Comparison of the Visibility Grading Forecast Method Based on Meteorological Factors and Environmental Factors

Te main visibility forecast factors were identifed with the support of data from routine meteorological observations from the Mianyang Airport and the Mianyang Environmental Monitoring Station from 2015 to 2018, and a visibility grading forecast model was established and tested by dint of the multiple linear regression and the KNN algorithm based on big data mining technology, and the variation characteristics of visibility in winter at the Mianyang Airport were studied. Te results showed that (1) in addition to having a signifcant positive correlation with wind speed, the visibility in winter at the Mianyang Airport has a signifcant negative correlation with relative humidity, dew point temperature, AQI, PM 2.5 concentration, PM 10 concentration, and CO, and it has the strongest correlation with relative humidity, and the correlation coefcient is − 0.76. (2) Te multivariate linear regression model and the KNN model were adopted for grading forecasting experiments on visibility, and the results revealed that both models could be used for visibility grading forecasts. Te multiple regression model secures an accuracy of over 70% for forecasts of level 1–5 visibility. In terms of the KNN model, the forecast accuracy is the best when K =3 or K =5, notably for level-2, level-4, and level-5 visibility. Te forecast accuracy rate is 100% for level-2 visibility, but the forecast for level-1 visibility is poor. (3) Te minimum value of the average daily visibility of the Mianyang Airport in winter appeared at 09: 00 and the maximum value appeared at 17: 00. Te level-1 visibility occurred and developed before 09 :00 and faded and vanished between 08 :00 and 15:00.


Introduction
Atmospheric visibility is closely related to life because low visibility can cause fight delays and induce trafc accidents, which directly afect the safety of railway, road, sailing, and air trafc [1][2][3].Te visibility forecast is a key part of the meteorological forecast of civil aviation airports and must be included in each forecast message.Low visibility makes it difcult for pilots to fnd the runway to land [4], which is a crucial factor afecting the safety and efciency of fight and fight training, so it is of great signifcance to improve the accuracy of airport visibility forecasts.Visibility in civil aviation is defned as the degree of turbidity of the atmosphere or the transparency of the atmosphere.In the daytime, visibility refers to the maximum distance from which a person with normal vision (visual contrast threshold of 0.05) can see or identify a moderately sized black target from the sky background under the weather conditions at that time.In the nighttime, visibility refers to the maximum distance from which a person with normal vision (visual contrast threshold of 0.05) can see or identify a moderately sized black target from the sky background or a luminous object of moderate intensity, under the assumption that the overall illumination increases to normal daytime levels [5].
As industry and transportation develop rapidly, the emission of atmospheric pollutants has intensifed, and a large number of dust, smoke, or salt particles suspended in the atmosphere seriously impact the visibility, making the study of the changing trend of visibility a popular topic worldwide [6][7][8][9][10][11][12].Visibility is afected not only by local pollution but also by meteorological factors and particles [13][14][15][16][17][18][19][20][21][22].Wen and Yeh [23] pointed out that the concentration of atmospheric pollutants has a substantial infuence on visibility, and wind speed is an important meteorological parameter afecting atmospheric turbidity for it facilitates the difusion of air pollutants.Li et al. [24] explored the association between visibility and relative humidity, wind speed, temperature, and air pressure to analyze the correlation between atmospheric visibility and meteorological elements in Dalian.Li et al. [25] discovered that atmospheric visibility varies with relative humidity and particulate concentration on a monthly and daily basis.Li et al. [26] noted that the decline in visibility in the Sichuan Basin is mainly caused by PM.While analyzing the causes of impact on visibility, some scholars have also built visibility forecasting models.Li et al. [27] employed the statisticaldynamic method and developed a visibility forecasting equation on the basis of the meteorological concept model.Liang and Hou [28] used the forecast factor index method to forecast visibility, and concluded that visibility had a strong correlation with humidity, temperature, and wind speed.Some scholars used diagnostic analysis methods to develop visibility forecasting equations.Hu et al. [29] adopted neural network step-by-step classifcation modeling to forecast visibility.Shu et al. [30] employed the least squares method to create a dynamic-statistical model of visibility forecasting in Shanghai with the support of historical data on visibility and atmospheric pollutants.Zhou et al. [31] used MM5 numerical forecasting products to establish regression equations and provide low-visibility grading forecasts.Te weather research and forecasting (WRF) model was utilized by Li et al. [32] to statistically simulate fog production.
Although scholars both at home and abroad have conducted a lot of research on visibility forecasting with diferent methods, the accuracy of current visibility forecast can still not reach 100% due to the complexity of the weather, and it is necessary to continuously explore fresh ideas and methods for visibility forecast research in order to improve the accuracy of forecasting.In view of the complexity of visibility forecasting and various impact factors of visibility in diferent regions, the visibility grading forecast of the fight training airport in the Civil Aviation Flight University of China will be studied in this paper.Te fight training airport at the Civil Aviation Flight University of China is located in the Mianyang city, Sichuan Province, which is the same airport as the Mianyang Airport (hereinafter referred to as Mianyang Airport), and mainly undertakes fight teaching tasks for primary teaching aircraft, medium-sized teaching aircraft, and high-level teaching aircraft.Since the airport is in the Sichuan Basin, the relative humidity is high, and low visibility events are common in the winter, which afects the safety and efciency of fight training [33], so it is necessary to study the factors afecting visibility in winter, build a forecast model, and provide a reference for fight training support.In this paper, with the support of data from routine meteorological observations from the Mianyang Airport and the Mianyang Environmental Monitoring Station, the visibility grading forecast model in winter is constructed by dint of multiple linear regression and the KNN algorithm based on big data mining technology, according to the fight training needs.Te two forecast models are tested and comparatively analyzed, in order to provide a reference for the visibility grading forecast at the training airport and an objective product for visibility forecast and early warning.

