The data for six conventional pollutants (PM_2.5, PM₁₀, sulfur (SO₂), nitrogen dioxide (NO₂), carbon monoxide (CO), ozone (O₃)) as well as meteorological parameters such as temperature, humidity, and wind speed were obtained from 10 national air quality monitoring stations. PM_2.5 and PM₁₀ are monitored using a particulate matter monitor (5030i, Thermo Fisher Scientific, Waltham, MA, USA), which utilizes a dual-method hybrid approach to measure PM_2.5 and PM₁₀, combining β-ray absorption (beta attenuation) and light scattering (nephelometry) technologies for high-precision, real-time monitoring. According to the relevant standards issued by the Ministry of Ecology and Environment of China [43,44], PM_2.5 and PM₁₀ are maintained and quality controlled, and data are reviewed by the relevant governmental departments to ensure the accuracy of the data. The chemical composition of PM_2.5, including elemental carbon (EC), organic matter (OM, which is calculated by organic carbon (OC) multiplied by 1.6 [45]), calcium ions (Ca²⁺), magnesium ions (Mg²⁺), potassium ions (K⁺), ammonium ions (NH₄⁺), sodium ions (Na⁺), sulfate ions (SO₄²⁻), nitrate ions (NO₃⁻), chloride ions (Cl⁻), and fluoride ions (F⁻), was observed at the South Gate site—the only national particulate-matter component monitoring site in Suzhou (Figure 1). The PM_2.5 data characteristics of the 10 national air quality monitoring stations and the South Gate site are consistent. The online measurement of water-soluble ions in PM_2.5 is conducted using online ion chromatography (2060 MARGA, Metrohm, Herisau, Switzerland). The concentrations of OC and EC in PM_2.5 are measured using the OC/EC online monitoring analyzer (RT4, Sunset Lab, Tigard, OR, USA), with a time resolution of 1 h. To ensure the accuracy of the data, regular checks on instrument flow rates, the replacement of accessories, air-tightness tests, and backups of original data are performed. The data were reviewed and audited by the National Environmental Monitoring Center and assessed in accordance with standards issued by the Ministry of Ecology and Environment of China, ensuring reliability and accuracy to analyze the effects of anthropogenic emissions and meteorological factors on PM_2.5 concentrations.

Electricity consumption data, used to evaluate the activity level in Suzhou, were sourced from the Suzhou Statistical Yearbook, published by the Suzhou Bureau of Statistics. Additionally, emissions data for key air-polluting enterprises, including sulfur dioxide, nitrogen oxides, and particulate matter, were obtained from automatically monitored data linked to the national network, with a valid data transmission rate of over 98%.

In this study, a random forest model [46,47,48,49] was used to predict pollutant concentrations based on time and meteorological variables, and to differentiate the relative contributions of meteorology and emissions to pollutant levels The random forest model is a widely used machine learning model. Specifically, the time variables include the Unix time, year, week, and hour, representing the long-term, annual, weekly, and daily variation characteristics of pollutant emissions. The meteorological variables used in the random forest model include the wind speed (WS, calculated from U10 and V10), wind direction (WD, calculated from U10 and V10), temperature (T), relative humidity (RH, calculated from T and the dewpoint temperature), surface solar radiation (SSR), surface pressure (SP), total precipitation (TP), boundary layer height (BLH), total cloud cover (TCC), length of each air mass trajectory (Length), and cluster of each air mass trajectory (Cluster), to fully account for the impact of meteorology on pollutants. The meteorological parameters above were extracted from the Global Reanalysis Data published by NOAA (National Oceanic and Atmospheric Administration), while Length and Cluster were generated by the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model.

A “random sampling–model prediction–concentration averaging” method was adopted to standardize the meteorological conditions and obtain the pollutant concentration changes unaffected by meteorology. First, a random forest model was developed to predict pollutant concentrations. Then, while keeping the Unix time column unchanged, historical meteorological data were randomly sampled for each data point to form a new dataset with shuffled sequences. This new dataset was fed into the random forest model to predict a time series of pollutant concentrations. This “random sampling–model prediction” process was repeated 1000 times, and the 1000 prediction results were averaged to obtain pollutant concentrations considering almost all meteorological conditions. These meteorologically standardized pollutant concentrations reflected changes solely due to emissions.

The WRF–CMAQ model was employed in this study to analyze the contributions of pollutant transport to Suzhou’s PM_2.5 levels on June 13–23 2023, as well as to conduct source apportionment for PM_2.5.

