500-m Raster Dataset Development of Fog Duration in Poyang Lake Basin (2018–2025)

LI Haijun^1,2  SHAN Jiusheng^1,2*  LIU Bo¹  YIN Xiaofei¹

1. Jiangxi Meteorological Observatory, Nanchang 330096, China;

2. Jiangxi Province Key Laboratory of Climate Change Risk and Meteorological Disaster Prevention, Nanchang 330096, China

Abstract: Clarifying the evolution and fine spatial distribution of fog is fundamental for refined fog forecast and early warning, which can mitigate adverse fog impacts on human daily activities, low-altitude economy, land transportation and shipping. Current operational fog observations rely exclusively on sparse surface visibility sensors (one station per county) with coarse spatial resolution, making it urgent to develop a high-resolution dataset capturing fine-scale fog characteristics. Based on 15-minute, 500 m visible Channel 2 data from Fengyun-4A/4B satellites, digital elevation model, ground visibility and precipitation observations, this study established a daytime fog dataset for the Poyang Lake Basin. A total of 139 daytime fog cases were identified, fog detection was conducted following the method of Li et al. (2026), and gridded accumulation was implemented to produce final gridded products. The dataset has a 500 m spatial resolution with 593,095 grid points, which is 6,817 times the number of the existing 87 operational surface stations. Validation results for 2018–2025 show a correlation coefficient of 0.85, a relative root mean square error of 0.29, and a mean relative error of 0.21. The dataset exhibits a significant positive correlation, high accuracy, and small deviation, demonstrating reliable applicability. The dataset contains 3 types of daytime fog products covering the Poyang Lake Basin from 2018 to 2025: (1) fog duration of 139 individual daytime fog cases; (2) annual cumulative daytime fog duration; (3) 8-year total cumulative daytime fog duration. The dataset is archived in .tif format, and consists of 149 data files with data size of 1.55 GB (compressed into 1 file with 26.4 MB).

Keywords: fog; spatial distribution; Poyang Lake; Fengyun satellite; high-resolution

DOI: https://doi.org/10.3974/geodp.2026.05.09

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2026.03.05.V1.

1 Introduction

Fog is a weather phenomenon in which a large number of tiny water droplets or ice crystals suspend in the near-surface atmosphere, reducing visibility to below 1 km^[1]. As a hazardous weather condition in a certain situation, fog brings risks and inconvenience to daily life, low-altitude economy, transportation, and shipping^[2–4]. To mitigate the adverse impacts of fog and provide timely fog forecasts and early warnings, it is essential to fully and accurately understand the spatial distribution and temporal evolution of fog in a given region^[5].

At present, meteorological departments mainly observe fog using visibility sensors at surface meteorological stations^[6–8], whose major limitation is the sparsity of stations. Moreover, since fog is strongly influenced by topography and underlying surface type, its spatial distribution is highly nonlinear^[5,6]. Therefore, high spatial precision data that can represent the actual distribution of fog are both scarce and necessary.

With stable and continuous observations and high spatial resolution, meteorological satellites have great potential to improve fog detection. Numerous studies have been conducted on satellite-based gridded fog detection. The classical dual-channel difference method using mid-infrared and long-wave infrared channels enables nighttime fog detection^[9]. Visible channels are used for daytime fog detection based on the unique optical characteristics such as texture and boundaries of fog^[10]. Statistical models^[11] and machine learning models^[12,13] have also been applied for both daytime and nighttime fog identification. Although remarkable progress has been made, these methods still struggle to effectively distinguish fog from low clouds.

In recent years, some researchers have incorporated digital elevation model (DEM) data and effectively improved the separation of fog and low clouds^[14,15]. However, these studies have not considered the spatial distribution characteristics of fog under topographic constraints. In fact, due to local microclimates, fog often concentrates in valleys or over specific underlying surfaces^[5,6]. Using elevation data combined with terrain delineated by natural watersheds as constraints is more consistent with the natural distribution characteristics of fog^[16]. Li^[17] proposed a fog detection method for the entire natural Poyang Lake Basin, which sets thresholds in groups based on the Gaussian distribution of visible reflectance over non-fog surfaces (referred to as the Li 26 algorithm). This method fully considers the topographic constraint of natural watersheds on fog, can partly distinguish fog from low clouds, and achieves high-resolution (500 m) daytime fog detection.

