An evaluated aerosol extinction coefficient dataset and its application to improve visibility forecasts in Xiong’an, China

With growing urbanization and industrialization in China, air pollution has become a major environmental issue affecting visibility across the country, especially in rapidly developing urban areas (Guo et al., 2014;Yue et al., 2017). Research demonstrates that particulate matter (PM2.5, PM10), aerosols, and other pollutants significantly reduce visibility by scattering and absorbing light, thereby causing severe impacts on public health, traffic safety, and economic activities (De Marco et al., 2019;Hao et al., 2021;Su et al., 2020). To quantify the influence of various factors on visibility, the atmospheric extinction coefficient is introduced; it represents the cumulative effect of multiple lightextinguishing substances in the atmosphere across visible and near-infrared wavelengths. Particularly during hazy conditions, aerosol extinction becomes a primary factor affecting visibility (Ting et al., 2022;Hu et al., 2017). The aerosol extinction coefficient (AEC), which is governed by particulate mass concentration, chemical composition, and hygroscopicity, constitutes a major source of uncertainty in visibility forecasting (Bhattacharjee et al., 2023).The primary technical methods for calculating the aerosol extinction coefficient include the physics-based Mie scattering theory, the statistics-based empirical formula method, and the remote sensing-based retrieval method, each possessing distinct advantages and application limitations. Mie scattering theory (Mie, 1908) calculates scattering and absorption coefficients based on aerosol particle size distribution, refractive index, and wavelength, forming the theoretical foundation for AEC computation. This method assumes aerosols are spherical particles and enables precise calculation of extinction efficiency by integrating particle size distribution with chemical composition (Ohta et al., 1990;Sumlin et al., 2018). Han et al. (2013) utilized a Mie scattering model within the Regional Atmospheric Modeling System-Community Multiscale Air Quality (RAMS-CMAQ) modeling system to simulate visibility during pollution episodes in Beijing and the North China Plain. Their results indicated that the low visibility events in this region were primarily caused by high concentrations of PM2.5, resulting from the accumulation of local pollutant emissions and long-range transport. In related studies, Mie scattering theory is also frequently employed to analyze the extinction characteristics of PM2.5. For instance, the hygroscopic growth of ammonium sulfate and ammonium nitrate under high-humidity conditions significantly enhances AEC (Gao et al., 2021). However, this method requires detailed particle size spectra and chemical composition data, which are difficult to obtain, and its high computational complexity constrains its application in real-time forecasting (Zhan et al., 2025;García-Arroyo and Osca, 2017).Secondly, the empirical formula method establishes a functional relationship between AEC and variables such as PM2.5 concentration and humidity through statistical regression, and it is widely used in air quality and visibility models. For example, Liu et al. (2022) (Hand and Malm, 2007;Malm et al., 1994). This program has become the most widely used empirical method for estimating the extinction coefficient. The IMPROVE empirical method estimates AEC based on PM2.5 chemical components such as sulfate, nitrate, and organic carbon (Pitchford et al., 2007). This algorithm incorporates the hygroscopic growth factor (f(Relative Humidity, RH)), effectively characterizing the influence of humidity on extinction (Zhao et al., 2019). For instance, researchers applying the IMPROVE algorithm to assess visibility impairment in China's Beijing-Tianjin-Hebei and Yangtze River Delta regions found that ammonium sulfate and organic matter contributed approximately 40% and 24%, respectively, to the extinction coefficient (Gao et al., 2021). However, this method relies on region-specific chemical composition data, exhibits poor versatility, and responds inadequately to dynamic changes in PM2.5 composition.Furthermore, satellite and radar retrieval methods are often used for AEC calculation, utilizing remote sensing data such as those from the Cloud-Aerosol Lidar (CALIPSO) and Moderate Resolution Imaging Spectroradiometer (MODIS). CALIOP data provide the aerosol vertical distribution, which, combined with MODIS Aerosol Optical Depth (AOD), enables the estimation of surface AEC via a two-layer model (He et al., 2012). Research in Shanghai, China, demonstrated that the AEC estimated using the two-layer model using combined CALIOP and MODIS data was highly correlated with observed visibility, with a correlation coefficient of 0.86 (He et al., 2012). However, satellite