Identifying dust events of Inner Mongolia based on multi-source observations

Using conventional meteorological observation data and environmental station PM,AQI,NO₂ data,as well as inversion products based on satellite data,the gradient modeling idea of random forest and support vector machine algorithm was used to identify whether 119 national meteorological observation stations in Inner Mongolia have experienced sandstorm weather.To improve the accuracy and spatial coverage of identification,environmental station data and satellite products were integrated,and the manually verified records of sandstorm,haze,and other factors at Baotou Airport from 2011 to 2020 were tested.A study on 64 dust and sand processes,29 haze processes,and 151 heavy fog processes from 2011 to 2023 found that:1.Meteorological elements such as visibility,relative humidity,air pressure,dew point temperature,and average wind speed,as well as weather phenomena (floating dust,blowing sand,sandstorms,haze,fog,etc.),have a significant contribution to identifying dust and sand weather.The contributions of visibility,weather phenomena,and relative humidity are 0.3,0.23,and 0.22,respectively.When dust and sand weather occurs,the average relative humidity decreases by 40% and the average wind speed increases by more than 2 times.However,when haze weather occurs,the changes in relative humidity and average wind speed are opposite to those of dust and sand weather,with an increase of 50-70% and decrease of 30-50%.The contribution of environmental factors in descending order is NO₂,CO,AQI,PM₁₀,PM,PM_2.5,PM₁₀_ratio,During sandstorm weather,AQI increase by nearly 6 times,while during haze weather it was only 0.7-2.5 times.During haze weather,the concentrations of NO₂ and CO increase by about 2 times,while during sandstorm weather,the average concentration of NO₂ decrease by 66%.3.Using a random forest model,calculate 53.5% relative humidity,1.4×10⁵ μg · m^-3 PM₁₀,1.1 μg · m^-3 CO,average wind speed exceeding 9.7 m · s^-1,and historical average increase of 1.2 m · s^-1 from the same month is a set of thresholds that can distinguish between dust and haze weather.4.By first establishing a random forest model for meteorological data,and then establishing a support vector machine model for environmental data,the final coverage and accuracy of the model are 92% and 93%.