale dependence of the constraint line method to ensure the validity and applicability of the analysis results at different scales.

[0059] By studying the shape and characteristics of the constraint line, and analyzing the specific constraint relationship of aerosol reduction in different LAI segments based on these characteristics, identifying the constraint relationship of different LAI segments on aerosol reduction is helpful to formulate targeted environmental improvement measures, and can also evaluate the potential impact of vegetation coverage on aerosol concentration, and then understand the ecological effect of vegetation on air quality. Using the constraint line as an evaluation tool for LAI aerosol reduction helps to confirm the threshold between vegetation coverage and aerosol reduction, that is, when a certain LAI value is reached, the aerosol concentration may change significantly. By identifying and analyzing the main segments that affect the constraint line, and using the constraint line method to simulate the potential changes of aerosols, we can understand which factors have a significant impact on aerosol concentration and distribution, understand how these segments affect the interaction and constraint effect of aerosols, and help reveal how the concentration and distribution of aerosols are affected by ecological processes. Identifying the aerosol concentration in different segments also helps to assess the air quality risks in different regions and provide a basis for formulating risk mitigation measures, which helps to achieve environmental protection and sustainable development goals. However, when using constraints, we must also pay attention to the scale dependence of the constraint line method, which helps to understand the differences in ecological processes and patterns at different scales and ensure the validity and applicability of the analysis results at different scales.

[0060] In one embodiment, the specific steps of “estimating and evaluating the model” in S36 are as follows:

[0061] S361. Model estimation: Use MATLAB to estimate the parameters of the quantile regression model for each segment;

[0062] S362. Model evaluation: Evaluate the goodness of fit and predictive ability of each quantile regression model, including checking the residuals and testing the significance of the model.

[0063] By using MATLAB to estimate the parameters of the quantile regression model for each segment, MATLAB can efficiently reveal the conditional distribution characteristics of LAI at different quantile levels, provide more comprehensive information, and accurately estimate the parameters of the quantile regression model, reduce the impact of outliers and extreme values, and correctly reflect the conditional distribution characteristics at different quantiles. At the same time, MATLAB can quickly perform segmentation and regression analysis on large amounts of data to improve work efficiency. The fit and predictive ability of each quantile regression model are evaluated. Through the evaluation, it can be diagnosed whether the model meets the basic assumptions, ensure that the model can accurately capture the relationship between LAI and aerosol concentration, and help select the model that best suits the LAI data. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a flow chart of a method for analyzing the effect of vegetation coverage on aerosol concentration using LAI data;

[0065] Figure 2 Schematic diagram of the process for acquiring data and producing LAI products with long time series, high consistency and spatiotemporal continuity;

[0066] Figure 3 A schematic diagram of the process of data collection and preprocessing;

[0067] Figure 4 This is a flowchart for training a random forest model based on clean samples;

[0068] Figure 5 A flow chart for patching and verifying the release of long-time LAI products;

[0069] Figure 6 Schematic diagram of the process for acquiring MODIS AOD aerosol data and processing it together with the LAI of the same period;

[0070] Figure 7 This is a schematic diagram of the process of studying the relationship between LAI and aerosol segments using the segmented quantile regression method;

[0071] Figure 8 A schematic diagram of the process for confirming the number of segments in the segmented quantile regression method and screening the segments;

[0072] Fig. 9 According to the fitted R 2 A schematic diagram of the process of identifying the types of S34's constraint lines and analyzing, studying and utilizing them;

[0073] Fig.10 Schematic diagram of the process of estimating and evaluating the quantile regression model. DETAILED DESCRIPTION

[0074] 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.

[0075] 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.

[0076] A method for analyzing the effect of vegetation coverage on aerosol concentration using LAI data according to some embodiments of the present invention will be described below with reference to the accompanying drawings.

[0077] Example

[0078] like Figures 1 to 10 As shown, this embodiment discloses a method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data, comprising the following steps:

[0079] S1. Acquire data and produce long-term, highly consistent, and spatially continuous LAI products;

[0080] S2, obtain MODIS AOD aerosol data and process it together with the LAI of the same period;

[0081] S3. The segmented quantile regression method was used to study the relationship between LAI and aerosol segments.