The WRF physical process and parameterization scheme are shown in Table 1. The WRF–CMAQ model considers chemical mechanisms such as the gas phase, liquid phase, and heterogeneous phase, so the calculation results of PM_2.5 concentrations in the model have included the influence of the generation of secondary particles such as nitrate. The 2 m temperature (T₂), 10 m wind speed (WS₁₀), and 2 m relative humidity (RH) simulated by the WRF model were compared with surface observation data to evaluate the accuracy of the meteorological field simulation. The performance of the model was assessed using statistical indicators such as Normalized Mean Bias (NMB), correlation coefficient (R), Index of Agreement (IOA), and root mean square error (RMSE) (Table 2). According to the evaluation results, the temperature simulation performed the best, with an R value of 0.84 and a RMSE of 2.7, although the model exhibited a slight overestimation overall. The RMSE of wind speed in the Suzhou area during this period, as calculated, was 2.0 m/s, indicating a certain degree of overestimation by the model. Regarding humidity, despite a relatively high RMSE, the correlation remained robust (R = 0.79) (detailed information is provided in the SI Figures S1 and S2).

This study used the three-level nested CMAQv5.3.2 model to simulate air quality across the Yangtze River Delta region [50,51]. The outermost domain (D01) had a horizontal resolution of 36 km, covering all of China, with 174 × 136 grid points and a central longitude–latitude of (110° E, 34° N). The second domain (D02) had a horizontal resolution of 12 km, covering eastern China, with 135 × 228 grid points. The innermost domain (D03) had a horizontal resolution of 4 km, with 192 × 216 grid points, covering the entire Yangtze River Delta region. The projection was consistent with the WRF model, using the Lambert Conformal Conic projection. Initial and boundary conditions for D01 were derived from the default CMAQ configuration, while the boundary conditions for D02 and D03 were obtained from the simulation results of D01 and D02, respectively. Since the initial conditions for all three layers used default configuration files, the simulation was run with a 5-day spin-up period to reduce the impact of initial conditions. The Multi-resolution Emission Inventory for China (MEIC inventory, Tsinghua University, Beijing, China), developed by Tsinghua University, was used for the 36 km, 12 km, and 4 km grids (except for the three provinces and one city in the Yangtze River Delta). The emission inventory for Jiangsu province (JS), Zhejiang province (ZJ), Anhui province, and Shanghai was updated using the 2018 city statistical yearbooks and 2017 environmental statistics. Natural VOC emissions were calculated using the Model of Emissions of Gases and Aerosols from Nature version 3.0 (MEGAN3.0 model) (Figure 2).

The model’s performance was evaluated by comparing observed PM_2.5 concentrations in Suzhou during 13–23 June 2023, with those simulated by the CMAQ model. Statistical indicators used for the evaluation included the Normalized Mean Bias (NMB) and the Index of Agreement (IOA). According to the evaluation results, both the NMB and IOA values met the recommended benchmarks proposed by Huang et al., 2021 [52], indicating that the model results are reliable and capable of accurately simulating the distribution of PM_2.5 concentrations during this period.

Meteorological conditions and pollutant emissions are the primary factors influencing PM_2.5 concentrations, with varying contributions depending on the time and location. The random forest model can analyze meteorological and pollution source contributions in a short amount of time through dynamic data learning. Therefore, this study quantified the contribution of meteorology and emissions to PM_2.5 in Suzhou from June 2019 to 2023 using the random forest model based on meteorological data and PM_2.5 data in Suzhou every June from 2019 to 2023. The specific method is detailed in the Materials and Methods section.

In comparison to the same period in 2022, June 2023 saw a 2% increase in the average humidity in Suzhou, a 0.5 m/s decrease in wind speed, a 1.3 °C drop in temperature, a 2 hPa increase in the average atmospheric pressure, and a 15.5% increase in rainfall (Table 3). These conditions were more stable and less favorable for the dispersion of pollutants compared to 2022.

The results indicated that meteorological conditions in June 2023 were more adverse for pollution dispersion than those of June 2022, while the contribution of emissions to PM_2.5 increased significantly in June 2023 compared to the same period in 2022. From the results of the random forest model, the meteorological and emission factors contributed to PM_2.5 concentration increases of 0.7 µg/m³ and 2.0 µg/m³, respectively, accounting for 26% and 74% of the total increase. Emissions were the dominant factor in the rise of PM_2.5 in Suzhou in June 2023.

The concentration of air pollutants is influenced by both local emissions and transboundary transport. Quantifying the contribution of external sources to Suzhou’s air quality, in addition to local emissions, is crucial for guiding regional joint pollution prevention and control.

