Analysis Of Influencing Factors And Prediction Of Atmospheric Visibility

Atmospheric visibility is an indicator that reflects the transparency of the atmosphere and it is closely related to the daily life of human beings.The visibility of a region can directly reflect the atmospheric environmental quality of the region.However,atmospheric visibility has shown a general downward trend in recent decades.The low atmospheric visibility and high frequency of haze in some areas have seriously affected people's normal life.In addition,low visibility will bring a variety of direct and indirect economic losses,such as reduced tourism income,traffic delays,frequent safety accidents,and potential casualties.With the cost of data collection and storage falling,various types of meteorological data can be accurately measured and saved in most areas,which provides abundant data for scientific research in this field.Analysis of the influencing factors of atmospheric visibility can reveal the potential relationship between atmospheric visibility and other influencing factors(such as temperature,humidity,precipitation,etc.),which is of great significance in terms of environmental relations,air pollution causes and haze control.On the other hand,the accurate prediction of atmospheric visibility can guarantee traffic safety,especially the prediction of low visibility weather information can effectively control and prevent the occurrence of air pollution incidents and reduce various losses caused by polluted weather,which has positive significance for the traffic operation management department,the traveling citizens.In the analysis of influencing factors of atmospheric visibility,traditional statistical methods cannot characterize the complex nonlinear relationship between atmospheric visibility and its influencing factors,so the analysis results of influencing factors can be improved.In the research of multi-site atmospheric visibility prediction,traditional methods are mostly based on modeling and forecasting each meteorological station separately,which not only consumes a lot of time,but also do not consider the geographical location relationship between meteorological stations,so the accuracy of prediction results can be improved.In order to improve this deficiency,this paper proposed an analysis method of influencing factors of atmospheric visibility based on the Granger causality.This method can learn complex non-linear causal relationships in meteorological data,and obtain qualitative and quantitative causality analysis of atmospheric visibility and its influencing factors.And in the multi-site atmospheric visibility prediction section,in order to obtain the influence of the spatial location and temporal feature between meteorological stations on atmospheric visibility,this paper proposed a STAM-GCN(Spatial Temporal Attention Mechanism-Graph Convolutional Network)model,and obtains a highly accurate prediction result by extracting the temporal and spatial features through the convolution module with attention mechanism.In the experimental section,MAE,MSE,RMSE and R~2 are used as the evaluation respectively.The results show that the methods proposed in this paper have achieved better results,which proving that the methods proposed in this paper have certain practical value.