[0082] The present application discloses a method for analyzing the effect of vegetation coverage on aerosol concentration using LAI data. Currently, long-term vegetation observation data is particularly important for monitoring and simulating the response and feedback of vegetation systems to climate change. The leaf area index (LAI) is identified as a key climate variable by the Global Climate Observing System. Current LAI products have uncertainty and it is difficult to capture LAI changes caused by changes in land cover types. Applications in fields such as agriculture and hydrology are in urgent need of LAI products with higher spatial resolution. Therefore, the present application uses high-resolution LAI data to evaluate and analyze the effect of vegetation coverage on aerosol concentration in the atmosphere. First, a large amount of data is obtained from various platforms to produce long-term, highly consistent, and spatiotemporally continuous LAI products. Through data for many years, the changing trends of vegetation growth, development, aging, and other changes throughout the life cycle can be captured. The long time series and high consistency of the data make the LAI values ​​obtained at different time points comparable, avoiding misjudgment of vegetation change trends due to data differences, and can greatly increase the depth and breadth of LAI product applications. The MODIS sensor is used to collect AOD aerosol data, and the high-resolution LAI data of the same period are used to process the two together. Based on the reliability of the high-resolution LAI data, by analyzing the difference in AOD aerosols between vegetation-covered areas and non-vegetation-covered areas, it is possible to intuitively infer whether the concentration of aerosols is related to the degree of vegetation coverage. Finally, the segmented quantile regression method is used to analyze and study the LAI data and aerosols. Since the relationship between LAI and aerosols may not be a simple linear relationship, the segmented quantile regression method can capture this nonlinear feature, more accurately describe the complex relationship between the two, better adapt to the characteristics of the data, reduce errors and uncertainties, and provide more accurate parameter estimates and model predictions. The method of using LAI data to analyze the effect of vegetation coverage on aerosol concentration can accurately quantify the growth state and coverage of vegetation, and then deeply understand whether vegetation affects the concentration of aerosols in the atmosphere, revealing the complex relationship between ecosystems and the atmospheric environment.

[0083] like Figure 2As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of S1 are as follows:

[0084] S11. Data collection and preprocessing;

[0085] S12. Data purification, optimization and strict screening, including element purity screening, outlier cleaning and thin cloud removal;

[0086] S13, due to the registration errors caused by the projection differences and spatial resolution differences between MODIS and Landsat, it is proposed to select homogeneous samples by expanding the spatial neighborhood window to improve the uncertainty caused by the spatial registration errors;

[0087] S14. Train a random forest model based on clean samples to produce a high-resolution LAI reference dataset covering a preset time and prediction range;

[0088] S15. Using the reference dataset of S14 and the PKU GIMMS NDVI product, a BP neural network model that takes into account the rapid expansion characteristics of forests is constructed, and the neural network model is used to produce long-term LAI products for the preset area.

[0089] S16: Patching, verifying and releasing the long-time LAI products produced by S15.

[0090] Through the collection and preprocessing of data, the first step of the study of LAI data and aerosols is achieved. The various data collected provide basic materials for subsequent analysis and modeling. Without sufficient data, it is impossible to deeply understand the relationship between LAI and aerosols and their impact on the environment and climate. Since there may be some values ​​in the data that are significantly deviated from the normal range, these outliers may be caused by measurement errors, emergencies or interference in the data collection process. Therefore, the collected data should be purified and optimized. For example, the use of element purity screening, outlier cleaning, thin cloud removal and other links can identify and correct these errors and errors, reduce the error level in the data, improve the accuracy of LAI data, and make the data closer to the actual situation. Due to the registration errors caused by the projection differences and spatial resolution differences between MODIS and Landsat, the method of expanding the spatial neighborhood window is used to select homogeneous samples to improve the uncertainty caused by spatial registration errors and improve the accuracy of data processing. Because the optimized data set can better represent the target group or phenomenon, the random forest model trained based on this data has stronger generalization ability, can perform well on new and unseen data, and can produce high-resolution LAI reference data sets covering the preset time and prediction range. By combining these reference data sets with the PKU GIMMS NDVI product, it is possible to construct a BP neural network model that takes into account the expression of rapid forest expansion characteristics. Based on the BP neural network model, Chinese long-term products can be produced, and after repairing and verifying the product, it can be released for use, providing strong support for ecological protection and sustainable development, and contributing to global ecological protection.

