Please wait a minute...
Journal of Arid Land
Journal of Arid Land  2025, Vol. 17 Issue (1): 93-111    DOI: 10.1007/s40333-024-0066-3     CSTR: 32276.14.JAL.02400663
Research article     
An improved GCN-TCN-AR model for PM2.5 predictions in the arid areas of Xinjiang, China
CHEN Wenqian1,*(), BAI Xuesong1, ZHANG Na1, CAO Xiaoyi2
1School of Information and Control Engineering, Qingdao University of Technology, Qingdao 266520, China
2Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, College of Atmospheric Sciences, Lanzhou University, Lanzhou 730000, China
Download: HTML     PDF(2137KB)
Export: BibTeX | EndNote (RIS)      

Abstract  

As one of the main characteristics of atmospheric pollutants, PM2.5 severely affects human health and has received widespread attention in recent years. How to predict the variations of PM2.5 concentrations with high accuracy is an important topic. The PM2.5 monitoring stations in Xinjiang Uygur Autonomous Region, China, are unevenly distributed, which makes it challenging to conduct comprehensive analyses and predictions. Therefore, this study primarily addresses the limitations mentioned above and the poor generalization ability of PM2.5 concentration prediction models across different monitoring stations. We chose the northern slope of the Tianshan Mountains as the study area and took the January-December in 2019 as the research period. On the basis of data from 21 PM2.5 monitoring stations as well as meteorological data (temperature, instantaneous wind speed, and pressure), we developed an improved model, namely GCN-TCN-AR (where GCN is the graph convolution network, TCN is the temporal convolutional network, and AR is the autoregression), for predicting PM2.5 concentrations on the northern slope of the Tianshan Mountains. The GCN-TCN-AR model is composed of an improved GCN model, a TCN model, and an AR model. The results revealed that the R2 values predicted by the GCN-TCN-AR model at the four monitoring stations (Urumqi, Wujiaqu, Shihezi, and Changji) were 0.93, 0.91, 0.93, and 0.92, respectively, and the RMSE (root mean square error) values were 6.85, 7.52, 7.01, and 7.28 μg/m³, respectively. The performance of the GCN-TCN-AR model was also compared with the currently neural network models, including the GCN-TCN, GCN, TCN, Support Vector Regression (SVR), and AR. The GCN-TCN-AR outperformed the other current neural network models, with high prediction accuracy and good stability, making it especially suitable for the predictions of PM2.5 concentrations. This study revealed the significant spatiotemporal variations of PM2.5 concentrations. First, the PM2.5 concentrations exhibited clear seasonal fluctuations, with higher levels typically observed in winter and differences presented between months. Second, the spatial distribution analysis revealed that cities such as Urumqi and Wujiaqu have high PM2.5 concentrations, with a noticeable geographical clustering of pollutions. Understanding the variations in PM2.5 concentrations is highly important for the sustainable development of ecological environment in arid areas.

Key words： air pollution      PM2.5 concentrations      graph convolution network (GCN) model      temporal convolutional network (TCN) model      autoregression (AR) model      northern slope of the Tianshan Mountains     
Received: 18 September 2024      Published: 31 January 2025
Corresponding Authors: *CHEN Wenqian (E-mail: chimmyqu@yeah.net)
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
CHEN Wenqian
BAI Xuesong
ZHANG Na
CAO Xiaoyi
Cite this article:

CHEN Wenqian, BAI Xuesong, ZHANG Na, CAO Xiaoyi. An improved GCN-TCN-AR model for PM2.5 predictions in the arid areas of Xinjiang, China. Journal of Arid Land, 2025, 17(1): 93-111.

URL:

http://jal.xjegi.com/10.1007/s40333-024-0066-3     OR     http://jal.xjegi.com/Y2025/V17/I1/93

