Method for detecting chemical properties and particle size distribution of aerosol organic matter

Abstract

The application discloses a method for detecting chemical characteristics and particle size distribution of aerosol organic matters, and belongs to the field of environmental monitoring technology and data analysis, and specifically comprises the following steps: step one, collecting atmospheric particulates by using a particle size grading particulate matter collector, and obtaining three-dimensional fluorescence spectra of the particle size grading atmospheric particulates by using a fluorescence spectrophotometer; step two, performing PARAFAC analysis on the three-dimensional fluorescence spectrum data set, decomposing the three-dimensional fluorescence spectrum data set into a plurality of optimal components, and obtaining fluorescence intensity values of the components; step three, taking the fluorescence intensity values of the components as input data of SOM simulation, and simulating a unified distance matrix, sample mapping distribution, component plane and k-means clustering; and step four, according to the k-means clustering result of the SOM of the particle size grading atmospheric particulates, the samples are divided into a plurality of groups for decision tree analysis, and a prediction model of the aerosol particle size range and the deposition position of the human respiratory tract is established.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring technology and data analysis, specifically relating to a method for detecting the chemical properties and particle size distribution of aerosol organic matter. Background Technology

[0002] Organic matter in aerosols has garnered widespread attention due to its potential impacts on climate, biogeochemical cycles, and human health. For example, it can absorb solar radiation, leading to radiation deficits; it can complex with heavy metals, affecting their migration and transformation; and it can generate reactive oxygen species, disrupting cellular redox balance and contributing to cardiovascular and respiratory diseases. However, due to its complex composition, elucidating the chemical properties, sources, and transformation processes of organic matter in aerosols remains challenging. Three-dimensional fluorescence spectroscopy, as a highly efficient spectroscopic technique, can provide high-resolution information for complex systems and has been applied in recent years to the study of aerosol characteristics, demonstrating advantages in characterizing the material composition, chemical properties, potential sources, and atmospheric chemical reaction mechanisms of organic matter in atmospheric particulate matter. Given the rich and complex data provided by three-dimensional fluorescence spectroscopy, the analysis results obtained using existing research methods in the aerosol field, such as regional integration, parallel factor analysis, and fluorescence index, show that the differences in fluorescence intensity among different samples are not significant, and the changes in fluorescence index values ​​are not obvious. It is difficult to accurately identify the differences in fluorophore characteristics in different seasons, sources, and atmospheric chemical processes, which limits the application of EEM in identifying aerosol chemical properties. EEM data still has great development potential in the aerosol field, and the problems of in-depth mining of EEM data and visualization of fluorophore chemical characteristics urgently need to be solved.

[0003] Obtaining accurate aerosol particle size distribution characteristics provides a scientific basis for analyzing aerosol pollution sources, exploring aerosol environmental behavior, and assessing respiratory health risks. Based on particle size and deposition location in the human respiratory system, atmospheric particulate matter is generally classified into three categories: particulate matter deposited in the head airways (HA, >4.7 μm), trachea and bronchi (TB, 1.1–4.7 μm), and alveoli (AR, <1.1 μm). Among these, fine particulate matter with a diameter less than 2.5 μm has a longer residence time in the atmosphere, a larger specific surface area, and can carry more toxic substances, posing a higher degree of harm to the human body. Toxic substances attached to atmospheric particulate matter, such as polycyclic aromatic hydrocarbons, heavy metals, pathogenic microorganisms, and allergens, act on the human respiratory system, leading to potential human health risks. Developing rapid and effective methods for identifying atmospheric particulate matter size distribution is of great significance for understanding the causes of air pollution and identifying the deposition locations of aerosols in the human respiratory tract, thereby assessing human health risk levels. Summary of the Invention

[0004] To address the problems in current aerosol research, such as the cumbersome acquisition of EEM data and insufficient application and information extraction of EEM data, which result in the inaccurate reflection of the chemical characteristics and particle size distribution of aerosol organic matter, this invention provides a method for detecting the chemical characteristics and particle size distribution of aerosol organic matter.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for detecting the chemical properties and particle size distribution of aerosol organic matter includes the following steps:

[0007] Step 1: Collect atmospheric particulate matter using a particle size fractionation sampler and obtain the three-dimensional fluorescence spectrum of the particle size fractionated atmospheric particulate matter using a fluorescence spectrophotometer.

