The atmospheric diffusion tracer experiment data used in the study were collected in the summer of 2022 in the Lianyungang area. The experimental area was generally flat, with the terrain having a relatively weak effect on the wind field. The overall terrain is depicted in Fig. 1, with the release point situated at the center. The elevation difference in terrain does not exceed 100 m, with the (+) indicating the position of the sampling points. Based on the statistical analysis of wind characteristics in the experimental area, sampling points were eventually established at various distances within 10 km to the west of the selected release location. The tracer gas used in the experiment is sulfur hexafluoride (SF₆), released at a height of 30 m and for a duration of 1 hour each time. Sampling is performed at all indicated sampling points in Fig. 1. After release, the arrival time of the SF₆ plume is estimated based on the current wind speed and the distance from each sampling point to the release point. Sampling commences 10 minutes after the estimated arrival time.

The experiment utilized the TWA-300 automatic electronic time-controlled atmospheric sampler manufactured by Shanghai Xingzhuoan Environmental Protection Instrument Co., Ltd. The instrument is equipped with a low-power steady flow pump, flow regulating valve, buffer, filter, and electronic time-controlled circuit. The flowmeter is controlled between 20 mL/min and 500 mL/min, which is particularly suitable for small flow timing sampling. The ambient temperature used for the instrument is 0–45 °C, and the relative humidity is ≤90%. The instrument employs a long-life air pump to extract gas from the site, with the required flow rate being adjusted via an adjustable three-way valve. The timing can be set by a time timer to pump the gas to be collected into the air bag at a fixed time. The effective sampling rate ranges from 0.05 L/min to 0.3 L/min, with a maximum range of 0.50 L/min. The sample bag is made of aluminum foil, possessing high airtightness and low absorption characteristics. The collected samples are analyzed by gas chromatography, which uses 5A molecular sieve packed column to separate SF₆, and is detected by gas chromatography electron capture detector. The concentration is determined according to the peak area. The information on sampling processes is listed in Table 1.

In this study, the meteorological data utilized are the meteorological tower data at the release point. This simulation uses 10 minutes meteorological data to drive diffusion model. The meteorological data used are as follows: wind direction, wind speed, temperature data at each layer of the meteorological tower and the data measured at 10 m height of other locations.

A three-dimensional objective diagnostic wind field model was used in the paper, which is based on principle of mass conservation. The simulated area of the diagnostic wind field is centered around the release point, extending 20 km in both the west-east and north-south directions. The simulation area is divided into 201×201 grids, each with a resolution of 100 m. Vertically, there are 20 layers, with the first layer being 10 m thick, and subsequent layers being 50 m thick.

Four sampling processes were simulated with a release height of 30 m and a release quantity of 24 kg. The wind speed was relatively high, and the atmospheric stability during the release period was slightly unstable. The first and second release processes experienced northeasterly winds, while the third and fourth release processes transitioned from northeast to easterly winds.

Fig. 2 displays the simulation results of the NACADOS diffusion model, including the distribution of simulated wind fields, concentration fields, and sampling points with concentrations above 0 during each release process. It can be seen that the distribution range of simulated plume is consistent with the distribution of sampling points with concentration values higher than 0 in actual sampling; that is, the range of simulated smoke clouds is consistent with the range of actual plume distribution.

The simulated concentration field does not appear to exhibit a Gaussian distribution. There are two reasons. Firstly, the wind direction varied over time, causing changes in the direction of the concentration fields. The direction of the concentration fields reflects a specific moment, and the fields may appear curved at certain points. Secondly, the particle model was utilized to simulate the diffusion of SF₆ in this paper, with the particle distribution characterizing the concentration distribution. Due to the relatively discrete particle distribution, the edges of the concentration field may appear different from those of a Gaussian distribution.

The study refers to indicators used in the effectiveness analysis of the DIsPersion over COmplex Terrain (DIPCOT) diffusion model [2] in the European Nuclear Accident Consequence Assessment System (JRODOS), and other indicators including: mean error (BIAS), mean absolute error (MAE), root mean square error (RMSE), normalised mean square error (NMSE), and the FACTor of x (FACTx, where x=2, 3.5, 5, and 10), which represents the fraction of Cp falling within Co/x and x·Co, where Cp and Co represent the observed and calculated concentrations, respectively.

Among the selected indicators, NMSE assigns greater weight to larger differences between observed and simulated values. The NMSE value provides information about bias rather than indicating overestimation or underestimation, resulting in positive values for this indicator. If the distribution of NMSE is relatively consistent, it indicates that the model performs well spatially and temporally. However, a high NMSE does not necessarily imply poor simulation performance of the model, as the sampling points in the tracer experiment reported are all in high concentration gradient areas. Significant differences between the simulated and actual plume directions result in higher NMSE values.