Based on the Li 26 algorithm, this study uses visible data from FY-4A/B meteorological satellites, DEM data, and in situ visibility and precipitation observations. A total of 139 daytime fog cases from 2018 to 2025 over the Poyang Lake Basin were screened and individually detected. By accumulating the detection results, a dataset of daytime fog spatial distribution from 2018 to 2025 was constructed, with a spatial resolution of 500 m. This dataset will help improve the understanding of the fine-scale distribution characteristics of fog in the Poyang Lake Basin, enhance the accuracy of fog forecasting and early warning by meteorological departments, and support more refined assessments of meteorological risks posed by fog to economy, population, transportation, and other sectors.

2 Metadata of the Dataset

The metadata of 500-m raster dataset of fog duration in Poyang Lake Basin (2018–2025)^[18] is summarized in Table 1. It includes the dataset full name, short name, authors, year of the dataset, spatial resolution, data format, data size, data files, data publisher, etc.

3 Methods

3.1 Study Area

The Poyang Lake Basin covers an area of 162,225 km², with a geographic extent of 113°35′– 118°29′ E, 24°29′–30°05′ N (Figure 1). The basin is located in Southern China, with 96.62% of its area lying in Jiangxi Province and the remaining 3.38% distributed across 5 provinces:

Table 1  Metadata summary of the 500-m raster dataset of fog duration in Poyang Lake Basin (2018–2025)

Fujian, Zhejiang, Anhui, Hunan, and Guangdong. The river system mainly consists of Poyang Lake (the largest freshwater lake in China) and its 5 major tributaries: the Ganjiang, Fuhe, Xinjiang, Raohe, and Xiushui Rivers. The basin is dominated by hills and mountains and forms an important part of the hilly region in southeastern China. Its highest peak is Huanggang Mountain in the Wuyi Mountains, with an elevation of 2.16 km, and there are no snow-capped mountains. The basin features typical ridge-valley topography, and local microclimates constrain the spatial distribution of fog. The basin is located among three large urban agglomerations: the Yangtze River Delta, the Pearl River Delta, and the Middle Yangtze River. It has a dense expressway network and is frequently affected by fog.

3.2 Data Sources

Satellite data were obtained from the National Satellite Meteorological Center[1]. Since the operational data of the FengYun-4A (FY-4A) satellite, the primary operational satellite of China’s meteorological department, are available from March 12, 2018, to March 4, 2024, the Channel 2 visible light data of the Advanced Geostationary Radiation Imager (AGRI) aboard FY-4A were used in this study for this period. The data have a central wavelength of 0.65 µm, a spectral bandwidth of 0.55–0.75 µm, and a temporal/spatial resolution of 15 min/500 m. As FY-4B took over operational tasks from FY-4A on March 5, 2024, the Channel 2 visible light data of AGRI aboard FY-4B were adopted from March 5, 2024, to December 31, 2025, with a central wavelength of 0.525 µm, a spectral bandwidth of 0.50–0.55 µm, and an identical temporal/spatial resolution of 15 min/500 m. Satellite data were primarily used for high spatiotemporal resolution identification of fog spatial distribution.

The Digital Elevation Model (DEM) data were acquired from the Geospatial Data Cloud[2]. With an original spatial resolution of 90 m, the DEM data were resampled to 500 m (consistent with the satellite data resolution) using ArcGIS 10.2 software. DEM data were primarily applied in the grouping step of fog identification.

The boundary data of the Poyang Lake Basin^[20] were obtained from the National Earth System Science Data Center[3], covering the year 2000 in .shp format. These boundary data were used to delineate the Poyang Lake Basin as the study region.

Visibility and precipitation data were retrieved from the “Tianqing” Meteorological Big Data Cloud Platform developed by the National Meteorological Information Center. Visibility and precipitation data were observed by visibility sensors and rain gauges at meteorological stations, respectively. Visibility data were mainly used for fog case screening and dataset result verification, while precipitation data were only used for fog case screening.

3.3 Algorithm

3.3.1 Fog Case Selection

Low visibility caused by clouds is frequently recorded at mountainous meteorological stations, so such stations are usually excluded when counting fog station observations^[5]. After excluding 2 mountainous stations (Lushan and Jinggangshan), a total of 85 national meteorological stations remain in the Poyang Lake Basin, and their locations are shown in Figure 1. Referring to Li 26 algorithm, fog cases were delineated using the 2018–2025 visibility and precipitation data from these 85 stations in the Poyang Lake Basin. A fog case was selected when the following two conditions were met simultaneously: the number of stations with visibility≤1 km was ≥15, and the number of stations with both visibility ≤1 km and rainfall intensity≥0.1 mm/h was ≤3 (Equation 1).

                                                                                                                                                                     (1)

where, N denotes the total number of stations, N_vis≤1 represents the number of stations with visibility ≤1 km, and N_{r≥0.1 and vis≤1} denotes the number of stations where visibility ≤1 km and rainfall intensity ≥0.1 mm/h at the same site.