[0008] Step 2: Perform PARAFAC analysis on the three-dimensional fluorescence spectral dataset to decompose the dataset into several optimal components and obtain the fluorescence intensity values ​​of each component.

[0009] Step 3: Use the fluorescence intensity values ​​of each component as input data for SOM simulation to obtain a unified distance matrix, sample mapping distribution, component plane and k-means clustering, and visualize the chemical characteristics of fluorophores and the differences of fluorophores between samples;

[0010] Step 4: Based on the k-means clustering results of atmospheric particulate matter SOMs classified by particle size, the samples are divided into several groups for decision tree analysis. The ratio of fluorescent components in each group is used as the independent variable, and the particle size range, i.e. the deposition location of particulate matter in the human respiratory tract, is used as the dependent variable. A prediction model for aerosol particle size range and deposition location in the human respiratory tract is established.

[0011] Step 5: Based on the established prediction model of aerosol particle size range and human respiratory tract deposition location, predict the aerosol particle size distribution and human respiratory tract deposition location of non-particle size fractionated particulate matter samples.

[0012] Furthermore, in step one, the parameters set for the fluorescence spectrophotometer are: excitation wavelength range 220-450 nm, emission wavelength range 250-550 nm, and scan speed 2000 nm / min.

[0013] Furthermore, in step two, the maximum fluorescence intensity value of each component is corrected using quinine sulfate.

[0014] Furthermore, in step three, the input data for the SOM simulation is standardized.

[0015] Furthermore, in step four, the established prediction model is cross-validated.

[0016] Furthermore, in step four, the number of cross-validation samples is 10% of the number of samples used to build the prediction model.

[0017] Compared with the prior art, the beneficial effects of the present invention are:

[0018] To address the current limitations in aerosol research, such as the cumbersome acquisition of EEM data and insufficient application and information extraction, which leads to inaccurate reflection of the chemical properties and particle size distribution of organic matter in aerosols, this invention addresses the issue of EEM detection and analysis of atmospheric particulate matter. It combines PARAFAC, SOM, and decision tree simulation to perform data mining on EEM data, demonstrating the visualization process of differences in fluorophore characteristics among different samples. This method enables rapid acquisition of aerosol chemical properties, particle size distribution, and deposition location information in the human respiratory tract without the need for other monitoring instruments. Furthermore, it avoids the need for filter membrane sample pretreatment and the use of chemical reagents, making the operation simple, fast, and economical, thus possessing significant potential for widespread application. This invention advances the comprehensive exploration of aerosol physicochemical properties and human health risks, as well as the application of EEM in aerosol research. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method for detecting the chemical properties and particle size distribution of organic matter in aerosols;

[0020] Figure 2 This is a fluorescence composition diagram obtained from PARAFAC analysis;

[0021] Figure 3 This is a visualization of the SOM simulation results based on PARAFAC components, including the U-matrix (a), Hit plot (b), sample mapping distribution (c), k-means clustering (d), and component plane (e). In Figure a, neurons are marked with dots; Figure b shows the number of samples in each neuron; in Figure c, numbers represent sample numbers, with 1-180 representing 20 groups of particle size-graded particulate matter samples arranged by sampling time. Numbers 1-9 represent the daytime sample group of the first sampling day, numbers 10-18 represent the nighttime sample group of the first sampling day, and numbers 19-27 represent the daytime sample group of the second sampling day. The samples are arranged in this manner; each group of samples is sorted from largest to smallest particle size, namely 9.0-10μm, 5.8-9.0μm, 4.7-5.8μm, 3.3-4.7μm, 2.1-3.3μm, 1.1-2.1μm, 0.65-1.1μm, 0.43-0.65μm, and <0.43μm; each neuron is numbered in order from top to bottom and from left to right. For example, neuron 1 contains samples 126, 153, and 162; neuron 2 has no samples; neuron 18 contains samples 132, 133, and 134; and neuron 19 contains samples 1, 27, 36, and 108.