[0091] like Figure 3 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of S11 are as follows:

[0092] S111. Download the Landsat reflectivity data and MODIS LAI data required for the project via the GEE platform;

[0093] S112, performing preprocessing steps including quality control, sample purification and outlier cleaning on the data of S111;

[0094] S113, obtain remote sensing data, including GLASS GLC land cover and PKU GIMMS NDVI;

[0095] S114, performing pre-processing steps such as projection conversion, time synthesis and space aggregation on the remote sensing data of S113.

[0096] By downloading Landsat reflectivity data and MODIS LAI data on the Google Earth Engine (GEE) platform and performing preprocessing steps such as quality control, sample purification, and outlier cleaning, critical, correct, and useful data are provided for the production of LAI products. Since Landsat reflectivity data can be used to calculate vegetation indices, quantitative assessments of vegetation growth, coverage, biomass, etc. are provided, this is very helpful for studying ecosystem productivity, seasonal changes in vegetation, ecosystem health, and the impact of climate change on vegetation. MODIS LAI data provides large-scale, continuous LAI information that can reflect the growth status and health of vegetation. By analyzing LAI data, we can understand the growth stage, growth rate, biomass accumulation, etc. of vegetation, and understand the long-term dynamic trend of vegetation. These data provide great help for the production of LAI products. In addition to the above data, remote sensing data including GLASS GLC land cover and PKUGIMMS NDVI were also obtained, and these remote sensing data were preprocessed by projection conversion, time synthesis and spatial aggregation. Remote sensing data can quickly monitor the distribution and coverage of vegetation over a large area, provide data support for vegetation resource inventory and management, and greatly improve the completeness of the collected data.

[0097] like Figure 4 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of "training a random forest model based on clean samples" in S14 are as follows:

[0098] S141. Evaluate the spatial and temporal representativeness of random forest model training samples and supplement samples for special periods and special areas;

[0099] S142. Introduce MODIS saturated LAI pixels to train the random forest model and test the effect of sample balance on model accuracy to improve the accuracy of the random forest model.

[0100] By evaluating the temporal and spatial representativeness of the training samples of the random forest model, it can be determined whether the training samples can fully and accurately reflect the characteristics and changes of the research objects in different time and space, avoid inaccurate prediction results, and supplement samples in special periods and special areas, so that the model can better capture the patterns and laws in various situations and improve the accuracy of the model. Reasonable sample selection and supplementation can reduce the computational cost of the model, improve the training and prediction efficiency of the model, and samples with good temporal and spatial representativeness can make the model more stable and reduce the risk of overfitting and underfitting. By introducing MODIS saturated LAI pixels to train the random forest model, the diversity of training data is increased, and the prediction accuracy of the model in high vegetation coverage areas is improved. The impact of sample balance on model accuracy cannot be ignored. When dealing with unbalanced data sets, random forests have better robustness than other models due to their integration characteristics based on multiple decision trees. However, when the category distribution of the data set is extremely unbalanced, even random forests may tend to the majority class, thereby ignoring the characteristics of the minority class. Therefore, the use of sample balancing technology can help the model better learn the characteristics of all categories and improve the prediction accuracy of the model for the minority class, thereby achieving the purpose of improving the accuracy of the random forest model.

[0101] like Figure 5 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of S16 are as follows:

[0102] S161. Develop a long short-term memory neural network algorithm to perform spatiotemporal data patching on long-term LAI products in a preset area, and use a variety of product verification methods to systematically evaluate the uncertainty of the produced LAI products;

[0103] S162. Release the long-term, highly consistent spatiotemporal continuous LAI products and their quality control documents in the preset area after product verification and quality assessment, and release the 30m Landsat LAI and 10m Sentinel LAI inversion models in the preset area online on the GEE platform.