Fig. 1 Overview of the study area (northern slope of the Tianshan Mountains) based on the digital elevation model (DEM) and the spatial distributions of PM2.5 monitoring stations and meteorological stations
Fig. 2 Framework using the developed GCN-TCN-AR (where GCN is the graph convolution network, TCN is the temporal convolutional network, and AR is the autoregression) model to predict PM2.5 concentrations
Parameter Value Parameter Value
Number of records
Number of PM2.5 monitoring stations
Number of meteorological stations
Training set
Validation set
Testing set		 365
21
15
60%
20%
20%		 Batch size
Loss function
Learning rate
Epochs
Adjusting rate
Optimizer		 64
Mean square error
0.001
200
0.6
Adam
Table 1 Parameter settings of the GCN-TCN-AR (where GCN is the graph convolution network, TCN is the temporal convolutional network, and AR is the autoregression) model for predicting the PM2.5 concentrations
Fig. 3 Plot showing the change in mean squared error (MSE) on the validation set with epochs and the training loss change during the training process
Fig. 4 Heatmap showing the relationship between PM2.5 concentration and meteorological factors. ** indicates a significant correlation at the P<0.01 level; * indicates a significant correlation at the P<0.05 level.
Meteorological factor Spring Summer Autumn Winter Annual
Instantaneous wind speed -0.26** -0.41** -0.39** 0.18** -0.47**
Temperature -0.11** -0.34** 0.05* -0.05* -0.39**
Pressure -0.04 0.06* -0.04 -0.03 -0.07*
Table 2 Correlation coefficients between PM2.5 concentration and meteorological factors at the seasonal and annual scales
Fig. 5 Predictions of PM2.5 concentrations at four monitoring stations from January to December in 2019 based on the GCN-TCN-AR model. (a), Urumqi; (b), Wujiaqu; (c), Shihezi; (d), Changji.
Fig. 6 Scatter plots showing the relationship between the predicted PM2.5 concentrations by the GCN-TCN-AR model and observed values at the daily scale in 2019 at four monitoring stations. (a), Urumqi; (b), Wujiaqu; (c), Shihezi; (d), Changji. n, sample size; R2, determination of coefficient; RMSE, root mean square error.
Fig. 7 Overall performance results for models used to predict the PM2.5 concentrations at the hourly scale. (a), RMSE; (b), MAE (mean absolute error). SVR, Support Vector Regression.
Fig. 8 Gravity center shift in the PM2.5 concentrations on the northern slope of the Tianshan Mountains in 2019. (a), Huyanghe City; (b), Karamay City; (c), Urumqi city; (d), Changji Hui Autonomous Prefecture; (e), Kuytun City; (f), Tacheng Prefecture; (g), Shihezi City; (h); Wujiaqu City; (i), Turpan City. The upper and lower edges of the box represent the upper quartile and the lower quartile, respectively, with the middle line representing the median; the whiskers represent the data range from the lower quartile to the minimum value and from the upper quartile to the maximum value, excluding outliers; the circles represent the outliers.
Fig. 9 Annual average PM2.5 concentration in the study area (the northern slope of the Tianshan Mountains) and its major cities and prefecture. Bars means standard errors.
Fig. 10 Spatial distribution characteristics of annual mean PM2.5 concentrations on the northern slope of the Tianshan Mountains in 2019
Fig. 11 Monthly spatial distribution characteristics of PM2.5 concentrations on the northern slope of the Tianshan Mountains from January to December in 2019 (a-l)
[1]   Bai S N, Shen X L. 2019. PM2.5 prediction based on LSTM recurrent neural network. Computer Application and Software, 36(1): 67-70, 104. (in Chinese)
[2]   Bhatt D, Patel C, Talsania H, et al. 2021. CNN variants for computer vision: history, architecture, application, challenges and future scope. Electronics, 10(20): 2470, doi: 10.1109/ACCESS.2021.3060744.
[3]   Chen T Q, Rubanova Y, Bettencourt J, et al. 2018. Neural ordinary differential equations. In: Bengio S. Proceedings of the 32nd International Conference on Neural Information Processing Systems (NIPS'18). NewYork: Curran Associates Inc., 6572-6583.
[4]   Chen Z Y, Chen D L, Zhao C F, et al. 2020. Influence of meteorological conditions on PM2.5concentrations across China: A review of methodology and mechanism. Environment International, 139: 105558, doi: 10.1016/j.envint.2020.105558.
[5]   Dong S, Wang P, Abbas K. 2021. A survey on deep learning and its applications. Computer Science Review, 40: 100379, doi: 10.1016/j.cosrev.2021.100379.