The time period of a diurnal fog case was defined to start at 0.5 h after sunrise in Nanchang (the central city of the Poyang Lake Basin) and end at the moment when the number of stations with visibility ≤1 km dropped below 3. If the number of stations with visibility ≤1 km remained ≥3 throughout the entire day for a fog case, the time period would end at 0.5 h before sunset in Nanchang.

3.3.2 Satellite Based Identification of Fog Spatial Distribution

The identification of a single fog case consisted of several fog detection processes at 15-minute intervals. For a single fog detection, the FY-4A/B data and DEM data at the corresponding moment were used, following the method Li 26 algorithm. The general workflow of this method is as follows: Considering the spatial distribution characteristics of fog that it concentrates along river valleys but is scarce along mountain ridges, the study area was divided into 52 subregions based on sub-basin boundaries (Equation 2), with the subregion division shown in Figure 1. For each subregion, a threshold was set as the minimum elevation value within the subregion plus 0.3 km using DEM data; areas above the threshold were classified as high elevation zones, and those below as low elevation zones (Equation 3). Based on the distribution features that the peak value of the Gaussian distribution of visible light reflectance over the surface of high elevation zones is much smaller than that over fog tops, the identification threshold for each of the 52 subregions was determined as the mean value of visible light reflectance in the high elevation zone plus three times the standard deviation (Equation 4). A visible light grid point with a value higher than the identification threshold of its corresponding subregion was identified as fog.

                          and                 (2)

                                                                                       and                                                                          (3)

                                                                                                                                                                                                                     (4)

where A denotes the Poyang Lake Basin, A₁, A₂, ^…, A₅₂ represent the 1st, 2nd, ^…, 52nd subregions after division, respectively, and A_m, A_n and A_i represent the m-th, n-th, and i-th subregions. A^h_i and A^l_i represent the high elevation zone and low elevation zone of the i-th subregion, respectively. t_i denotes the identification threshold for the i-th subregion, and μ^h_i and σ^h_i represent the mean visible reflectance and standard deviation over the high elevation zone of the i-th subregion, respectively.

3.4 Technical Route

Fog does not persist continuously over time; it typically forms at night and dissipates after sunrise, appearing as temporally discrete fog events. The risk of fog disasters increases significantly only when the density and extent of a fog event expand, resulting in low visibility over a large area. Most fog studies focus on fog cases with a certain spatial scale^[3,5,7–17]. Therefore, datasets developed based on fog cases meet the current general practical research needs.

The technical route for developing this dataset is shown in Figure 2.

Step 1: Selection of fog cases. According to the algorithm described in Section 3.3.1, 139 fog cases were selected during 2018–2025 (Figure 3). Satellite data at 15-minute intervals during the daytime period were obtained for each fog case individually.

Step 2: Satellite-based fog identification. According to the algorithm described in Section 3.3.2, fog areas were identified for each of the 139 fog cases one by one.

Step 3: Accumulation into the dataset. First, grid alignment was carried out. Annual accumulation was performed to obtain the spatial distribution of daytime fog duration for each year from 2018 to 2025. All cases were then accumulated to generate the grid data of the spatial distribution of total daytime fog duration from 2018 to 2025.

4 Data Results and Validation

4.1 Dataset Composition

The dataset consists of spatial distribution raster data of daytime fog duration for three components from 2018 to 2025: (1) the 139 individual fog events; (2) the annual cumulative

Figure 2  Flowchart of the dataset development

Figure 3  Statistical analysis of daily fog station numbers (2018–2025)

values; and (3) the 8-year cumulative value. The dataset is archived in .tif format, with a total of 149 files and a compressed size of 26.4 MB.

4.2 Data Validation

The fog duration derived from in situ observations was compared with that extracted from the dataset at corresponding stations (Table 2), with correlation coefficient (C), relative root mean square error (R), and mean relative error (M) adopted for quantitative accuracy evaluation. For the 8-year cumulative fog duration from 2018 to 2025, the correlation coefficient reached 0.85, indicating a significant positive correlation. The corresponding R and M values were 0.29 and 0.21, respectively, suggesting high overall accuracy, small deviations, and reliable performance of the 8-year cumulative dataset. For annual fog duration during 2018–2025, the correlation coefficients ranged from 0.77 to 0.87, with significant correlations detected for all years. The highest correlation coefficient (0.87) occurred in 2020, while the lowest value (0.77) appeared in 2025. Annual R and M values varied within 0.33–0.53 and 0.24–0.38, respectively. The minimum R value of 0.33 was found in both 2019 and 2021, and the minimum M value of 0.24 was recorded in 2019. Both R and M peaked in 2020, which was attributed to fewer fog cases and lower absolute fog duration values in this year. In general, the annual fog duration products of this dataset from 2018 to 2025 are credible.