Sources of Materials.
Te meteorological observation data used in this paper are collected from the hourly monitoring data provided by the Automated Weather Observing Systems (AWOS) from 2015 to 2018, including air pressure, corrected sea level pressure, temperature, relative humidity, dew point temperature, total cloud cover, low cloud cover, wind (wind direction and wind speed), and visibility.Te height of the air pressure data is the runway elevation.Te height of the corrected sea level pressure data is the sea level height.Te height of the temperature data is 1.5 meters above the ground.Te height of the relative humidity data is 1.5 meters above the ground.Te height of dew point temperature data is 1.5 meters above the ground.In this paper, the quality of the aforementioned data is well controlled, the data format is unifed, the missing measurement and abnormal data are eliminated, and the interpolation method is also used to interpolate the data.Due to the inconsistencies in data dimensions, the data are standardized in this paper.

Multiple Linear Regression Model.
Te linear regression model with the dependent variable y and the independent variables x 1 , x 2 , . .., x p is where β 0 , β 1 , . .., β p denote unknown parameters, β 0 is the regression constant, β 1 , . .., β p refer to the regression coefcient, and ε signifes a random error.When p ≥ 2, the model is called a multivariate linear regression model.

KNN Algorithm.
In recent years, the KNN algorithm has attracted much attention in the feld of meteorology.As one of the ten classic algorithms of data mining, it is a nonparametric supervision algorithm proposed by Cover and Hart [34], and a nonparameter estimation technology for classifcation through the calculation of the distance between diferent eigenvalues of objects.Tis rapidly developing practical data mining technology has applications in precipitation forecasting [35], wind forecasting [36][37][38], and cloud classifcation [39].Te idea is that if the majority 2 Advances in Meteorology of K most similar (the closest proximity) samples in the feature space belong to a certain category, then the sample also belongs to that category, where K is usually an integer not greater than 20.In the KNN algorithm, the selected neighbors are objects that have been correctly categorized.
Te principle of KNN is to calculate the distance between the sample to be labeled and each sample in the dataset, and take the nearest K samples.Te category of samples to be labeled is determined through a vote based on K nearest samples.
Te K nearest neighbor algorithm system design is as follows: 1. calculate the distance.In this paper, the Euclidean distance method is used to measure the distance.2. Choose a neighbor.For each sample to be classifed, the distance between it and the training sample is calculated, and the nearest K samples are selected as their neighbors.K is determined by cross-validation.3. Determine the category.According to the categories of K neighbors, the categories of samples to be classifed are determined by majority voting.
Te concrete steps are as follows: suppose there is a set of historical weather sample sets defned as S. S consists of i weather samples, and each weather sample consists of m attribute variations and 1 label quantity.Its mathematical expression is In this paper, the attribute variable X ij represents six factors, namely, AQI, PM 2.5 , PM 10 , relative humidity, dew point temperature, and wind speed.L i is called the label quantity, and the label quantity refers to the visibility level.Te mathematical model for predicting visibility level by dint of the KNN algorithm is as follows: we suppose that the set of six elements of AQI, PM 2.5 , PM 10 , relative humidity, dew point temperature, and wind speed on the forecast day is Y i � (y 1 , y 2 , . .., y m ), which is called the prediction sample.When conducting the forecast, we frst fnd K nearest neighbors that are most similar to the prediction sample Y i in the training sample set S (K is usually an odd number); Ten, we fnd the set of K label quantities (visibility levels) L � (L 1 , L 2 , . .., L k ).Finally, in accordance with the majority rule in voting, L i was selected as the prediction result of the prediction sample Y i .

