the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 26 May 2020
Developing a novel hybrid model for the estimation of surface 8 h ozone (O3) across the remote Tibetan Plateau during 2005–2018
We developed a two-stage model called the random-forest–generalised additivemodel (RF–GAM), based on satellite data, meteorological factors, and othergeographical covariates, to predict the surface 8 hO3 concentrationsacross the remote Tibetan Plateau. The 10-fold cross-validation resultsuggested that RF–GAM showed excellent performance, with the highestR2 value (0.76) and lowest root-mean-square error (RMSE)(14.41 µg m−3), compared with other seven machine-learning models. Thepredictive performance of RF–GAM showed significant seasonaldiscrepancy, with the highestR2 value observed in summer (0.74),followed by winter (0.69) and autumn (0.67), and the lowest one in spring(0.64). Additionally, the unlearning ground-observedO3 data collectedfrom open-access websites were applied to test the transferring ability of thenovel model and confirmed that the model was robust in predicting the surface8 hO3 concentration during other periods (R2=0.67, RMSE = 25.68 µg m−3). RF–GAM was then used to predict the daily 8 hO3 level over the Tibetan Plateau during 2005–2018 for the first time. Itwas found that the estimatedO3 concentration displayed a slow increase,from64.74±8.30 µg m−3 to66.45±8.67 µg m−3 from 2005 to 2015, whereas it decreased from the peak to65.87±8.52 µg m−3 during 2015–2018. Besides this, the estimated 8 hO3 concentrations exhibited notable spatial variation, with the highestvalues in some cities of the northern Tibetan Plateau, such as Huangnan (73.48±4.53 µg m−3) and Hainan (72.24±5.34 µg m−3), followed by the cities in the central region, including Lhasa(65.99±7.24 µg m−3) and Shigatse (65.15±6.14 µg m−3), and the lowestO3 concentration occurred in a city ofthe southeastern Tibetan Plateau called Aba (55.17±12.77 µg m−3).Based on the 8 hO3 critical value (100 µg m−3) provided bythe World Health Organization (WHO), we further estimated the annual meannonattainment days over the Tibetan Plateau. It should be noted that most of thecities on the Tibetan Plateau had excellent air quality, while severalcities (e.g. Huangnan, Haidong, and Guoluo) still suffered from more than40 nonattainment days each year, which should be given more attention in order toalleviate localO3 pollution. The results shown herein confirm that thenovel hybrid model improves the prediction accuracy and can be applied toassess the potential health risk, particularly in remote regions withfew monitoring sites.
- Article
(4889 KB) - Full-text XML
- Supplement
(2090 KB) - BibTeX
- EndNote
Along with the rapid economic development and urbanisation, theanthropogenic emissions of nitrogen oxides (NOx) and volatile organiccompounds (VOCs) displayed high-speed growth. The chemical reactions betweenNOx and VOCs in the presence of sunlight triggered ambient ozone(O3) formation (Wang et al., 2017, 2019). As a strongoxidant, ambientO3 could play a negative role in human health throughaggravating the cardiovascular and respiratory function (Ghude et al., 2016;Marco, 2017; Yin et al., 2017a). Apart from the effect on human health,O3 also posed a great threat to vegetation growth (Emberson et al., 2018;Feng et al., 2015, 2019; Qian et al., 2018). Moreover, thetroposphericO3 can perturb the radiative energy budget of theearth–atmosphere system, as it is the third most important greenhouse gas next tocarbon dioxide (CO2) and methane (CH4), thereby changing theglobal climate (Bornman et al., 2019; Fu et al., 2019; Wang et al., 2019).Recently, the particulate matter with a concentration of less than 2.5 µm (PM2.5)showed a persistent decrease, while theO3 issue hasbeen increasingly prominent in China (Li et al., 2017b, 2019b).Therefore, it was critical to accurately reveal the spatiotemporal variation inO3 pollution and assess its heath risk in China.