[0022] Figure 4 It is a decision tree structure based on all samples of the PARAFAC component;

[0023] Figure 5 It is the decision tree structure of cluster I (a) and cluster II (b) samples after k-means clustering. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] This invention acquires aerosol excitation-emission spectroscopy (EEM) datasets and performs data mining on these EEM data to explore the chemical properties of aerosol organics, enabling rapid acquisition of information on atmospheric particulate matter size distribution and respiratory system deposition locations. First, EEM data of aerosols are tested. Second, parallel factor analysis (PARAFAC) is used to simulate the EEM data to obtain the optimal fluorescence components, and self-organizing map (SOM) simulations are used to interpret the chemical properties of atmospheric particulate matter organics. Third, based on the k-means clustering results of SOM simulations of size-graded particulate matter samples, the samples are grouped, with the fluorescence component ratio as the independent variable and the particulate matter size range as the dependent variable. A decision tree tool is used to establish a predictive model for aerosol deposition locations in the human respiratory system. Finally, the particle size of non-size-graded particulate matter samples is predicted based on the established model. This invention provides a scientific basis for revealing the environmental fate of aerosols and assessing the exposure risks to the human respiratory system.

[0026] The specific plan is as follows:

[0027] A method for detecting the chemical properties and particle size distribution of aerosol organic matter, such as... Figure 1 As shown, this method includes the following steps:

[0028] Step 1: Aerosol Sample Collection and 3D Fluorescence Spectroscopy Testing: To eliminate differences in the spatiotemporal distribution characteristics and environmental behavior of aerosol fluorophores in different atmospheric environments, sampling locations were selected based on the spatial distribution characteristics of different cities. Sampling times were determined by dividing the sampling locations into intervals according to the actual season. Atmospheric particulate matter was collected onto a filter membrane using a particle size fractionation collector. The filter membrane was fixed on the solid test stage of the fluorescence spectrophotometer, allowing direct testing of the 3D fluorescence spectra of the atmospheric particulate matter collected on the filter membrane.

[0029] Step 2, PARAFAC simulation: To avoid differences in PARAFAC analysis of sample datasets from different regions and seasons, PARAFAC analysis was performed on EEM datasets from different sampling points and seasons. The EEM was decomposed into several optimal components, and the fluorescence intensity values ​​of each component were obtained.

[0030] Step 3, SOM simulation: The fluorescence intensity values ​​of each component are used as input data for SOM simulation to obtain the unified distance matrix (U-matrix), sample mapping distribution, component plane and k-means clustering, and to visualize the chemical characteristics of fluorophores and the differences of fluorophores between samples;

[0031] Step 4: Establish an aerosol particle size distribution prediction model based on the EEM dataset of particle size graded particulate matter samples: Classify the samples according to the k-means clustering results of the SOM of particle size graded particulate matter samples, use the ratio of fluorescent components as the independent variable, and use the particle size range, i.e. the deposition location of particulate matter in the human respiratory tract, as the dependent variable, to conduct decision tree simulation and cross-validation, and establish a prediction model for aerosol particle size range and deposition location in the human respiratory tract.

[0032] Step 5: Prediction of aerosol particle size distribution and human respiratory tract deposition location for non-size-fed particulate matter samples: Based on the prediction model for aerosol particle size range and human respiratory tract deposition location established in Step 4, the aerosol particle size distribution and human respiratory tract deposition location for non-size-fed particulate matter samples are predicted.

[0033] Furthermore, in step one, the aerosol collection filter membrane is a glass fiber filter membrane or a quartz filter membrane. Before use, the filter membrane is wrapped in aluminum foil and pre-baked in a muffle furnace at 450℃ for 6-8 hours to remove contaminants. The baked filter membrane needs to be placed in a desiccator for 24-48 hours before use, with the desiccator conditions being a relative humidity of 30%-35% and a temperature of 19-25℃. The fluorescence spectrophotometer is set with excitation wavelength range of 220-450nm, emission wavelength range of 250-550nm, and a scan speed of 2000nm / min; at least one site blank sample is collected for each batch of sampling.