[0104] By developing a long short-term memory neural network algorithm, it is possible to predict the LAI value in the future time period based on the spatiotemporal correlation of the data according to the existing LAI time series data, and it is possible to perform spatiotemporal repair on the long-time series LAI products in the preset area. The development of a long short-term memory neural network algorithm can smooth the LAI time series, reduce data fluctuations caused by sensor noise or environmental factors, improve data quality, and provide support for further data analysis and ecological research. A variety of product verification methods are then used to systematically evaluate the uncertainty of the produced LAI products, which can identify errors and uncertainties in LAI products, provide a basis for further optimizing algorithms and improving data processing, and a comprehensive assessment of uncertainty helps to enhance users' trust in LAI products and ensure the reliability of data in scientific research and practical applications. Release long-term, highly consistent spatiotemporal continuous LAI products and their quality control files within the preset area after product verification and quality assessment. At the same time, release the 30m Landsat LAI and 10m Sentinel LAI inversion models within the preset area online on the GEE platform. By using the powerful computing power and user interface of GEE, users around the world can easily access and process LAI data. These data and tools can not only support the research on aerosols, but also provide help in many fields such as ecology, agriculture, environmental science, climate change, etc. At the same time, release the inversion model code to the GEE platform to encourage other researchers to use, modify and expand these models to promote scientific collaboration and knowledge sharing.

[0105] like Figure 6 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of S2 are as follows:

[0106] S21. Obtain MODIS AOD aerosol data, use ENVI software to crop and calibrate, use ENVI modeler and LUTS table to invert aerosol AOD, and use machine learning methods to improve accuracy;

[0107] S22. Select LAI data of the same period as aerosol data, and use ArcGIS software to perform fishing net segmentation on LAI and AOD data based on urban planning management units;

[0108] S23. Count the attribute data of the urban planning management unit of each LAI and AOD data, match the data of the same management unit in pairs, and then export them to Excel and save them as a file in CSV format.

[0109] By acquiring MODIS AOD aerosol data and using ENVI software to remove unnecessary areas after cropping and correction, the data set is made more targeted and refined. Cropping is used to extract data in the study area, while correction ensures the accuracy of the data's geographic positioning, providing a reliable basis for subsequent analysis. ENVI modeler and LUTS table are then used to invert aerosol AOD, which can more accurately simulate the optical properties of aerosols under different conditions, thereby improving the accuracy of inverted AOD and reusing it. Finally, machine learning methods are used to further improve the accuracy of MODIS AOD aerosol data. Based on the processed MODIS AOD aerosol data, we obtain the LAI data of the same period. Based on the urban planning management unit, ArcGIS software is used to perform fishing net segmentation on the LAI and AOD data respectively. These large-area data sets can be divided into smaller units that are easier to manage and process, which is convenient for users to perform spatial statistics and analysis. Then, the attribute data of the urban planning management unit of each LAI and AOD data is counted, and the data of the same management unit are matched two by two, which can more intuitively observe the impact of LAI data on AOD aerosol data. Finally, these data are exported to Excel and saved as CSV format files, which can be easily transferred and used between multiple operating systems and different programs, facilitating data sharing between different users or systems.

[0110] like Figure 7 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of S3 are as follows:

[0111] S31. confirm the number of segments in the segmented quantile regression method and screen the segments;

[0112] S32, segmenting the LAI, and establishing a quantile regression model for each segment based on the segmented quantile regression method, selecting an appropriate quantile and estimating parameters using the quantile regression method;

[0113] S33, extracting a boundary line around the “scattered point cloud”, and determining whether the boundary line belongs to an upper boundary or a lower boundary type according to the shape of the scattered point cloud;

[0114] S34. In order to eliminate the influence of outliers, the 95% quantile of each part was obtained using MATLAB program, and 10%, 25%, 50%, 75%, and 90% were tried as boundary points, and Origin 8 software was used to fit these boundary points to obtain the constraint lines;

[0115] S35, according to the fitted R 2 Identify the types of S34's restraining lines and analyze, study and utilize them;

[0116] S36. Obtain a quantile regression model based on the constraint line type of S35, and estimate and evaluate the model.