Visibility Grading Standard.
Visibility is a major factor infuencing general aviation fight training, and its value can determine whether captain training or solo training is conducted.Winter has the highest number of low-visibility days and has the greatest impact on fight training.In order to better provide fight training support and accurate visibility forecast, the visibility is graded according to the aircraft fight standards.Visibility grading forecast has emerged as a new topic in recent years [31,40,41], but there are no grading standards in place.In this paper, visibility is divided into fve levels according to the aircraft fight standards (Table 1), with 800 m, 1,600 m, 2,000 m, and 5,000 m as the demarcation values based on the level of visibility.

Visibility Grading Forecasting Methods of the
Mianyang Airport in Winter and Its Characteristics

Correlation Analysis of Visibility with Each Parameter in
Winter.Since low visibility events occur most frequently in winter (December, January, and February) and afect the safety and efciency of fight training, this paper focuses on the winter visibility at the Mianyang Airport.14 meteorological factors and environmental factors were selected for analysis on their correlation with corresponding visibility, and the correlation coefcients were obtained.Tose factors with correlation coefcients greater than or equal to 0.3 were taken as high-impact physical quantities and passed the signifcance level test of a � 0.05. Figure 1 shows the correlation coefcients between winter visibility and meteorological factors at the Mianyang Airport, and the analysis results reveal that the correlation coefcients between visibility and relative humidity, dew point temperature, and wind speed are high.Among all factors, relative humidity has a strong negative correlation with visibility, and the correlation coefcient is −0.76.Visibility and dew point temperature are strongly adversely connected (the correlation coefcient: −0.55), while visibility and wind speed are signifcantly positively correlated (the correlation coefcient: 0.33).
Relative humidity has a strong negative correlation with visibility, and with the increase in relative humidity, visibility gradually decreases.Tis is because with the increase in relative humidity, the radius of wet particles increases, the extinction increases, and the visibility decreases.Dew point temperature has a strong negative correlation with visibility, and with the increase in dew point temperature, visibility gradually decreases.Tis is because the dew point temperature is high and the water vapor content is high, which will lead to a decrease in visibility.Relative humidity and dew point temperature have the greatest correlation with visibility, which also shows that the change in visibility in this airport is most infuenced by these two factors and it is also the focus of visibility forecast.Other factors do not change much, so the correlation is also small.Figure 2 presents the correlation coefcients between visibility in winter at the Mianyang Airport and environmental factors in four diferent monitoring sites in Mianyang city.Te analysis shows that the environmental factors of the Municipal People's Congress are more indicative of the visibility of Mianyang Airport, and the Municipal People's Congress is located 3.09 km to the northeast of the Mianyang Airport.To be specifc, the following indexes have a strong correlation with visibility at the Municipal People's Congress: AQI has a signifcant negative correlation with visibility and the correlation coefcient is −0.59;PM 2.5 concentration has a signifcant negative Advances in Meteorology correlation with visibility and the correlation coefcient is −0.61;PM 10 concentration has a signifcant negative correlation with visibility and the correlation coefcient is −0.54.Visibility correlates more signifcantly with PM 2.5 concentrations than with PM 10 concentrations.Visibility is also signifcantly inversely correlated with CO, with a correlation coefcient of −0.5.
Figure 3 is a scatter plot for the relationships between visibility and various meteorological factors.Figure 3(a) shows that the relationship between visibility and relative humidity can be represented by y � −0.0031x + 97.568 (y is visibility and x is relative humidity).As the relative humidity increases, the visibility gradually decreases, and the relative humidity directly improves the scattering efciency by affecting the moisture absorption growth of the particles, thus reducing the visibility.Figure 3(b) shows that with the increase in dew point temperature, which indicates the increase in water vapor content, visibility gradually decreases, and the ftting relationship can be represented by y � −0.0006x + 5.8168 (y is visibility and x is dew point temperature).Figure 3(c) demonstrates that as the wind speed increases, the better the difusion conditions become, and the visibility gradually increases.Lower wind speeds do not promote the dispersion of pollutants and result in lower visibility.When the wind speed is less than or equal to 2 m/s (that is, a small wind speed in the traditional sense), the visibility is generally less than 10 km, and when the wind speed is greater than 3 m/s, the visibility is better.
Figure 3(d) shows that AQI and visibility can be ftted through the following relationship formula: y = 157.68e−1E−4x (y is visibility and x is AQI index), and R square is 0.3937, indicating that the ftting efect is great.It can also be found from the fgure that as the AQI index increases, the visibility decreases.When the AQI is less than 100, the visibility slowly decreases as the index increases, and conversely, when the AQI is more than 100, the visibility decreases drastically as the index increases.Figure 3(e) demonstrates that the relationship between PM 2.5 and visibility can be represented through the following formula: y = 122.59e−1E−4x (y is visibility and x is PM 2.5 concentration), and R square is 0.4208, indicating that the ftting efect is great.It can also be found from the fgure that as the PM 2.5 concentration increases, the visibility decreases.Excessive PM 2.5 concentration may lead to low visibility.When the PM 2.5 concentration is less than 70 μg/m³, the visibility slowly decreases as the concentration increases, and conversely, when the PM 2.5 concentration is more than 70 μg/ m³, the visibility decreases drastically as the concentration increases.Figure 3(f ) shows that PM 10 and visibility can be ftted through the following relationship formula: y = 163.09e−1E−4x (y is visibility and x is PM 10 concentration).It can also be found from the fgure that as the    Advances in Meteorology PM 10 concentration increases, the visibility decreases.When the PM 10 concentration is less than 100 μg/m³, the visibility slowly decreases as the concentration increases, and conversely, when the PM 10 concentration is more than 100 μg/m³, the visibility decreases drastically as the concentration increases.Figure 3(g) shows that the relationship between CO and visibility can be represented by y = 4E − 9x 2 − 0.0001x + 1.6092 (y is visibility and x is CO concentration).It can also be found from the fgure that visibility below 10 km decreases as the CO concentration increases.