[0034] Furthermore, in step two, PARAFAC is performed to remove most of the Rayleigh and Raman scattering and correct the internal filtering effect. The maximum fluorescence intensity value of each component is corrected with quinine sulfate. The model is performed under non-negative constraints. The fluorescence intensity of the actual sample should be reduced by the fluorescence intensity of the site blank sample. To ensure the accuracy of the simulation results, the number of simulation samples is no less than 20 each time. The same PARAFAC analysis is performed on particle size fractionated particulate matter samples and non-particle size fractionated particulate matter samples to ensure the consistency of the obtained fluorescence components.

[0035] Furthermore, in step three, the SOM simulation: the fluorescence intensity values ​​of each component are used as input data for the SOM simulation. Data standardization is performed to avoid concentration effects, and the simulation yields a unified distance matrix (U-matrix), sample mapping distribution, component plane, and k-means clustering, visualizing the chemical characteristics of fluorophores and the differences in fluorophores between samples. Considering the complexity of the organic composition in atmospheric particulate matter, fluorescence components and fluorescence indices can only partially describe fluorophore characteristics. Therefore, this invention uses SOM, which is suitable for analyzing unknown and complex systems, to visualize the fluorophore characteristics of organic matter in atmospheric particulate matter. SOM can map EEM to two-dimensional space through nonlinear transformation, providing high visualization capabilities. Specifically, the fluorescence component intensities obtained from PARAFAC are used as input data for SOM, and the fluorescence intensity is standardized to reduce concentration effects. The simulation results include a unified distance matrix (U-matrix), sample mapping distribution, component plane, and k-means clustering. The U-matrix represents the differences in fluorescence characteristics among samples. The fluorescence difference between a neuron and its neighboring neurons is represented by the neuron's own color, which corresponds to the average Euclidean distance. Empty neurons indicate significant differences in fluorophore characteristics between neurons. Neurons that are far apart in the U-matrix typically exhibit large differences in fluorescence characteristics. The sample mapping distribution corresponds to the sample's position in the U-matrix. The component plane represents the distribution of each fluorescent component in the sample; similar component planes between different components indicate a strong correlation between them. K-means clustering is used to identify the similarity of fluorescence characteristics. Samples within the same cluster typically have similar fluorophore characteristics. Three fluorescence indices (HIX, FI, and BIX) are used to verify the differences in fluorescence characteristics between clusters. The formulas for calculating HIX, FI, and BIX are as follows:

[0036]

[0037]

[0038]

[0039] In the formula, F is the fluorescence intensity when Ex = i and Em = j;

[0040] An increase in the HIX value is associated with an increase in the condensation reaction of dissolved organic matter and is also used to represent the degree of humification or aromaticity of dissolved organic matter. Aging processes also lead to an increase in the HIX value. FI can be used to determine the source of dissolved organic matter precursors; FI values ​​of 1.2 and 1.8 represent terrestrial origin and microbial activity, respectively. BIX is related to the freshness of dissolved organic matter.

[0041] Furthermore, in step four, the samples are divided into several groups according to the k-means clustering results of the SOM of the particle size fraction samples for decision tree analysis. In each group, the ratio of fluorescent components is used as the independent variable, and independent variables with a denominator of zero are removed. The particle size range is used as the dependent variable. Based on the difference in the location of particles of different sizes entering the human respiratory system, the particle size range is set as head airway (HA) (>4.7μm), trachea and bronchi (TB) (1.1-4.7μm), and alveolar region (AR) (<1.1μm). The number of cross-validation samples is 10% of the number of samples used to build the prediction model.