[0117] By segmenting the LAI and confirming the number of segments in the segmented quantile regression method, the nonlinear relationship and complex patterns in the data can be better captured. Reasonable selection of the number of segments can avoid the overfitting problem caused by the model being too complex. At the same time, the number of segments also affects the explanatory power and prediction accuracy of the quantile regression model. A quantile regression model is established for each LAI segment, and the appropriate quantile is selected and the parameters are estimated using the quantile regression method. First, quantile regression provides a robust method to estimate the central tendency and dispersion of the data. Secondly, different quantiles can help reveal the influencing factors of LAI data changes under different environmental conditions. By comparing the regression results of different quantiles, the robustness of the model for different data subsets can be evaluated. In the LAI data segment, the boundary line around the "scatter cloud" is extracted. According to the shape of the scatter cloud, it is determined whether the boundary line belongs to the upper bound or lower bound type, which is convenient for better confirmation of the mathematical equation and provides assistance for establishing the quantile regression model. The MATLAB program was used to obtain the 95% quantile of each part to eliminate the influence of outliers, and then the Origin 8 software was used to fit these boundary points to obtain the constraint lines. The constraint lines were analyzed and studied, and a quantile regression model was established. Finally, the model was estimated and evaluated, and the reasons why LAI data affected the reduction of aerosol concentration were found.

[0118] like Figure 8 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of S31 are as follows:

[0119] S311. Preset the number of LAI segments;

[0120] S312. Data exploration: Use box plot analysis to find outliers or natural groupings of data, or use histograms and cumulative distribution functions to observe the distribution characteristics of data;

[0121] S313. Determine the number of segments based on current knowledge and experience;

[0122] S314, Model selection criteria: Use statistical criteria to assist in determining the number of segments;

[0123] S315, cross validation: cross validation is performed on different numbers of segments, and the number of segments with the best prediction performance is selected;

[0124] S316, trade-off between model complexity and interpretability: Increasing the number of segments can increase the complexity of the model and may result in a better fit, but it may also make the model too complex to interpret. It is necessary to strike a balance between the complexity and interpretability of the model;

[0125] S317, Stepwise Regression: Determine the optimal number of segments by stepwise regression method, and evaluate the performance change of the model by gradually adding or removing segments;

[0126] S318, Model diagnosis: Diagnose models with different numbers of segments to check whether there are violations of model assumptions, including the distribution and heteroscedasticity of residuals.

[0127] Segmenting the LAI data is a very important step. Segmentation helps to analyze LAI data at different scales, from local to global, and better understand the scale effect. Segmentation has the following steps and methods. First, the number of segments is preset based on existing knowledge and experience. Secondly, the number of segments can be confirmed through data exploration, model selection criteria and cross-validation. In the process of segmentation, it should be noted that the number of segments should not be too many. Although too many segments can increase the complexity of the model and may obtain a better fit, it may also cause the model to be too complex and difficult to explain. It is necessary to balance the complexity and interpretability of the model. The optimal number of segments can also be confirmed by stepwise regression. Finally, the models with different numbers of segments are diagnosed to check whether there are violations of model assumptions, including the distribution and heteroscedasticity of residuals, to ensure that the selected model is suitable for the specific segmentation of the data, and to check whether the model can reasonably capture the characteristics of the data.

[0128] like Fig. 9 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of S35 are as follows:

[0129] S351. Study the shape and characteristics of the constraint line, including convex wave type, hump type or exponential type, analyze the specific constraint relationship of aerosol reduction in different LAI segments through characteristics, use it as an evaluation tool for LAI aerosol reduction, and extract the threshold when a threshold exists;

[0130] S352. Identify and analyze the main sections that affect the constraint line, and understand how these sections affect the aerosol interaction and constraint effect;

[0131] S353. Use the constraint line method to simulate the potential changes in aerosols and predict the response of aerosol concentrations under different segment scenarios;

[0132] S354. Evaluate the scale dependence of the constraint line method to ensure the validity and applicability of the analysis results at different scales.