[6]   Gao X, Li W D. 2021. A graph-based LSTM model for PM2.5 forecasting. Atmospheric Pollution Research, 12(9): 101150, doi: 10.1016/j.apr.2021.101150.
[7]   Guo B, Wang Y Q, Zhang X Y, et al. 2020. Temporal and spatial variations of haze and fog and the characteristics of PM2.5 during heavy pollution episodes in China from 2013 to 2018. Atmospheric Pollution Research, 11(10): 1847-1856.
[8]   Jiang F X, Zhang C Y, Sun S L, et al. 2021. Forecasting hourly PM2.5 based on deep temporal convolutional neural network and decomposition method. Applied Soft Computing, 113: 107988, doi: 10.1016/j.asoc.2021.107988.
[9]   Li K J, Talifu D, Gao B, et al. 2022. Temporal distribution and source apportionment of composition of ambient PM2.5 in Urumqi, North-West China. Atmosphere, 13(5): 781, doi: 10.3390/atmos13050781.
[10]   Li X L, Qin D, He X L, et al. 2024. Spatial and temporal changes in land use and landscape pattern evolution in the economic belt of the northern slope of the Tianshan Mountains in China. Sustainability, 16(16): 7003, doi: 10.3390/su16167003.
[11]   Liu X P, Zou B, Feng H H, et al. 2020. Anthropogenic factors of PM2.5 distributions in China's major urban agglomerations: A spatial-temporal analysis. Journal of Cleaner Production, 264(10): 121709, doi: 10.1016/j.jclepro.2020.121709.
[12]   Liu Y, He L J, Qin W M, et al. 2021. The effect of urban form on PM2.5 concentration: evidence from China's 340 prefecture-level cities. Remote Sensing, 14(1): 7, doi: 10.3390/rs14010007.
[13]   Lu J, Li B, Li H, et al. 2021. Expansion of city scale, traffic modes, traffic congestion, and air pollution. Cities, 108: 102974, doi: 10.1016/j.cities.2020.102974.
[14]   Lu Y, Li K. 2023. Multistation collaborative prediction of air pollutants based on the CNN-BiLSTM model. Environmental Science and Pollution Research, 30: 92417-92435.
[15]   Luo Y T, Xu L P, Li Z Q, et al. 2023. Air pollution in heavy industrial cities along the northern slope of the Tianshan Mountains, Xinjiang: characteristics, meteorological influence, and sources. Environmental Science and Pollution Research, 30: 55092-55111.
[16]   Ma J, Ding Y X, Cheng J C, et al. 2020. Identification of high impact factors of air quality on a national scale using big data and machine learning techniques. Journal of Cleaner Production, 244: 118955, doi: 10.1016/j.jclepro.2019.118955.
[17]   Ma W, Ding J L, Wang R, et al. 2022. Drivers of PM2.5 in the urban agglomeration on the northern slope of the Tianshan Mountains, China. Environmental Pollution, 309: 119777, doi: 10.1016/j.envpol.2022.119777.
[18]   Mo H H, You Y C, Wu L P, et al. 2023. Potential impact of industrial transfer on PM2.5 and economic development under scenarios oriented by different objectives in Guangdong, China. Environmental Pollution, 316: 120562, doi: 10.1016/j.envpol.2022.120562.
[19]   Mohamed S A. 2019. MicroRNA detection in the pathogenesis of BAV‐associated aortopathy‐mediated vascular remodelling through EndMT/EMT. Journal of Internal Medicine, 285(1): 115-117.
doi: 10.1111/joim.12856 pmid: 30478994
[20]   Mohammadzadeh A K, Salah H, Jahanmahin R, et al. 2024. Spatiotemporal integration of GCN and E-LSTM networks for PM2.5 forecasting. Machine Learning with Applications, 15: 100521, doi: 10.1016/j.mlwa.2023.100521.
[21]   Peng J B, Huang Y, Liu T, et al. 2019. Atmospheric nitrogen pollution in urban agglomeration and its impact on alpine lake-case study of Tianchi Lake. Science of the Total Environment, 688: 312-323.
[22]   Qi Y L, Li Q, Karimian H, et al. 2019. A hybrid model for spatiotemporal forecasting of PM2.5 based on graph convolutional neural network and long short-term memory. Science of the Total Environment, 664: 1-10.
[23]   Ren Y, Wang S Y, Xia B S. 2023. Deep learning coupled model based on TCN-LSTM for particulate matter concentration prediction. Atmospheric Pollution Research,14(4): 101703, doi: 10.1016/j.apr.2023.101703.
[24]   Saha S, Gan Z T, Cheng L, et al. 2021. Hierarchical deep learning neural network (HiDeNN): an artificial intelligence (AI) framework for computational science and engineering. Computer Methods in Applied Mechanics and Engineering, 373: 113452, doi: 10.1016/j.cma.2020.113452.
[25]   Shi T, Li P Y, Yang W, et al. 2023. Application of TCN-biGRU neural network in PM2.5 concentration prediction. Environmental Science and Pollution Research, 30: 119506-119517.