[0042] Example 1

[0043] (1) Atmospheric particulate matter samples were collected from the rooftop of a building in Harbin City. The collection site was a mixture of commercial and residential areas. Particle size classification samples were collected using a particle size classification sampler. The particle size ranges collected for each stage were: Stage 0: 9.0-10 μm, Stage 1: 5.8-9.0 μm, Stage 2: 4.7-5.8 μm, Stage 3: 3.3-4.7 μm, Stage 4: 2.1-3.3 μm, Stage 5: 1.1-2.1 μm, Stage 6: 0.65-1.1 μm, Stage 7: 0.43-0.65 μm, and Stage F: <0.43 μm. The particle size range for particles that can be deposited in the HA (High Aperture Zone) during human respiration includes grades 0, 1, and 2; the particle size range for particles that can be deposited in the TB (Thickness Zone) includes grades 3, 4, and 5; and the particle size range for particles that can be deposited in the AR (Aperture Zone) includes grades 6, 7, and F. Furthermore, in this embodiment, particles with a diameter ≤2.1 μm and 2.1-10 μm are defined as fine particles and coarse particles, respectively. In this embodiment, particulate matter samples were collected continuously for 10 days, with two sets of samples collected daily: daytime and nighttime samples. The sampling duration for daytime and nighttime samples was allocated according to the winter diurnal pattern of the sampling point, with a daytime sampling duration of 8 hours and a nighttime sampling duration of 15 hours. A total of 20 sets of particle size-graded particulate matter samples were obtained, each set containing 9 glass filter membranes, totaling 180 samples. After each sampling, the sampler was disinfected with 75% alcohol, and the sampler was assembled and sampling was carried out only after the alcohol had completely evaporated. Ultrapure water was used to extract water-soluble substances from the filter membrane using ultrasonic extraction for three-dimensional fluorescence spectroscopy testing. The fluorescence spectrophotometer was set with excitation wavelength range of 220-450 nm, emission wavelength range of 250-550 nm, and scan speed of 2000 nm / min.

[0044] (2) The EEM data of the samples were processed using PARAFAC. No outliers were found during the processing. Figure 2 As shown, five optimal components were obtained.

[0045] (3) Considering the complexity of the fluorescence characteristics of water-soluble organic compounds (WSOCs) in atmospheric particulate matter, SOM was performed to visualize the fluorophore characteristics of WSOCs in atmospheric particulate matter. The fluorescence component intensity values ​​obtained from PARAFAC were used as input data for SOM. The simulation results included U-matrix, sample mapping distribution, component plane, and k-means clustering. Three fluorescence indices (HIX, FI, and BIX) of samples in clusters I and II were calculated to indicate the differences in the chemical characteristics of WSOCs between the two clusters. The results showed that there were significant differences in all three fluorescence indices between the two clusters (t-test, p<0.05), confirming the effectiveness of SOM in identifying differences in the chemical characteristics of fluorophores. The FI (1.95) and BIX (1.08) values ​​in cluster I were significantly higher than those in cluster II (FI: 1.82, BIX: 0.992), indicating that the WSOCs in cluster II mainly originated from microbial activity, while the WSOCs in cluster I were fresher than those in cluster II. The HIX level in cluster II (2.28) was significantly higher than that in cluster I (0.719), indicating that WSOC in cluster II had stronger aromaticity.

[0046] The differences in WSOC fluorophore distribution during particulate matter of different sizes and atmospheric pollution processes were investigated using SOM. Figure 3 As shown, the k-means clustering results indicate that the fluorophores are divided into two clusters (Cluster I and Cluster II). Cluster I consists entirely of fine particulate matter samples, while Cluster II is mainly composed of coarse particulate matter samples, accounting for more than 74%. The increasing legend values ​​in the U-matrix indicate an increase in fluorescence property differences; therefore, the differences between WSOC fluorophores in fine particles are greater than those in coarse particles. Furthermore, in Cluster I, we observed greater fluorescence differences among neurons in the lower left part. The samples in these neurons all came from severe air pollution weather (average AQI: 215). Given that Cluster I consists entirely of fine particulate matter samples, the influence of particle size was excluded, indicating that air pollution processes increased the differences in WSOC fluorescence characteristics within fine particulate matter.

[0047] The component planes reveal the correlations between components. Overall, the distribution characteristics of the five component planes are similar, all showing a trend of fluorescence intensity gradually increasing from top to bottom, indicating a certain degree of correlation among the components. The component plane values ​​of C1 and C3 gradually increase from the upper right to the lower left, indicating a strong correlation between C1 and C3. Correlation analysis of the fluorescence intensity of each component shows significant correlations, with C1 and C3 exhibiting the largest correlation coefficient, confirming the conclusions drawn from the component planes. In conclusion, PARAFAC-SOM is a powerful tool for further exploring the environmental behavior of organic aerosols. It can effectively characterize the fluorescence properties of dissolved organic matter in aerosols and the influence of particle size and air pollution processes on fluorophores.