The WRF–CMAQ model was employed to analyze the sources of pollution during the high PM_2.5 period from 13 to 23 June 2023. Local sources contributed 46.3% to Suzhou’s PM_2.5, while contributions from Shanghai, Jiaxing, and Huzhou contributed 12.5%, 7.3%, and 1.9%, respectively, to the PM_2.5 concentration in Suzhou. Other cities in Anhui, Jiangsu, and Zhejiang provinces also had significant impacts on Suzhou. The results show that local sources contributed the most to the PM_2.5 concentrations in Suzhou during this pollution event, indicating that local anthropogenic emissions play a decisive role in this pollution process. At the same time, the contribution of exogenous transmission mainly comes from southern and southeastern cities.

In detail, we investigated the hourly contribution ratio of PM_2.5 in Suzhou from 13 to 23 June 2023 (see Figure 6 and Figure 7). The local source contribution of PM_2.5 concentrations in Suzhou has always been at a high level throughout the whole pollution process. During the increasing PM_2.5 episode, local contributions in Suzhou rose significantly, especially at night when wind speeds decreased, leading to a stable atmosphere. During the late night of 22 June and early morning of 23 June, local sources accounted for over 60% of the PM_2.5 increase, especially highlighting the importance of controlling local emissions. From the transmission of surrounding cities, on 17 and 18 June, the prevailing southeast winds brought pollution from Shanghai, while on 23 June, southerly winds transported pollutants from Jiaxing. The above result suggests that strengthening coordinated regional pollution control, particularly from cities upwind of Suzhou, is essential for reducing transboundary impacts to PM_2.5 concentrations in Suzhou. However, it should be noted that the WRF model overestimated the wind speeds during 18 June to 20 June, which might lead to underestimated contributions from local emissions.

Moreover, based on these findings, we applied the WRF–CMAQ air quality model to analyze the contribution of local emission sources to PM_2.5 in Suzhou during the period from 13 June to 23 June 2023 (Figure 8), where the ‘other’ sources encompass the contributions of agriculture, biomass burning, and residential activities to PM_2.5. Overall, dust sources contributed the most to PM_2.5 concentrations, accounting for 39.9% (ranging from 31.9% to 50.0%), followed by industrial sources, which contributed 30.4% (ranging from 21.3% to 38.4%). The day-by-day source apportionment results show that during the pollution period (21–22 June), compared to the clean period (17–19 June), the contribution rates of dust sources, mobile sources, and industrial boilers and kilns increased significantly. This indicates that dust sources, mobile sources, and industrial boilers and kilns were the dominant contributors during the high PM_2.5 period. The source of dust was dominant during the high-concentration period of PM_2.5, which was consistent with the results of a significant increase in calcium ion concentration during the pollution period in the previous analysis of the chemical composition of PM_2.5.

To further verify the significance of the impact of local industrial pollution emissions, we conducted an electricity consumption analysis of the whole society and pollutant emissions of key polluting enterprises in Suzhou. The electricity consumption of the whole society includes electricity consumption in primary industries, secondary industries, and tertiary industries, and the domestic electricity consumption of urban and rural residents. Firstly, the electricity consumption data were used to evaluate Suzhou’s activity level, showing a clear increase in social activity following the end of the Coronavirus Disease 2019 (COVID-19) pandemic. This activity level is reflected in and total electricity consumption, which directly correlates with pollutant emissions. As is shown in Figure 9a, from February to May 2023, the total electricity consumption in Suzhou increased significantly compared to the same period in the previous year, with year-on-year increases ranging from 1% to 21%, and slightly decreasing in June. In June 2023, the total electricity consumption in Suzhou reached 14.5 billion kilowatt-hours, the highest since 2023. As is illustrated in Figure 9b, the monthly emissions of pollutants from key polluting enterprises in Suzhou followed a pattern similar to the electricity consumption. In January 2023, NOx and SO₂ emissions decreased year on year compared to January 2022 due to the impact of the COVID-19 pandemic. From February to June 2023, NOx and SO₂ emissions rose year on year compared with the same period in 2022 as the economy recovered. In particular, in June 2023, the NOx and SO₂ emissions of key polluting enterprises in Suzhou reached 58 t and 22 t, with both being the highest since 2023. To sum up, both the electricity consumption of the whole society and pollutant discharge of key polluting enterprises in Suzhou are at a high level, and local emissions have a significant contribution to the concentration of PM_2.5 in Suzhou, so the local industrial emission control should be strengthened.