[0133] By studying the shape and characteristics of the constraint line, and analyzing the specific constraint relationship of aerosol reduction in different LAI segments based on these characteristics, identifying the constraint relationship of different LAI segments on aerosol reduction is helpful to formulate targeted environmental improvement measures, and can also evaluate the potential impact of vegetation coverage on aerosol concentration, and then understand the ecological effect of vegetation on air quality. Using the constraint line as an evaluation tool for LAI aerosol reduction helps to confirm the threshold between vegetation coverage and aerosol reduction, that is, when a certain LAI value is reached, the aerosol concentration may change significantly. By identifying and analyzing the main segments that affect the constraint line, and using the constraint line method to simulate the potential changes of aerosols, we can understand which factors have a significant impact on aerosol concentration and distribution, understand how these segments affect the interaction and constraint effect of aerosols, and help reveal how the concentration and distribution of aerosols are affected by ecological processes. Identifying the aerosol concentration in different segments also helps to assess the air quality risks in different regions and provide a basis for formulating risk mitigation measures, which helps to achieve environmental protection and sustainable development goals. However, when using constraints, we must also pay attention to the scale dependence of the constraint line method, which helps to understand the differences in ecological processes and patterns at different scales and ensure the validity and applicability of the analysis results at different scales.

[0134] like Fig.10 As shown, in addition to the features of the above embodiment, this embodiment further defines: the specific steps of "estimating and evaluating the model" in S36 are as follows:

[0135] S361. Model estimation: Use MATLAB to estimate the parameters of the quantile regression model for each segment;

[0136] S362. Model evaluation: Evaluate the goodness of fit and predictive ability of each quantile regression model, including checking the residuals and testing the significance of the model.

[0137] By using MATLAB to estimate the parameters of the quantile regression model for each segment, MATLAB can efficiently reveal the conditional distribution characteristics of LAI at different quantile levels, provide more comprehensive information, and accurately estimate the parameters of the quantile regression model, reduce the impact of outliers and extreme values, and correctly reflect the conditional distribution characteristics at different quantiles. At the same time, MATLAB can quickly perform segmentation and regression analysis on large amounts of data to improve work efficiency. The fit and predictive ability of each quantile regression model are evaluated. Through the evaluation, it can be diagnosed whether the model meets the basic assumptions, ensure that the model can accurately capture the relationship between LAI and aerosol concentration, and help select the model that best suits the LAI data.

[0138] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] The above-mentioned 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 ordinary technicians in this field, 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 analyzing the effect of vegetation coverage on aerosol concentration using LAI data, characterized in that: The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data comprises the following steps: S1. Acquire data and produce long-term, highly consistent, and spatially continuous LAI products; S2, obtain MODIS AOD aerosol data and process it together with the LAI of the same period; S3. The segmented quantile regression method was used to study the relationship between LAI and aerosol segments.

2. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 1, characterized in that: The specific steps of S1 are as follows: S11. Data collection and preprocessing; S12. Data purification, optimization and strict screening, including element purity screening, outlier cleaning and thin cloud removal; S13, due to the registration errors caused by the projection differences and spatial resolution differences between MODIS and Landsat, it is proposed to select homogeneous samples by expanding the spatial neighborhood window to improve the uncertainty caused by the spatial registration errors; S14. Train a random forest model based on clean samples to produce a high-resolution LAI reference dataset covering a preset time and prediction range; S15. Using the reference dataset of S14 and the PKU GIMMS NDVI product, a BP neural network model that takes into account the rapid expansion characteristics of forests is constructed, and the neural network model is used to produce long-term LAI products for the preset area. S16: Patching, verifying and releasing the long-time LAI products produced by S15.

3. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 2 is characterized in that: The specific steps of S11 are as follows: S111. Download the Landsat reflectivity data and MODIS LAI data required for the project via the GEE platform; S112, performing preprocessing steps including quality control, sample purification and outlier cleaning on the data of S111; S113, obtain remote sensing data, including GLASS GLC land cover and PKU GIMMS NDVI; S114, performing pre-processing steps such as projection conversion, time synthesis and space aggregation on the remote sensing data of S113.

4. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 2 is characterized in that: The specific steps of "training a random forest model based on clean samples" in S14 are as follows: S141. Evaluate the spatial and temporal representativeness of random forest model training samples and supplement samples for special periods and special areas; S142. Introduce MODIS saturated LAI pixels to train the random forest model and test the effect of sample balance on model accuracy to improve the accuracy of the random forest model.

5. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 2, characterized in that: The specific steps of S16 are as follows: S161. Develop a long short-term memory neural network algorithm to perform spatiotemporal data patching on long-term LAI products in a preset area, and use a variety of product verification methods to systematically evaluate the uncertainty of the produced LAI products; S162. Release the long-term, highly consistent spatiotemporal continuous LAI products and their quality control documents within the preset area after product verification and quality assessment, and release the 30m Landsat LAI and 10m Sentinel LAI inversion models within the preset area online on the GEE platform.

6. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 1, characterized in that: The specific steps of S2 are as follows: S21. Obtain MODIS AOD aerosol data, use ENVI software to crop and correct, use ENVImodeler and LUTS table to invert aerosol AOD, and use machine learning methods to improve accuracy; S22. Select LAI data of the same period as aerosol data, and use ArcGIS software to perform fishing net segmentation on LAI and AOD data based on urban planning management units; S23. Count the attribute data of the urban planning management unit of each LAI and AOD data, match the data of the same management unit in pairs, and then export them to Excel and save them as a file in CSV format.

7. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 1, characterized in that: The specific steps of S3 are as follows: S31. confirm the number of segments in the segmented quantile regression method and screen the segments; S32, segmenting the LAI, and establishing a quantile regression model for each segment based on the segmented quantile regression method, selecting an appropriate quantile and estimating parameters using the quantile regression method; S33, extracting a boundary line around the "scattered point cloud", and determining whether the boundary line belongs to an upper boundary or a lower boundary type according to the shape of the scattered point cloud; S34. In order to eliminate the influence of outliers, the 95% quantile of each part was obtained using MATLAB program, and 10%, 25%, 50%, 75%, and 90% were tried as boundary points, and Origin 8 software was used to fit these boundary points to obtain the constraint lines; S35, according to the fitted R 2 Identify the types of S34's restraining lines and analyze, study and utilize them; S36. Obtain a quantile regression model based on the constraint line type of S35, and estimate and evaluate the model.

8. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 7, characterized in that: The specific steps of S31 are as follows: S311, preset the number of LAI segments; S312. Data exploration: Use box plot analysis to find outliers or natural groupings of data, or use histograms and cumulative distribution functions to observe the distribution characteristics of data; S313. Determine the number of segments based on current knowledge and experience; S314, Model selection criteria: Use statistical criteria to assist in determining the number of segments; S315, cross validation: cross validation is performed on different numbers of segments, and the number of segments with the best prediction performance is selected; S316, trade-off between model complexity and interpretability: Increasing the number of segments can increase the complexity of the model and may result in a better fit, but it may also make the model too complex to interpret. It is necessary to strike a balance between the complexity and interpretability of the model; S317, Stepwise Regression: Determine the optimal number of segments by stepwise regression method, and evaluate the performance change of the model by gradually adding or removing segments; S318, Model diagnosis: Diagnose models with different numbers of segments to check whether there are violations of model assumptions, including the distribution and heteroscedasticity of residuals.

9. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 7, characterized in that: The specific steps of S35 are as follows: S351. Study the shape and characteristics of the constraint line, including convex wave type, hump type or exponential type, analyze the specific constraint relationship of aerosol reduction in different LAI segments through characteristics, use it as an evaluation tool for LAI aerosol reduction, and extract the threshold when a threshold exists; S352. Identify and analyze the main sections that affect the constraint line, and understand how these sections affect the aerosol interaction and constraint effect; S353. Use the constraint line method to simulate the potential changes in aerosols and predict the response of aerosol concentrations under different segment scenarios; S354. Evaluate the scale dependence of the constraint line method to ensure the validity and applicability of the analysis results at different scales.

10. The method for analyzing the influence of vegetation coverage on aerosol concentration using LAI data according to claim 7, characterized in that: The specific steps of "estimating and evaluating the model" in S36 are as follows: S361. Model estimation: Use MATLAB to estimate the parameters of the quantile regression model for each segment; S362. Model evaluation: Evaluate the goodness of fit and predictive ability of each quantile regression model, including checking the residuals and testing the significance of the model.

Images