Table 2  Verification table of total and annual fog duration from 2018 to 2025

Note: The correlation coefficient is bolded when the significance level of the linear correlation is less than 0.01.

4.3 Data Products

This study developed a 500-m raster dataset of daytime fog duration spatial distribution over the Poyang Lake Basin. The basin covers an area of 162,200 km², but is currently monitored by only 87 surface meteorological stations (including 2 mountain stations). This dataset improves the spatial resolution of fog observation to 500 m, with a total of 593,095 grid points, representing a 6,817-fold increase relative to the number of stations. This significant improvement helps to further understand the fine scale characteristics of fog spatial distribution in this region.

The spatial distribution of the 8-year cumulative daytime fog duration from 2018 to 2025 is shown in Figure 4. Compared with previous understanding of fog spatial distribution^[6], this dataset clearly identifies 3 high-value centers in the northeast, northwest, and west, and reveals their detailed spatial patterns. The high value center in the northeast is larger and more contiguous than those in the northwest and west, which are more isolated. In addition, there are several connected sub-high-value centers in the central basin. Overall, fog duration is higher in the north and central parts than in the south; the variance is larger in the surrounding mountainous areas, while it is more uniform in the central region.

From the perspective of annual cumulative fog duration from 2018 to 2025 (Figure 5), the spatial distribution varies considerably among the 8 years. Overall, fog duration is lower in 2020 and 2022, while the other years are comparable. The northeast region is a high-value center in all 8 years, whereas the northwest and west are not always high-value centers. The central part still tends to form connected high- or sub-high-value centers. In 2024, long duration fog also appeared in the southern part of the basin.

Figure 5  Spatial distribution maps of annual cumulative duration of daytime fog over the Poyang Lake Basin (2018–2025)

5 Discussion and Conclusion

This study developed a daytime fog dataset for the Poyang Lake Basin using 15-minute and 500-m Channel 2 visible observations from the Advanced Geostationary Radiation Imager  onboard Fengyun-4A/4B satellites, combined with digital elevation model data, surface visibility and precipitation measurements. A total of 139 daytime fog cases were identified with the Li 26 algorithm, and gridded fog products were generated via grid alignment and cumulative calculation. The final 500-m resolution dataset contains 593,095 grid points, with spatial resolution improved by 6,817 times compared with the existing 87 in situ meteorological stations. Three statistical indicators—correlation coefficient (C), relative root mean square error (R), and mean relative error (M)—were used for dataset validation against station observations. The multi-year averaged C, R and M values for 2018–2025 were 0.85, 0.29 and 0.21, while the annual values ranged within 0.77–0.87, 0.33–0.53 and 0.24–0.38, respectively. Validation results reveal that the dataset has significant positive correlation, high accuracy and small deviations, and both annual and 8-year cumulative fog products are reliable. This study demonstrates that domestic Fengyun-4A/4B satellites have great application potential for satellite-based daytime fog detection.

Benefiting from the high spatial resolution of 500 m, this dataset greatly improves the observational refinement of daytime fog spatial distribution across the Poyang Lake Basin, and further advances our understanding of regional fog distribution characteristics. Nevertheless, several limitations still exist. Firstly, the visible satellite data adopted in this study depend on solar reflectance during daytime, so nighttime data are not included, and the dataset fails to describe and capture the initial formation stage of nocturnal fog. Secondly, this study only selected fog cases observed by at least 15 meteorological stations, which conforms to the general research convention of current fog case studies^[3,5,7–17]. However, only 139 qualified fog cases were extracted in total, which may underestimate the cumulative fog duration in local fog-prone areas (such as patchy fog zones) affected by small-scale topographic variations. Thirdly, accurate discrimination between fog and low clouds remains one of the most intractable challenges for existing satellite fog monitoring techniques^[14–17]. Although the introduction of the DEM algorithm enhances the capability to separate fog from low clouds in this dataset, there is still considerable room for overall performance improvement.

Author Contributions

Li, H. J. and Shan, J. S. conducted the overall design for the dataset development. Li, H. J. and Liu, B. collected and processed the data. Li, H. J. designed the models and algorithms. Li, H. J. and Yin, X. F. performed the data validation. Li, H. J. wrote the data paper.