A growing body of studies began to investigate the spatiotemporal variation in theO3 level worldwide. Wang et al. (2014b) demonstrated that the 8 hO3 concentrations in nearly all of the provincial cities experiencedremarkable increases during 2013–2014. Following this work, Li et al. (2017b) reported that the annual meanO3 concentration over Chinaincreased by 9.18 % during 2014–2016. In other Asian countries exceptChina, Vellingiri et al. (2015) performed long-term observation and foundthat theO3 concentration in Seoul, South Korea, has displayed a gradualincrease in recent decades. In the southeastern United States, Li et al. (2018) observed that the surfaceO3 concentration has displayed a gradualdecrease in the past 10 years. Although the number of ground-levelmonitoring sites has been increasing globally, the limited monitoring sitesstill cannot accurately reflect the fine-scaleO3 pollution statusbecause each site shows little spatial representativeness(0.25–16.25 km2) (Shi et al., 2018). Furthermore, the number of monitoring sites inmany countries (e.g. China and the United States) displays an unevendistribution characteristic at the spatial scale. In China, most of thesesites are concentrated in the North China Plain (NCP) and Yangtze River Delta (YRD), whilewestern China has an extreme lack of ground-levelO3 data, which oftenincreases the uncertainty of health assessment. Therefore, many studies usedvarious models to estimate theO3 concentrations without monitoringsites. Chemical transport models (CTMs) were often considered to be the typicalmethods to predict the surfaceO3 level. Zhang et al. (2011) employedthe GEOS-Chem model to simulate the surfaceO3 concentration over theUnited States, suggesting that the model could capture the spatiotemporalvariation in surfaceO3 concentration at a large spatial scale. Lateron, Wang et al. (2016) developed a hybrid model called land use regression(LUR) coupled with CTMs to predict the surfaceO3 concentration in theLos Angeles Basin, California. In recent years, these methods were alsoapplied to estimate the surfaceO3 level over China. Liu et al. (2018) used the Community Multiscale Air Quality (CMAQ) model to simulate thenationwideO3 concentration over China in 2015. Nonetheless, thehigh-resolutionO3 prediction using CTMs might have widely deviated fromthe measured value, owing to the imperfect knowledge about the chemicalmechanism and the higher uncertainty of the emission inventory. Moreover, thecontinuous emission data ofNOx and VOCs were not always open access,which restricted the long-term estimation of the surfaceO3 concentrationusing CTMs.
Fortunately, the daily satellite data enable the fine-scale estimations of theO3 level at a regional scale due to broad spatial coverage and hightemporal resolution (McPeters et al., 2015). Shen et al. (2019) confirmedthat the satellite-retrievedO3 column amount could accurately reflect thespatiotemporal distribution of the surfaceO3 level. Therefore, somestudies tried to use traditional statistical models coupled withhigh-resolution satellite data to estimate the ambientO3 level.Fioletov et al. (2002) used the satellite measurement to investigate theglobal distribution ofO3 concentrations based on a simple linear model.Recently, Kim et al. (2018) employed the integrated empirical geographicregression method to predict the long-term (1979–2015) variation in ambientO3 concentration over the United States based onO3 column amountdata. Although the statistical modelling of ambientO3 concentration iswidespread around the world, most of the traditional statisticalmodelling only utilised the linear model to predict the ambientO3concentration, which generally decreased the prediction performance becausethe nonlinearity and high-order interactions betweenO3 and predictorscannot be managed by a simple linear model.
As an extension of traditional statistical model, machine-learning methodshave been widely applied to estimate the pollutant levels in recent yearsbecause of their excellent predictive performances. Among these machine-learning algorithms, decision tree models such as random forest (RF) andextreme gradient boosting (XGBoost) generally showed fast training speed andexcellent prediction accuracy (Li et al., 2020; Zhan et al., 2018).Furthermore, decision tree models can obtain the contribution of eachpredictor to air pollutants, which was beneficial to the parameter adaptionand model optimisation. Chen et al. (2018b) has firstly employed the RF model tosimulate the PM2.5 level in China since 2005. Following this work, werecently used the XGBoost model to estimate the 8 hO3 concentration onthe island of Hainan for the first time and captured the moderate predictiveperformance (R2=0.59) (Li et al., 2020). While the decision tree modelshowed many advantages in predicting the pollutant level, the spatiotemporalautocorrelation of pollutant concentration was not a concern in thesestudies. Li et al. (2019a) confirmed that the prediction error by the decisiontree model varied greatly with space and time. Thus, it is imperative toincorporate the spatiotemporal variables into the original model to furtherimprove the performance. To resolve the defects of decision tree models,Zhan et al. (2018) developed a hybrid model called RF-spatiotemporal Kriging(STK) to predict theO3 concentration over China and achievedbetter performance (overall –R2=0.69; southwestern China –R2=0.66). Unfortunately, the RF–STK model still showed some weaknesses inpredictingO3 concentration. First of all, the predictive performanceof the STK model was strongly dependent on the number of monitoring sitesand their spatial densities. The model often showed worse predictiveperformance in regions with few monitoring sites (Gao et al., 2016).Moreover, the ensemble model cannot simulate theO3 level during theperiods without ground-level-measured data. In contrast, the generalised additivemodel (GAM) not only considers the time autocorrelation ofO3concentration but also shows better extrapolation ability (Chen et al.,2018a; Ma et al., 2015). Thus, the ensemble model of RF and GAM was proposedto predict the spatiotemporal variation in the surface 8 hO3concentration.