Multiple Linear Regression Models and Tests.
Based on the results of the correlation analysis between visibility and each factor, the multivariate linear regression model is used to establish the visibility grading forecast equation, and after repeated introduction and deletion of parameters, six parameters are selected for modeling optimization for corresponding visibility level, and the visibility grading forecast model is obtained (Table 2), where Y is the forecast value of visibility, X 1 is AQI, X 2 is PM 2.5 , X 3 is PM 10 , X 4 is relative humidity, X 5 is the dew point temperature, and X 6 is the wind speed.
A comparative analysis of forecast and observation was carried out by dint of the visibility grading forecast model.As can be seen in Figure 4, the trend of the forecast and that of the observation are consistent, indicating that the visibility grading forecast model has a better simulation efect.
In order to objectively evaluate the visibility forecast results, based on visibility observations, the forecast accuracy rate P C , the underreport rate P O , and the false report rate F AR are calculated to evaluate the forecasting capability of the model.Te formulas are as follows: where N A is the correct number of forecasts, N B is the number of missing reports, and N C is the number of false reports.
Table 3 shows that the visibility grading forecast model has an accuracy of 75% for level-1 visibility forecasting, 70% accuracy for level-2 visibility forecasting, 75% accuracy for level-3 visibility forecasting, 84.6% accuracy for level-4 visibility forecasting, and 95.8% accuracy for level-5 visibility forecasting.Te rate of missing and false reports is less than 30%.Level-5 visibility forecasting is more accurate because level-5 visibility is better, with a value greater than 5,000 m.Tis kind of weather is generally relatively stable and it is less afected by the changes of meteorological factors and environmental factors.Level-2 visibility forecasting is the lowest, because level-2 visibility is relatively low, with the value between 800 m and 1,600 m.At this time, the weather is unstable, and it is in the stage of inversion layer destruction, which is greatly infuenced by the changes in meteorological factors and environmental factors, so the forecast is more difcult and the forecast is the lowest.
Based on the multivariate linear regression visibility grading forecast model, the forecast efect is good, and it is obviously better than the nongrading model.In the experiment, the visibility nongrading forecast model has an accuracy of 16.7% for level-1 visibility forecasting, 20% accuracy for level-2 visibility forecasting, 16.7% accuracy for level-3 visibility forecasting, 55.9% accuracy for level-4 visibility forecasting, and 90.5% accuracy for level-5 visibility forecasting.So, the visibility grading forecast model has a higher accuracy and it can provide a reference for visibility grading forecasting, and this is also a basic method for the interpretation and application of current numerical weather prediction.