Acknowledgements

Special thanks go to Engineer Huang, Y. J. of the Jiangxi Meteorological Observatory for her help in English translation.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]        China Meteorological Administration. Specifications for surface meteorological observation. Part 4: observation of weather phenomenon (QX/T 48—2007) [S]. Beijing: Standards Press of China, 2007.

[2]        Guo, J. P., Zhang, W., Zhou, B. W., et al. Progress and prospect of research on the key frontier scientific and technological issues in low-altitude economy meteorology [J]. Meteorological Monthly, 2025, 51(11): 1455–1476.

[3]        Gong, X. Q., Zhao, Y., Wang, W., et al. Analysis of traffic risk due to fog events using numerical forecast model outputs [J]. Journal of Safety and Environment, 2026, 26(1): 142–150.

[4]        Yang, Z. N., Gu, T. T., Pan, Y. Y. GIS-based risk regionalization of fog disaster on highway in Zhejiang Province [J]. Meteorological Science and Technology, 2017, 45(4): 756–760.

[5]        Shi, C. E., Zhang, H., Yang, G. Y., et al. Climatic characteristics and influencing factors of regional extremely dense fog in Anhui Province [J]. Acta Meteorologica Sinica, 2022, 80(6): 924–939.

[6]        Yang, H., Liu, Y. X., Zhang, K. M. The temporal and spatial distribution of fog days in Jiangxi Province [J]. Meteorology and Disaster Reduction Research, 2015, 38(1): 66–72.

[7]        Xu, A. H., Chen, X. X., Xiao, A., et al. Analysis on the characteristics of meteorological factors and forecast ideas for regional advection fog in Jiangxi [J]. Meteorological Monthly, 2016, 42(3): 372–381.

[8]        Sang, X. Z., Chen, S. J., Liu, D. Y., et al. Circulation anomalies and formation mechanisms for late autumn and early winter fog occurrences over eastern China [J]. Acta Meteorologica Sinica, 2025, 83(2): 240–256.

[9]        Eyre, J. R., Brownscombe, J. L., Allam, R. J. Detection of fog at night using Advanced Very High Resolution Radiometer ( AVHRR) imagery [J]. Meteorology Magazine, 1984, 113: 266–271.

[10]     Ma, H. Y., Li, Y. N., Wu, X. J., et al. Daytime land fog detection based on H8/AHI satellite data [J]. Remote Sensing Technology and Application, 2024, 39(4): 784–792.

[11]     Lu, B., Wang, B., Yu, B., et al. Research on analysis of daytime sea fog detection threshold in the East China Sea and visibility inversion based on Himawari-8 [J]. Marine Forecasts, 2022, 39(6): 62–73.

[12]     Liang, X. N., Ren, C. P., Wang, Z., et al. Identification of heavy fog on the Jinliwen expressway in Zhejiang Province by a new generation geostationary meteorological satellite [J]. Climatic and Environmental Research, 2023, 28 (5): 471−482.

[13]     Huang, Y. H., Fu, R. D., Fang, X. Y., et al. Day time sea fog recognition based on multi-scale feature fusion generation under attention mechanism [J]. Remote Sensing Technology and Application, 2025, 40(1): 258–264.

[14]     Shang, H. Z., Chen, L. F., Letu, H. S., et al. Development of a daytime cloud and haze detection algorithm for Himawari‐8 satellite measurements over central and eastern China [J]. Journal of Geophysical Research Atmospheres, 2017. 122, 3528–3543.

[15]     Ran, Y. Z., Ma, H. Y., Liu, Z. W., et al. Satellite fog detection at dawn and dusk based on the deep learning algorithm under terrain-restriction [J]. Remote Sensing, 2022, 14, 4328.

[16]     Kurose, Y., Fukaishi, K., Hayashi, Y. et al. Climatological characteristics of basin fog at the Ohzu Basin, Ehime Prefecture [J]. Journal of Agricultural Meteorology, 1998, 54(1): 13–21.

[17]     Li, H. J., Shan, J. S., Xin, J. J., et al. High-resolution fog dection over a mountainous lake basin [J]. Journal of Meteorological Research, 2026, 40(1), 173–193.

[18]     Li, H. J., Shan, J. S., Liu, B., et al. 500-m raster dataset of fog duration in Poyang Lake Basin [J/DB/OL]. Digital Journal of Global Change Data Repository, 2026. https://doi.org/10.3974/geodb.2026.03.05.V1.

[19]     GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[20]     Ma, R. H. 1:1,000,000 scale vector boundary dataset of the Poyang Lake Basin (2000) [DS/OL]. National Earth System Science Data Center, 2015. https://www.geodata.cn/main/face_science_detail?typeName= face_science&guid=201525230652027.