The Tibetan Plateau, the highest plateau in the world, shows highersurface solar radiation compared with the regions outside the plateau. It waswell documented that high solar radiation tended to generate a large amount ofthe OH radical, resulting in theO3 formation via the reaction of VOC andthe OH radical (Ou et al., 2015). While the totalO3 column amount on the Tibetan Plateau has displayed a slight decrease since the 1990s, the convergentairflow formed by subtropical anticyclones could bring ozone-rich airsurrounding the plateau to the low atmosphere (Lin et al., 2008), therebyleading to a higher surfaceO3 concentration over the plateau. Moststudies focused on the stratosphere–troposphere transport ofO3 on the Tibetan Plateau, whereas limited effort was given to investigatingthe ground-levelO3 level over this region. To date, only several studieswere concerned with the spatiotemporal variation in the surfaceO3concentration in this region based on field-observation data (Chen et al.,2019; Shen et al., 2014; Yin et al., 2017b). Unfortunately, the fewmonitoring sites on the Tibetan Plateau cannot capture the realO3 pollutionstatus, especially in remote areas (e.g. the northern part of the Tibetan Plateau), because each site only possessed limited spatialrepresentativeness. Apart from these field measurements, Liu et al. (2018) (R=0.60) and Zhan et al. (2018) (R2=0.66) used CTMs and the machine-learning model to simulate the surfaceO3 concentration over China in2015, respectively. Both of these studies included the predictedO3level on the Tibetan Plateau. Although they finished the pioneering work,the predictive performances of both studies were not excellent.Therefore, it was imperative to develop a higher-quality model to enhancethe modelling accuracy.
Here, we developed a new hybrid-method (RF–GAM) model integrating satellitedata, meteorological factors, and geographical variables to simulate thegridded 8 hO3 concentrations over the Tibetan Plateau for the first time.Based on the estimated surfaceO3 concentration, we clarified thelong-term variation (2005–2018) of the surfaceO3 concentration andquantified the key factors for the annual trend. Filling the gap ofstatistical estimation of the 8 hO3 level in a remote region, this studyprovides useful datasets for epidemiological studies and air qualitymanagement.
2.1 Study area
The Tibetan Plateau is located in southwestern China, which ranges from 26.00 to39.58∘ N and from 73.33 to 104.78∘ E.The Tibetan Plateau is surrounded by the Taklamakan Desert to the north and Sichuan Basinto the southeast. The land area of the Tibetan Plateau reaches2.50×106 km2 (Chan et al., 2006). Based on the air circulation pattern, the Tibetan Plateaucan be roughly classified into the monsoon-influenced region and thewesterly-wind-influenced region (Wang et al., 2014a). The annual mean airtemperature in most regions is below 0 ∘C. The annual meanrainfall amount on the Tibetan Plateau ranges from 50 to 2000 mm. The terrainconditions are complex, and higher altitudes are concentrated in the central region.The Tibetan Plateau is generally treated as a remote region lacking inanthropogenic activity, and most of the residents are concentrated in the southeastern andsouthern parts of the Tibetan Plateau. The Tibetan Plateau consists of 19prefecture-level cities, and their names and corresponding geographicallocations are shown in Figs. 1 and S1.
Figure 1The geographical locations and annual mean 8 hO3concentrations in the ground-observed sites (red dots) over the Tibetan Plateauduring 2014–2018. The elevation data are collected from geographical andspatial data cloud at a 30 m spatial resolution.
2.2 Data preparation
2.2.1 Ground-level 8 hO3 concentration
The daily 8 hO3 data in 37 monitoring sites over the Tibetan Plateau from13 May 2014 to 31 December 2018 were collected from the national airquality monitoring network. TheO3 levels in all of these sites weredetermined using an ultraviolet-spectrophotometry method. The highest 8 hmoving averageO3 concentration each day was calculated as the daily8 hO3 level after data quality assurance. The data quality of all themonitoring sites was assured on the basis of the HJ 630-2011 specifications.The data with no more than two consecutive hourly measurements missing in allthe days were treated as the valid data.