[26]   Sivarethinamohan R, Sujatha S, Priya S, et al. 2021. Impact of air pollution in health and socio-economic aspects:review on future approach. Materials Today: Proceedings, 37(2): 2725-2729.
[27]   Sun L X, Yu X, Li B S, et al. 2020. Coupling analysis of the major impact on sustainable development of the typical arid region of Turpan in Northwest China. Regional Sustainability, 1(1): 48-58.
doi: 10.1016/j.regsus.2020.08.002
[28]   Wang H, Gu Z J, Wang D, et al. 2024a. Evolution characteristics of Akdala PM2.5 and correlation analysis with meteorological elements. Sichuan Environment, 43(1): 8-15. (in Chinese)
[29]   Wang J, Wu T, Mao J J, et al. 2024b. A forecasting framework on fusion of spatiotemporal features for multi-station PM2.5. Expert Systems with Applications, 238: 121951, doi: 10.1016/j.eswa.2023.121951.
[30]   Wu X H, Song L H, Li Q L, et al. 2021. Characteristics of temporal and spatial distribution of atmospheric PM2.5and PM10 in urban Taiyuan, China. Journal of Ecological Environment, 30(4): 756-762. (in Chinese)
[31]   Xia X S, Chen J J, Wang J J, et al. 2020. China PM2.5 based on random forest model analysis of 5 factors influencing concentration. Environmental Science, 41(5): 2057-2065. (in Chinese)
[32]   Xing H T, Guo J L, Liu S A, et al. 2022. NOx emission prediction based on CNN-LSTM hybrid neural network model. Electronic Measurement, 45(2): 98-103. (in Chinese)
[33]   Yang Q Q, Yuan Q Q, Li T W, et al. 2017. The relationships between PM2.5 and meteorological factors in China: seasonal and regional variations. International Journal of Environmental Research and Public Health, 14(12): 1510, doi: 10.3390/ijerph14121510.
[34]   Yao Y J, Gong Y G, Liu J, et al. 2023. Overview of intelligent question answering systems based on deep learning. Computer System Applications, 32(4): 1-15. (in Chinese)
[35]   Ye S, Wang P, Huang Y, et al. 2023. The spatial form of cities in the Yangtze River Delta urban agglomeration affects PM2.5 study on the influence of spatial heterogeneity characteristics of O3 pollution. Journal of Ecological Environment, 32(10): 1771-1784. (in Chinese)
[36]   Yin Z M, Cui K P, Chen S D, et al. 2019. Characterization of the air quality index for Urumqi and Turfan cities, China. Aerosol and Air Quality Research, 19(2): 282-306.
[37]   Yu Z Q, Qu Y H, Zhou G Q, et al. 2020. Numerical study on the sources of PM2.5 pollution in the Yangtze River Delta region in autumn and winter 2018. China Environmental Science, 40(10): 4237-4246. (in Chinese)
[38]   Zeng Q L, Wang L H, Zhu S Y, et al. 2023. Long-term PM2.5 concentrations forecasting using CEEMDAN and deep Transformer neural network. Atmospheric Pollution Research, 14(9): 101839, doi: 10.1016/j.apr.2023.101839.
[39]   Zhang M J, Wu Q Q, Zhang J, et al. 2022. Fluid micelle network for image super-resolution reconstruction. IEEE Transactions on Cybernetics, 53(1): 578-591.
[40]   Zhang S Q, Hu W, Zhao X M. 2024. Multi site air quality prediction model based on adaptive hierarchical graph convolution. Application of Computer System, 33(5): 127-135. (in Chinese)
[41]   Zhang Y W, Yuan H W, Sun X, et al. 2021. PM2.5 concentration prediction method based on Adam attention mechanism. Journal of Atmospheric and Environmental Optics, 16(2): 117-126. (in Chinese)
[42]   Zhao G Y, He D H, Huang Y F, et al. 2021. Near-surface PM2.5 prediction combining the complex network characterization and graph convolution neural network. Neural Computing and Applications, 33: 17081-17101.
[43]   Zhen Z, Liu J Y, Niu Y Z, et al. 2022. Analysis of factors influencing PM2.5 in Harbin City based on multivariate time series. Journal of Henan Normal University (Natural Science Edition), 50(1): 98-107. (in Chinese)
[44]   Zhou Y L, Chang F J, Chang L C, et al. 2019. Multi-output support vector machine for regional multi-step-ahead PM2.5 forecasting. Science of the Total Environment, 651(1): 230-240.
[1] ZHANG Shubao, LEI Jun, TONG Yanjun, ZHANG Xiaolei, LU Danni, FAN Liqin, DUAN Zuliang. Temporal and spatial responses of ecological resilience to climate change and human activities in the economic belt on the northern slope of the Tianshan Mountains, China[J]. Journal of Arid Land, 2023, 15(10): 1245-1268.
[2] Jingxin LI, Shigong WANG, Jinhua CHU, Jiaxin WANG, Xu LI, Man YUE, Kezheng SHANG. Characteristics of air pollution events over Hotan Prefecture at the southwestern edge of Taklimakan Desert, China[J]. Journal of Arid Land, 2018, 10(5): 686-700.