[0048] (4) This example first simulates the particle size range of all samples using a decision tree, such as... Figure 4 As shown, the ratio of fluorescent components was the independent variable, and ratios with a denominator of zero were discarded. The particle size range was the dependent variable. Based on the differences in the entry points of particles of different sizes into the human respiratory system, the particle size ranges were set as HA (>4.7 μm), TB (1.1-4.7 μm), and AR (<1.1 μm), and 10% of the sample size was selected for cross-validation. The results showed that the prediction accuracies for HA, TB, and AR were 91.7%, 85.0%, and 95.0%, respectively. Secondly, to address the low prediction accuracy of decision trees for TB, we divided the samples into two groups, Cluster I and Cluster II, based on the k-means clustering results of SOM. Then, based on the ratio of fluorescent components in the two groups, we used decision trees to predict the particle size range. Figure 5 As shown, the results indicate that cluster I, which is dominated by coarse particulate matter samples, exhibits high accuracy in predicting the particle size of HA (88.3%) and TB (91.7%) particles, while cluster II, which is dominated by fine particulate matter, exhibits high accuracy in predicting the particle size of AR (100%) particles. This demonstrates that decision tree simulation after k-means clustering using SOM can significantly improve the accuracy of atmospheric particulate matter size prediction, thus confirming the effectiveness of this invention in predicting particulate matter size.

[0049] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for detecting the chemical properties and particle size distribution of aerosol organic matter, characterized in that, Includes the following steps: Step 1: Collect atmospheric particulate matter using a particle size fractionation sampler and obtain the three-dimensional fluorescence spectrum of the particle size fractionated atmospheric particulate matter using a fluorescence spectrophotometer. Step 2: Perform PARAFAC analysis on the three-dimensional fluorescence spectral dataset to decompose the dataset into several optimal components and obtain the fluorescence intensity values ​​of each component. Step 3: Use the fluorescence intensity values ​​of each component as input data for SOM simulation to obtain a unified distance matrix, sample mapping distribution, component plane and k-means clustering, and visualize the chemical characteristics of fluorophores and the differences of fluorophores between samples; Step 4: Based on the k-means clustering results of atmospheric particulate matter SOMs classified by particle size, the samples are divided into several groups for decision tree analysis. The ratio of fluorescent components in each group is used as the independent variable, and the particle size range, i.e. the deposition location of particulate matter in the human respiratory tract, is used as the dependent variable. A prediction model for aerosol particle size range and deposition location in the human respiratory tract is established.

2. The method for detecting the chemical properties and particle size distribution of aerosol organic matter according to claim 1, characterized in that: Based on the established prediction model of aerosol particle size range and human respiratory tract deposition location, the aerosol particle size distribution and human respiratory tract deposition location of non-size fractionated particulate matter samples are predicted.

3. The method for detecting the chemical properties and particle size distribution of aerosol organic matter according to claim 1, characterized in that: In step one, the parameters set for the fluorescence spectrophotometer are: excitation wavelength range 220-450 nm, emission wavelength range 250-550 nm, and scan speed 2000 nm / min.

4. The method for detecting the chemical properties and particle size distribution of aerosol organic matter according to claim 1, characterized in that: In step two, the maximum fluorescence intensity value of each component is corrected using quinine sulfate.

5. The method for detecting the chemical properties and particle size distribution of aerosol organic matter according to claim 1, characterized in that: In step three, the input data for the SOM simulation is standardized.

6. The method for detecting the chemical properties and particle size distribution of aerosol organic matter according to claim 1, characterized in that: In step four, the established prediction model is cross-validated.

7. The method for detecting the chemical properties and particle size distribution of aerosol organic matter according to claim 6, characterized in that: In step four, the number of cross-validation samples is 10% of the number of samples used to build the prediction model.

Images