2.2.2 Satellite-retrievedO3 column amount
TheO3 column amounts (DU: total molecules cm−2) during 2005–2018 were downloaded from theOzone Monitoring InstrumentO3 (OMIO3) level-3 data with a0.25∘ spatial resolution from the website of the National Aeronauticsand Space Administration (NASA) (https://acdisc.gsfc.nasa.gov/data/Aura_OMI_Level3/OMDOAO3e.003/, last access: 19 May 2020). The OMIO3product shows global coverage and traverses the earth once a day. TheO3 column amount with a cloud radiance fraction >0.5, terrainreflectivity >30 %, and solar zenith angles >85∘ should be removed. In addition, the cross-track pixelssignificantly influenced by the row anomaly should be deleted.
2.2.3 Meteorological data and geographical covariates
The daily meteorological data were obtained from ERA-Interim datasets with0.125∘ resolution. These meteorological data consisted of the 2 m dew-point temperature (d2m), 2 m temperature (t2m), 10 mU windcomponent (u10), 10 mV wind component (v10), boundary layer height(blh), sunshine duration (sund), surface pressure (sp), and totalprecipitation (tp). The 30 m resolution elevation data (DEM) were downloadedfrom the China Resource and Environmental Science Data Center (CRESDC). The dataof the gross domestic product (GDP) and population density with 1 kmresolution were also extracted from CRESDC. Population density and GDP in2005, 2010, and 2015 were integrated into the model to predict the surface8 hO3 concentration over the Tibetan Plateau because these data wereavailable every 5 years. Additionally, the land use data of 30 mresolution (e.g. water, grassland, urban, forest) were also extracted fromCRESDC. Lastly, the latitude, longitude, and time were also incorporatedinto the model.
All of the explanatory variables collected were resampled to 0.25∘ × 0.25∘ grids to predict theO3 level. The originalmeteorological data with 0.125∘ resolution were resampled tothe 0.25∘ grid. The land use area, elevation, GDP, and populationdensity in each grid were calculated using spatial clipping. Lastly, all ofthe predictors were integrated into an intact table to train the model.
2.3 Model development and assessment
The RF–GAM model was regarded as the hybrid model of RF and GAM. The RF–GAMmodel was a two-stage model in which the prediction error estimated by the RFmodel was then simulated by GAM. The prediction results of RF and GAM weresummed as the final result of the RF–GAM model (Fig. 2). The detailed equationis as follows:
whereZ(s, t) is the estimated 8 hO3 level at the locations and timet,P(s, t) represents the 8 hO3 concentration predicted by the RFmodel, andE(s, t) denotes the prediction error by GAM.
Figure 3Density scatterplots of model fitting and cross-validation results ata daily level. Panels(a),(b), and(c) represent RF–GAM, RF–STK, and RF models,respectively. The red dotted line denotes the fitting linear-regressionline. MPE, RMSE, and RPE are mean prediction error (µg m−3), root-mean-square error (µg m−3), and relativepercentage error (%), respectively.
In the RF model, a large number of decision trees were planted based on thebootstrap sampling method. At each node of the decision tree, the randomsamples of all predictors were applied to determine the best split amongthem. Following the procedure, a simple majority vote was employed topredict the 8 hO3 level. The RF model avoided a priori linear assumptionofO3 concentration and predictors, which was often not in goodagreement with the actual state. The RF model has two key parameters, includingntree (the number of trees grown) andmtry (the number ofexplanatory variables sampled for splitting at each node). The predictionperformance of the RF model was strongly dependent on the two parameters.The optimalntree andmtry were determined based on the leastout-of-bag (OOB) errors. Based on the iteration result, the optimalntree andmtry reached 500 and 5, respectively. Besides this, thebackward variable selection method was performed on the RF submodel toachieve better performance. At each step of the predictor selection, thevariable with the least important value was excluded from the next step.This one-variable-at-a-time exclusion method was repeated until only twoexplanatory variables remained in the submodel. Finally, all of the selectedvariables except the area of water were integrated into the model toachieve the best prediction performance. The detailed RF model is asfollows:
whereO3 denotes the observed 8 hO3 level in the monitoring site;theO3 column represents theO3 column amount in the correspondinggrid; Elevation denotes the corresponding elevation of the site; and Agr, Urban,Forest, and Grassland are the agricultural land, urban land, forest land, andthe grassland, respectively. Population represents the population density inthe corresponding site. Prec,T, WS,P,tsun, and RH are precipitation, airtemperature, wind speed, air pressure, sunshine duration, and relativehumidity, respectively. Additionally, another five models, including the RF,generalised regression neutral network (GRNN), backward-propagation neuralnetwork (BPNN), Elman neural network (ElmanNN), and extreme learning machine(ELM), also used the backward variable selection method. TheR2 valuewas treated as an important parameter for adding or reducing the variable. Thevariable should be removed when theR2 value of the submodel showed aremarkable decrease with the integration of this variable. Lastly, theoptimal variable group was applied to establish the submodel.