KNN Model and Test.
Due to the chaotic and nonlinear nature of atmospheric motion, it is difcult to accurately describe and simulate weather models with one equation or a set of simple linear or nonlinear regression equations.In this paper, we also try to use the KNN algorithm to establish a grading visibility forecasting model.
In this paper, the K value is determined by crossvalidation.Table 4 shows the classifcation accuracy when K � 3, 4, and 5. Te results demonstrate that when K � 3 or K � 5, the cross-validation accuracy rate of visibility forecasting is 70%, and the rate is only 63.5% when K � 4. When K � 3 or K � 5, the accuracy rate is better, in comparison to the rate when K � 4. Terefore, K � 3 is used to build the KNN model.
To understand the predictive performance of the KNN classifer on the visibility of each level, Table 5 further shows the cross-validation results of the KNN classifer for each level of visibility classifcation.Te horizontal axis in the table represents the forecast of visibility at all levels, the vertical axis represents the observation of visibility at all levels, and the cross line represents the forecast accuracy.
It can be seen from the table that the KNN classifer performs better in high-visibility weather.Te accuracy rate is as high as 75% for the observation visibility greater than 5,000 m (level 5), and as high as 77.8% for observation visibility greater than 2,000 m but not greater than 5,000 m (level 4).In medium-visibility weather, the KNN classifer forecasts a one-level lower visibility.When the observation visibility is level 3, the forecast is level 2, so we can conclude that the visibility forecast value is lower than the observation.Te KNN classifer performed very well in low-to-mediumvisibility weather.When the observation visibility is greater than 800 m but not greater than 1,600 m (level 2), the accuracy rate reaches as high as 100%.Te KNN classifer performs poorly in low-to-medium-visibility weather.When the observation visibility is not greater than 800 m (level 1), the accuracy rate is only 33.3%.
Te possible reason for poor calculations for lowvisibility weather is that there are many infuencing factors in low-visibility weather.(1) In the vertical space, in addition to infuencing factors such as temperature, 6 Advances in Meteorology pressure, humidity, wind, and AQI, the temperature profle within the boundary layer is also a signifcant infuencing factor.For example, fog and heavy pollution weather at airports are often accompanied by the phenomenon of "temperature inversion," which puts a lid on the vertical difusion of pollutants.(2) In the horizontal space, in addition to local pollutants and water vapor and other infuencing factors, pollutants or water vapor transportation in the surrounding area may also have an impact on local visibility.(3) From the perspective of the time series, the continuous accumulation of pollutants or water vapor also has a greater infuence on the low-visibility weather.For instance, on day 1, day 2, and day 3, all meteorological factors (high humidity and small wind) are the same, but the visibility gradually decreases as time goes by.What was not taken into account in the analysis include temperature inversions in the vertical space, contaminant transport in the horizontal direction, and cumulative efects in terms of time series, which may become possible causes of unsatisfactory forecasts for the low-visibility weather.5).Te daily variation of average visibility at the Mianyang Airport is obvious, with the minimum value of daily visibility appearing at 9 : 00 (Beijing time, the same as below) and the maximum value appearing at 17 : 00.Te average low value of visibility appeared at 4 : 00 to 9 : 00 in the morning.Te visibility began increasing slowly from 9 : 00, reached the maximum value at 17 : 00, and then decreased slowly.Te weather that afects visibility in Mianyang Airport in winter is mainly radiation fog.Radiation fog has obvious daily variation, which mainly occurs in the morning and generally lasts until noon and afternoon.Te formation of radiation fog is mainly due to the ground radiation cooling at night, so it is easier to produce radiation fog in the morning, resulting in low visibility.As the sun rises, the intensity of solar radiation increases, the ground temperature increases, the cooling efect of ground radiation begins to weaken, the dew point defcit increases, the water vapor content near the ground decreases, the intensity of radiation fog decreases, and the visibility is greatly improved.At the same time, the height of the inversion layer rises further, which is benefcial to the difusion of water vapor in the lower layer, and also reduces the relative humidity and increases the visibility.Terefore, the visibility at noon is generally greater than that in the morning, until it reaches its maximum at 17 : 00.
Te visibility at all levels in winter was analyzed (Figure 6).Days of level-5 visibility were the least, which accounted for 5% of the total.Te proportion of days with level-4 visibility was 31%, which is the highest, followed by days with level-3 visibility at 29%, days with level-2 visibility at 24%, and days with level-1 visibility at 11%.