retrieval is often limited by cloud contamination and relatively coarse spatial resolution, which restricts its ability to meet the demands of numerical forecasting.The atmospheric extinction coefficient is a core physical quantity that quantifies the interaction between light and the atmosphere. As a key input for numerical simulations of visibility, it supports the analysis of aerosol physical and chemical properties, research on atmospheric radiative transfer, and assessment of climate effects, thereby forming a scientific bridge connecting fundamental research with practical needs such as pollution control and climate policy formulation. China's Xiong'an New Area, officially established in 2017 as a national-level new district, is designed to relieve Beijing of functions non-essential to its role as the capital and promote the coordinated regional development of the Beijing-Tianjin-Hebei area (Liao and Huang, 2020). As shown in Supplementary Material Figure S1, this region is characterized by unique geographical attributes: nestled against the eastern slopes of the Taihang Mountains in the west, adjacent to the North China Plain in the east, fronting the Bohai Bay, and cradling the renowned Baiyang Lake Wetland, often called the "Pearl of North China" Xiong'an has a warm-temperate monsoon continental climate. Situated at the lower reach of nine rivers, the area is jointly influenced by its complex terrain and the northern fringe of the East Asian monsoon belt, with prevailing northeasterly-tosouthwesterly airflow throughout the year. This makes the climate highly sensitive and variable, and prone to meteorological hazards such as low-visibility events. This specific geographical and climatic context, compounded by multifaceted pressures such as persistently high PM2.5 concentrations, pronounced humidity variability, and concentrated local emissions (Zhang et al., 2023), collectively establishes the region as a prototypical area for investigating atmospheric extinction characteristics, visibility evolution, and aerosol effects amid rapid urbanization.This study focuses on the Chinese Xiong'an New Area as a case study. By establishing a multi-source data fusion approach integrated with numerical simulation techniques, we have constructed a comprehensive dataset of aerosol extinction coefficients. This effort compensates for the limitations inherent in relying solely on the IMPROVE empirical formula or satellite retrieval methods for constructing extinction coefficient datasets, thereby providing a novel methodology for related fields. The dataset enabled regional visibility simulations using the China Meteorological Administration Mesoscale model (CMA-MESO), the outcomes of which were rigorously evaluated against ground-based observational data. The dataset encompasses an optimized aerosol extinction coefficient series and visibility forecast products. Its construction methodology is not only applicable to visibility simulation and forecasting evaluation in urban environments but also provides scientific data support and practical reference for regional air quality management, sustainable urban planning, atmospheric radiation transfer modeling, and climate impact analysis.This research is dedicated to constructing an aerosol extinction coefficient dataset through a multi-source data fusion approach, primarily to serve visibility prediction in mesoscale numerical weather modeling. This study is based on the CMA-MESO model, a mesoscale numerical weather prediction system independently developed by the China Meteorological Administration (CMA) which has been under continuous development and refinement since 2001 (Chen et al., 2008;Zhuang et al., 2023;Jishan, 2004). The system operates on the principles of mesoscale meteorological dynamics, utilizing high-resolution numerical simulation and multi-source data assimilation to generate refined weather element forecasts. It provides critical support for both daily meteorological operations and the safeguarding of major public events (Chen and Shen, 2006;Huang et al., 2017).In this study, two methods were applied in the CMA_MESO model to simulate visibility data in China, which were labelled as S1 and S2.a. The S1 scheme S1 is a set of algorithms based on historical station observation data. Wang et al. (2020) found the PM2.5 concentration showed a good linear relationship with the extinction coefficients at different relative humidity in most parts of China, according to observation data. Especially in Beijing, the correlation coefficients between extinction coefficient and PM2.5 concentration always fluctuate around 0.90, with relative humidity ranging from 20% to 90%. The relationship equations between PM2.5 concentration and extinction coefficient are given by equation (1).