Following the RF submodel, the prediction error estimated by the RF submodelwas further modelled by the GAM. GAM could reflect the time autocorrelationof the predictive error of RF model, and thus the ensemble model of RF and GAMmight decrease the modelling error of the one-stage model. All of the variableswere incorporated into the models to establish the second-stage model, andthe backward variable selection was also used to determine the optimalvariable group.
The 10-fold cross-validation (CV) technique was employed to evaluate thepredictive performances for all of the machine-learning models. All of thetraining datasets were randomly classified into 10 subsets uniformly. Ineach round of validation, nine subsets were used to train, and the remainingsubset was applied to test the model performance. The process was repeated10 times until every subset has been tested. Some statistical indicators,including theR2, root-mean-square error (RMSE), mean prediction error(MPE), relative percentage error (RPE), and the slope, were calculated toassess the model performance. The optimal model with the best performancewas used to estimate the 8 hO3 concentration in recent decades.
Table 1TheR2 values, RMSE, MPE, and RPE of RF–GAM in different yearsduring 2014–2018 over the Tibetan Plateau.
Figure 4The transferring-ability validation of RF–GAM method based on themeasured daily 8 hO3 concentration during December 2013–May 2014.
3.1 The validation of model performance
Figures 3 and S2 show the density scatterplots of the fitting and10-fold cross-validation results for eight machine-learning models forChina. The 10-fold cross-validationR2 values followed the order ofRF–GAM(R2=0.76)>RF–STK(R2=0.63)>RF(R2=0.55)>GRNN(R2=0.53)>BPNN(R2=0.50)>XGBoost(R2=0.48)>ElmanNN(R2=0.47)>ELM(R2=0.32). The RMSE values of RF–GAM, RF–STK, RF, GRNN, XGBoost, BPNN,ElmanNN, and ELM were 14.41, 17.79, 19.13, 19.41, 20.73, 20.06, 20.61, and23.36 µg m−3, respectively. Both the MPE and RPE showed similarcharacteristics to RMSE of the order of RF–GAM (10.97 µg m−3 and26.50 %) < RF–STK (13.48 µg m−3 and 35.15 %) < RF (14.71 µg m−3 and 35.51 %) < GRNN (14.89 µg m−3 and 35.82 %) < BPNN (15.43 µg m−3 and36.19 %) < ElmanNN (15.75 µg m−3 and 37.05 %) < XGBoost (15.80 µg m−3 and 38.13 %) < ELM (18.23 µg m−3 and 44.05 %) (Figs. 3 and S2). Besides this, the slope of theRF–GAM model was closer to 1 compared with other models. It was welldocumented that the RF model generally showed better performance thanother models because this method did not need to define complexrelationships between the explanatory variables and theO3concentration (e.g. linear or nonlinear). Furthermore, the variableimportance indicators calculated by the RF model can help the user todistinguish the key variables from the noise ones and make full use of thestrength of each predictor to assure the model robustness. Although BPNN,GRNN, XGBoost, ElmanNN, and ELM have been widely applied to estimate the airpollutant concentrations (Chen et al., 2018c; Zang et al., 2018; Zhu et al.,2019), these methods suffered from some weaknesses in predicting thepollutant level. For instance, both the BPNN and ElmanNN models could capturethe locally optimal solution when the training subsets were integrated intothe final model, which decreased the predictive performance of the model(Wang et al., 2015). Moreover, BPNN generally showed slow training speed,especially with the huge training subsets (Li and Park, 2009; Wang et al.,2015). ELM often consumed more computing resources and experienced theoverfitting issue due to the increase in sampling size (Huang et al., 2015;Shao et al., 2015). The GRNN method advanced the training speed compared withthe BPNN model and avoided the locally optimal solution during the modellingprocess (Zang et al., 2019), whereas the predictive performance is stillworse than that of the RF model. XGBoost was often considered to be robust inpredicting the air pollutant level (Li et al., 2020), while the model did notdisplay excellent performance in the present study. This might beattributable to the sampling size in the present study not being big enoughbecause the model generally showed better performance with big samples.Moreover, we found that the two-stage model was superior to the one-waymodel in the predictive performance. This encouraging result suggested thatthe relationship between the predictors and the 8 hO3 concentrationvaried with space and time. The two-stage model used the GAM method tofurther adjust the prediction error of the RF model and considered thespatiotemporal correlation of the predictor error on the Tibetan Plateau. Althoughthe STK model incorporated space and time into the model simultaneously, theRF–GAM model outperformed the RF–STK model. It was assumed that the STKmodel showed higher uncertainty in predicting theO3 concentrationin regions with few sampling sites (Gao et al., 2016; Li et al.,2017a). Overall, the ensemble RF–GAM model showed significantimprovement in predictive performance.