Advances in Meteorology
Since level 1 is the most important level for fight training, so we analyzed the occurring and disappearing of the visibility below 800 m, and the period of visibility that is most likely to develop or fade in a day, which lays a groundwork for the visibility forecast below 800 m.
As can be seen from Figure 7, visibility below 800 m mainly occurred and developed before 09 : 00, and the occurrence frequency is the highest at 03 : 00, indicating that low visibility tends to occur in the early morning.Tere are three peak periods (01 : 00, 03 : 00, and 07 : 00) for visibility occurrence below 800 m.As for the visibility disappearing time below 800 m, 08 : 00 to 15 : 00 is the peak period for visibility to diminish and dissipate, with 09 : 00 being the time for most dissipation, and most of low visibility dissipates before 15 : 00.
Visibility below 800 m occurs 24 times in total, and its duration is analyzed (Figure 8).Te longest duration was 14 hours, which occurred on December 8, 2016, followed by 13 hours on December 4, 2016.In December 2016, low visibility lasted longer and occurred in more days in December 2016 than in other months.

Conclusions
In this paper, the relationship between winter visibility and meteorological factors and environmental factors is studied.Te authors construct a visibility grading forecast model in winter by dint of the multiple linear regression and the KNN algorithm based on big data mining technology, and perform the testing and comparative analysis.At the same time, the visibility characteristics in winter at the Mianyang Airport are analyzed.Te conclusions are drawn as follows: (1) Mianyang Airport's winter visibility has a signifcant correlation with relative humidity, dew point temperature, wind speed, AQI, PM Note that, the accuracy rate of level-2 visibility is 100%.However, the forecast performance is poor in the case of low visibility (level 1).In practice, the two models can complement each other to further increase the accuracy of the forecast.(3) At the Mianyang Airport, the minimum value of average daily visibility in winter appeared at 09 : 00,  Advances in Meteorology the maximum value appeared at 17 : 00, and the average low value of visibility appeared from 04 : 00 to 09 : 00 in the morning.According to the statistical analysis of visibility at all levels in winter, level-5 visibility days account for the least proportion of the total, and most are level-4 visibility days.Level-1 visibility occurred and developed before 09 : 00, 01 : 00, 03 : 00, and 07 : 00 and these were the peak time for the occurrence of level-1 visibility, and 08 : 00-15 : 00 was the peak period for the visibility to diminish and dissipate.Te maximum duration of level-1 visibility was 14 hours.
In the future, we will also work in the following areas: frst, collecting vertical space data and including the temperature inversion into forecast factors; second, taking into account the transport, convergence, and dispersion of surrounding pollutants and water vapor; third, considering the cumulative efect of pollutants and water vapor, in hope of improving the efects of the KNN classifer and increasing the forecast accuracy.

Figure 4 :
Figure 4: Forecast and observation comparison through the visibility grading forecast model.

Figure 7 :
Figure 7: Occurrence time-frequency (a) and disappearing time-frequency (b) of visibility below 800 m.
Te height of wind speed data is 10 meters above the ground.

Table 3 :
Visibility grading forecast model test.

Table 5 :
Cross-validation results of KNN classifers for all levels of visibility (unit: %).
Tis indicates that low visibility does not occur very often, but it is the most critical level afecting fight training.
2.5 concentration, PM 10 concentration, and CO.It has a signifcant positive correlation with wind speed, and a signifcant negative correlation with relative humidity, dew point temperature, AQI, PM 2.5 concentration, PM 10 concentration, and CO, of which relative humidity has the largest impact on the visibility with a correlation coefcient of −0.76.Te correlation coefcient of dew point temperature is −0.55, the correlation coefcient of wind speed is 0.33, the correlation coefcient of AQI at the Municipal People's Congress is −0.59, the correlation coefcient of PM 2.5 concentration is −0.61, the correlation coefcient of PM 10 concentration is −0.54, and the correlation coefcient of CO is −0.5.(2)Te multivariate linear regression model and the KNN model were adopted to conduct grading forecasting experiments on visibility, respectively, and the results showed that both models can be used for visibility grading forecasts, but the forecast efect for diferent levels of visibility is diferent.Te multiple regression model can provide great level-1, level-2, level-3, level-4, and level-5 visibility forecasts, with an accuracy rate of more than 70%.Te KNN model has a better grading accuracy rate at K � 3 or K � 5 than at K � 4, and the KNN model can better forecast level-2, level-4, and level-5 visibility.