□□ □□□□□□ = □□(□□□□) × □□□□ 2.5 + □□ (1)Where □□□□□□□□ is the extinction coefficient, □□ -1 ; □□(□□□□) is the extinction efficiency of the aerosol produced by PM2.5 under different relative humidity; PM2.5 is the fine particulate matter concentration, μg/m 3 . b is the additional extinction coefficient arising from Rayleigh scattering and gas absorption, □□ -1 .The value of (□□□□) and b were referred to Wang et al. (2020). Based on hourly observational data of PM2.5 and relative humidity from 2018 to 2023, the aerosol extinction coefficient in Xiong'an was calculated using Formula (1). Considering the significant differences in particulate pollution characteristics across seasons and distinct diurnal variation patterns, the hourly data for each respective month over the five-year period were averaged, forming the basis for the extinction coefficient dataset under the S1 scheme. The formula for calculating the average value is as follows:□□ ̅ (□□,ℎ) = ∑ □□ □□,□□,ℎ 5 □□=1 5 (2)Where y stands for year, m for month, and h for hour. □□ ̅ (□□,ℎ) represents the final required average extinction coefficient for the m-th month and h-th hour, while □□ □□,□□,ℎ denotes the original extinction coefficient data for the h-th hour of the m-th month in the y-th year.b. The S2 scheme S2 is generated from S1 through the integration of satellite aerosol extinction data.This approach employs the Space-Time Multiscale Analysis System (STMAS), an advanced data assimilation system developed by the Earth System Research Laboratory (ESRL) of the National Oceanic and Atmospheric Administration (NOAA), to merge the baseline data produced by Formula (1) with the MERRA-2 reanalysis product, thereby creating the new dataset. MERRA-2 provides globally covered reanalysis data and significantly improves the spatiotemporal consistency and accuracy of atmospheric variables (such as aerosols) by assimilating multi-source satellite observations. The core algorithm of STMAS is the multigrid sequential variational method, which performs iterative analysis from coarse to fine grids. This method utilizes coarse grids to accelerate the convergence of large-scale, low-frequency patterns, thereby providing an initial field for the fine-grid analysis. This process effectively eliminates aliasing effects across different scales, ultimately producing a fused product with high spatiotemporal resolution (Xie et al., 2011).The objective function on each grid takes the form:J (□□) = 1 2 □□ (□□)□□ □□ (□□) + 1 2(□□ (□□) □□ (□□) -□□ (□□) ) □□ □□ (□□)-1 (□□ (□□) □□ (□□) -□□ (□□) ) (□□ = 1,2,3, □□, □□)The Y value on each grid is calculated as follows：{ □□ (1) = □□ (□□□□□□) -□□ (1) □□ □□ (□□ = 1) □□ (□□) = □□ (□□-1) -□□ (□□-1) □□ (□□-1) (□□ = 2,3, □□, □□)(4)The final analysis result is the superposition of the results of each heavy grid analysis:□□ □□ = □□ □□ + □□ □□ = □□ □□ + ∑ □□ (□□) □□ □□=1(5)Among them Y = □□ □□□□□□ -□□□□ □□ , X = □□ □□ -□□ □□ . O is the observation error covariance matrix; □□ □□ , □□ □□ □□□□□□ □□ □□□□□□ are the background field, analysis field, and observation field vectors, respectively; H is the bilinear interpolation operator from the model grid to the observation point; X represents the correction vector relative to the model field vector, calculated by the variational data assimilation system; Y is the difference between the observation field and the model field; n represents the nth grid, and N is the total number of grids. The element data is fused according to the multi-grid variational analysis method to generate a fusion product.The development of the S2 extinction coefficient dataset involved transforming the S1-derived data into a 3-km resolution grid through the application of the Cressman (Cressman, 1959) interpolation method, which was selected for its demonstrated advantages in spatiotemporal distribution accuracy. The Cressman objective analysis method employs a successive correction technique to interpolate data from discrete observation stations onto regular grid points. The specific procedure is as follows: First, an initial guess field for the grid points is established, which in this study is defined as the regional mean within the scanning radius R. Subsequently, observational data within this radius are used to correct the initial guess field by calculating the analysis increment(Δ□□ □□□□ )between the observed values and the initial field. This correction process is repeated iteratively until the adjusted field converges satisfactorily with the observations. In practice, a single objective analysis typically requires configuring the configuration of multiple scanning radii R and several iterations of the Cressman method to achieve optimal results. The formula is expressed as follows: □□ □□□□ □□ =Where R represents the scanning radius, □□ □□□□,□□ denotes the distance between the kth observation within the scanning radius and the grid point (i,j), □□ □□□□,□□ is the influence weight of the k-th observation