Table 2TheR2 values, RMSE, MPE, and RPE of RF–GAM in four seasonsover the Tibetan Plateau.
Table 3The predictive performances of RF–GAM in different provinces overthe Tibetan Plateau.
Figure 6The mean value of estimated 8 hO3 concentration during2005–2018 over the Tibetan Plateau.
The performances of RF–GAM displayed slight differences for each year during2014–2018. As shown in Table 1, theR2 value showed the highest value(0.76) in 2016, followed by that in 2018 (0.75), 2017 (0.73), and 2015 (0.72),and showed the lowest one in 2014 (0.69). Both the RMSE and MPE exhibited the lowestvalues in 2014, while these parameters did not show significant variationduring 2015–2018. The lowestR2 value and the highest RPE were found in2014 due to having the lowest sample size, while the highestR2 value andlowest RPE in 2016 were due to the maximum sample size. Geng et al. (2018) found that the predictive performance of the machine-learning model wasstrongly dependent on the number of training samples and sampling frequency.The lower RMSE and MPE in 2014 might be attributable to the lack of measuredO3 data in spring, which decreased the higher value ofO3concentration. The performances of the RF–GAM model in four seasons werealso assessed by 10-fold cross validation (Table 2). The predictiveperformance of the RF–GAM model showed significant seasonal differences,with the highestR2 value observed in summer (0.74), followed by winter(0.69) and autumn (0.67), and the lowest one in spring (0.64). However, boththe RMSE and MPE displayed different seasonal characteristics from theR2 value. Both the RMSE and MPE for RF–GAM followed the order of spring(15.32 and 11.94 µg m−3) > summer (15.13 and 11.75 µg m−3) > winter (14.58 and 11.44 µg m−3) > autumn (13.23 and 10.52 µg m−3). The lowestR2value in spring might be caused by multipleO3 sources and complicateO3 formation mechanisms. On the one hand, theO3 in spring mightbe generated from the local anthropogenic emission or long-range transport(Li et al., 2017b, 2019b). On the other hand, a strongstratosphere–troposphere exchange process due to the lower height of the troposphereon the Tibetan Plateau might lead to the higherO3 concentration in spring(Skerlak et al., 2014). Unfortunately, both the long-range transport andstratosphere–troposphere exchange process were missing in the RF–GAM model,which restricted the accuracy ofO3 estimation in spring. The largeestimation errors (e.g. RMSE, MPE, and RPE) in spring and summer wereattributable to the high 8 hO3 concentration in these seasons, whilethe low prediction error observed in autumn was due to the lowO3 level.
Figure 7The inter-annual variation in predicted 8 hO3 level (µg m−3) from 2005 to 2018 across the Tibetan Plateau.
Apart from the seasonal variation, we also investigated the spatialvariabilities in the predictive accuracy for the RF–GAM model. The Tibetan Plateauwas classified into five provinces, and then the predictive performance ofRF–GAM model in each province was calculated. Among the five provinces,Gansu displayed the highestR2 value (0.74), followed bySichuan Province (0.71), Qinghai Province (0.70), the autonomous region of Tibet(0.69), and Yunnan Province (0.54) (Table 3). The results shown herein werenot in agreement with the previous studies by Geng et al. (2018), whoconfirmed that the predictive performance of the machine-learning model waspositively associated with the sampling size. It was assumed that thespatial distribution of the sampling sites in Tibet was uneven and thesampling density was low, though Tibet possessed the highest number of monitoringsites of the provinces. The prediction errors (RMSE and MPE)did not exhibit the same characteristics as theR2 value. The higherRMSE and MPE were found in the autonomous region of Tibet (14.81 and 11.24 µg m−3) and Qinghai Province (14.83 and 11.33 µg m−3) due tothe higher values of blh and sund. The lowest values of the RMSE and MPE couldbe observed in Yunnan Province, which was due to the higher rainfallamount. The highest RPE was found in Yunnan Province (25.85 %),followed by Tibet (22.90 %), Qinghai (22.65 %), and Sichuan (22.62 %), andthe lowest one was found in Gansu Province (22.51 %), which might be linked with thesample size.
Figure 8The trend analysis of predicted 8 hO3 concentration. Panels (a) and(b) represent the result of Mann–Kendall method and discrepancy of estimatedO3 level during 2005–2018 across the Tibetan Plateau.
Although 10-fold cross-validation verified that the RF–GAM model showedbetter predictive performance in estimating the surface 8 hO3concentration, the test method cannot validate the transferring ability ofthe final model. The monitoring site on the Tibetan Plateau before May 2014 isvery limited, and only the daily 8 hO3 data in Lhasa from the open-accesswebsite (https://www.aqistudy.cn/historydata/daydata.php?city=%E6%8B%89%E8%90%A8&month=2013-12, last access: 19 May 2020) were available tocompare with the simulated data. As depicted in Fig. 4, theR2 value ofthe unlearning 8 hO3 level against the predicted 8 hO3 concentrationreached 0.67, which was slightly lower than that of the 10-foldcross-validationR2 value. Overall, the extrapolation ability of theRF–GAM model is satisfactory, and thus it was assumed that the model couldbe applied to estimate theO3 concentration in other years. Both theRMSE and MPE for the unlearning 8 hO3 level against the predicted 8 hO3 concentration were significantly higher than those of the 10-foldcross validation. It was assumed that Lhasa showed a higher surface 8 hO3 concentration over the Tibetan Plateau.
To date, some previous studies also simulated the surfaceO3concentration on the Tibetan Plateau using statistical models (Zhan et al.,2018). For instance, Zhan et al. (2018) employed the RF–STK model toestimate the surfaceO3 concentration over China and explained the66 % spatial variability in theO3 level on the Tibetan Plateau. Apart fromthese statistical models, some classical CTMs were also applied to estimatetheO3 concentration in remote areas. Both Liu et al. (2018) andLin et al. (2018) used CMAQ to estimate theO3 level across China,while theR2 values in most of cities were lower than 0.50. In terms ofthe predictive performance, the RF–GAM model in our study showedsignificant advantages compared with previous studies. It should be notedthat our RF–GAM model could outperform most of current models, chieflybecause of (1) accounting for the temporal autocorrelation of the surfaceO3 concentration and (2) the use of high-quality satellite data.
Figure 9The seasonal variability in estimated 8 hO3 level acrossthe Tibetan Plateau. Panels(a–d) represent the predicted 8 hO3concentrations in spring, summer, autumn, and winter, respectively.
Figure 10The spatial distributions of nonattainment days on the Tibetan Plateauduring 2005–2018.
3.2 Variable importance
The results of variable importance for key variables are depicted in Fig. 5.In the final RF–GAM model, it was found that time was the dominant factorfor the 8 hO3 concentration on the Tibetan Plateau, indicating that theambientO3 concentration displayed significant temporal correlation.Following the time, meteorological factors served as the main factors fortheO3 pollution in remote regions. The sum of sund, sp, d2m, t2m,and tp constituted 34.43 % of the overall variable importance. Among others,sund was considered to be the most important meteorological factor for theO3 pollution. It was assumed that strong solar radiation and longduration of sunshine favoured the photochemical generation of ambientO3 (Malik and Tauler, 2015; Stähle et al., 2018). Tan et al. (2018)demonstrated that the chemical reaction betweenNOx and VOCs wasstrongly dependent on the sunlight. Besides this, the atmospheric pressure (sp)was also treated as a major driver for theO3 pollution over the Tibetan Plateau. Santurtún et al. (2015) have demonstrated that sp was closelylinked to the atmospheric circulation and synoptic-scale meteorologicalpattern, which could influence the long-range transport of ambientO3.Apart from sund and sp, d2m and t2m played significant role in theO3pollution, which was consistent with many previous studies (Zhan et al.,2018). Zhan et al. (2018) observed that cold temperatures were not favourableto theO3 formation. d2m can affect the surfaceO3 pollutionthrough two aspects. On the one hand, RH affected heterogeneous reactions ofO3 and particles (e.g. soot, mineral) (He et al., 2017; He and Zhang,2019; Yu, 2019). On the other hand, high RH could increase the soil moistureand evaporation, and thus the water-stressed plants tended to emit morebiogenic isoprene, thereby promoting the elevation ofO3 concentration(Zhang and Wang, 2016). It should be noted that the effect of precipitationonO3 pollution was weaker than those of othermeteorological factors. Zhan et al. (2018) also found a similar result andbelieved that rain scavenging served as the key pathway for theO3removal only whenO3 pollution was very serious. The effect of theO3column amount on surfaceO3 concentration seemed to be lower than thoseof most meteorological factors, suggesting that vertical transport ofambientO3 was complex. Although socioeconomic factors and land usetypes were not dominant factors for theO3 pollution on the Tibetan Plateau, they still cannot be ignored in the present study because thepredictive performance would worsen if these variables were excluded fromthe model. It was widely acknowledged that the anthropogenic emissionsfocused on developed urban areas with high population density, especiallyon the remote plateau (Zhang et al., 2007; Zheng et al., 2017). Comparedwith the urban land, the grassland played a more important role in theO3pollution on the Tibetan Plateau. It was thus assumed that the grassland waswidely distributed on the Tibetan Plateau, which could release a large amount ofbiogenic volatile organic compounds (BVOCs) (Fang et al., 2015). It was wellknown that photochemical reactions of BVOCs andNOx in the presence ofsunlight caused theO3 formation (Calfapietra et al., 2013; Yu et al.,2006). Furthermore, Fang et al. (2015) confirmed that BVOC emission on the Tibetan Plateau displayed a remarkable increase in the wet seasons.