on grid point (i,j), □□ □□ represents the k-th observational value within the scanning radius, □□ □□□□ is the value of the initial guess field, and □□ □□□□ □□ denotes the Cressman analysis value after the (n-1)-th iteration. □□ □□ represents the Cressman analysis weight for scanning radius R, where the dimensions of V and R are identical. □□ □□□□ (□□,□□) is the Cressman analysis result for scanning radius R, while □□ □□□□ (□□,□□-1) corresponds to the analysis result for scanning radius R-1.The 3-km gridded data, obtained through the aforementioned interpolation method, were subsequently fused with the MERRA-2 reanalysis aerosol extinction data using the data assimilation approach described in Equation ( 3). This procedure yielded the integrated grid data that form the S2 extinction coefficient dataset.The aerosol extinction coefficients calculated by the S1 and S2 schemes will be converted into visibility via Koschmieder's law for the purpose of evaluating the dataset quality and its application effectiveness.3.912 □□ × ln 10 (10)To verify the accuracy of the extinction coefficient, this study assessed its impact on visibility forecasting. The performance of the two calculation schemes (S1 and S2)was evaluated through statistical and graded verification tests, following the methodology detailed in Table S1 (see Supplkmentary Material) (Jongeward et al., 2016).The statistical test metrics comprise the mean error (ME), the standardized mean bias (NMB), and the root-mean-square error (RMSE). Their respective formulas are as follows:□□□□ = ∑(□□ □□ -□□ □□ ) □□ □□=1 □□ ⁄ (10) □□□□□□ = ∑(□□ □□ -□□ □□ ) □□ □□=1 ∑ □□ □□ □□ □□=1 ⁄ × 100% (11) □□□□□□□□ = √ 1 □□ ∑(□□ □□ □□ □□=1 -□□ □□ ) 2 (12)Where i represents the values for each hour, □□□□ is the i-th forecast value of the scheme, □□□□ is the i-th observed value, and □□ is the total number of samples.For the grading test, the Threat Score (TS) was computed using the three visibility classes in Table S1. The parameters were calculated as follows.Where □□□□ is the number of correct forecasts, □□□□ is the number of empty reports, and □□□□ is the number of missed reports. A higher TS signifies superior forecast quality.The surface observation data used in this case study include PM2.5 concentration, relative humidity, and visibility. The pollution data were obtained from the Environmental Meteorological Service Platform (http://eia-data.com/), and the meteorological data are from the Xiaozuanfeng platform (https://meteo.agrodigits.com/), a publicly accessible platform providing meteorological observations for research applications. All observational data were collected from stations in the Xiong'an area. The case study focused on the Xiong'an region and utilized data from these observation station covering the period from December 1, 2018, to December 1, 2023, covering a five-year period.The MERRA-2 (Modern Era Retrospective-Analysis for Research, version 2) (https://disc.gsfc.nasa.gov/datasets/M2TMNXAER_5.12.4/summary) is a long-term reanalysis product that includes various meteorological variables. In this study, we used the monthly average aerosol extinction coefficient at 550 nm data from the MERRA-2 reanalysis dataset, covering the date range December 1, 2018, to December 1, 2023.The outputs of this study have been systematically archived as the following four 3 Technical ValidationTo validate the dataset constructed using Schemes S1 and S2, we assessed its impact on visibility forecasting for the period from September to November 2023.Supplementary Figure S2 and As shown in Table 2 and Supplementary Figure S3, the S2 scheme achieves higher TS values than S1 across most visibility intervals, demonstrating its improved capability in simulating aerosol effects on visibility. In the 1-3 km interval, the S2 scheme achieves higher TS values (0.0440-0.0663) than S1 (0.0048-0.0295). The peak TS for S2 in September (0.0663) coincides with an observed PM2.5 concentration of approximately 100 μg/m³, underscoring its accuracy in capturing haze episodes. For the 3-5 km interval, S2 performs slightly better than S1 in September (TS =0.0373) and October (TS=0.0697). Similarly, in the higher visibility range (5-10 km), S2 outperforms S1 in September (0.1673) and November (0.1730). The consistently higher TS values of S2 across most ranges confirm its enhanced ability to simulate the aerosol effect. In Evaluation of visibility forecasts in the Xiong'an area shows that S2 outperforms S1, with ME ranging from -6.823 km to 1.360 km and RMSE between 4.290 km and 10.981 km. Moreover, S2 exhibits a superior ability to capture haze events, as indicated by its higher TS in the visibility range of 1-3 km.This study provides a methodological framework and a validated dataset that offer a scientific basis for air quality management and visibility forecasting in rapidly urbanizing regions.