3.3 The spatial distribution of estimated 8 hO3 concentration over the Tibetan Plateau
Figure 6 depicts the spatial distribution of the 8 hO3 level estimatedby the novel RF–GAM model. The spatial distribution pattern modelled by theRF–GAM model showed a similar characteristic to the results found byprevious studies except on the northern Tibetan Plateau (Liu et al., 2018). Theestimated 8 hO3 concentration displayed the highest value in somecities of the northern Tibetan Plateau, such as Huangnan (73.48±4.53 µg m−3) and Hainan (72.24±5.34 µg m−3), followed by thecities in the central region, including Lhasa (65.99±7.24 µg m−3) and Shigatse (65.15±6.14 µg m−3), and the lowestone was found in a city of the southeastern Tibetan Plateau (Aba) (55.17±12.77 µg m−3). The spatial pattern of the 8 hO3 concentration is highlyconsistent with the result predicted by Liu et al. (2018) using the CMAQ model,while it is not in agreement with the result estimated by Zhan et al. (2018)using the RF–STK model. The difference between the present study and Zhan et al. (2018) is seen on the northern Tibetan Plateau, which lacks a monitoring siteand still has the higher uncertainty. Firstly, this might be due to theweakness of RF–STK mentioned above. Moreover, Zhan et al. (2018) only usedthe ground-level-measured data in 2015 to establish the model, and the datafrom new sites since 2015 were not incorporated into the model, which couldincrease the model uncertainty (Zhan et al., 2018). As shown in Fig. 6, mostof the cities in Qinghai Province (e.g. Huangnan, Hainan, and Guoluo)generally showed a higher 8 hO3 concentration over the Tibetan Plateau,which was in a good agreement with the spatial distribution of theO3column amount (Fig. S3). Besides this, some cities in Tibet, such as Shigatse andLhasa, also showed higher 8 hO3 levels. It was assumed that theprecursor emissions in these regions were significantly higher than those inother cities of the Tibetan Plateau (Fig. S4). Zhang et al. (2007) used thesatellite data to observe that the higher VOCs andNOx emission was concentratedin the residential areas with high population density on the remote TibetanPlateau. Apart from the effect of anthropogenic emission, the meteorologicalconditions could also be important factors for the 8 hO3concentration. As shown in Figs. S5–S10, a higher blh and sp on thenortheastern Tibetan Plateau might promote theO3 formation through thereaction of the VOC and OH radical, leading to a higher 8 hO3concentration in these cities (Ou et al., 2015). In addition, a lower tpoccurred on the northern Tibetan Plateau and thenortheastern Tibetan Plateau, both ofwhich were unfavourable to the ambientO3 removal (Yoo et al., 2014). Incontrast, the higher tp observed on the southeastern Tibetan Plateau resultedin slightO3 pollution.
Table 4The estimated 8 hO3 concentration in 19 prefecture-level cities over the Tibetan Plateau during four seasons, including spring, summer,autumn, and winter.
3.4 The temporal variation in the simulated 8 hO3 concentration over the Tibetan Plateau
The annual mean estimated 8 hO3 concentration on the Tibetan Plateaudisplayed a slow increase, from64.74±8.30 to66.45±8.67 µg m−3 in 2005–2015 (Table S1), whereas itdecreased from the peak to65.87±8.52 µg m−3 during2015–2018 (Fig. 7). Based on the Mann–Kendall method (Fig. 8a), it was foundthat the surfaceO3 concentration exhibited a slight increase on thewhole, while the degree of increase was not significant (p>0.05).Besides this, it should be noted that theO3 concentrations in variousregions showed different rates of increase. As depicted in Fig. 8b, we foundthat the 8 hO3 concentrations on the northern, western, and eastern Tibetan Plateaudisplayed significant an increasing trend at the rate of 1–3 µg m−3during 2005–2018. The middle region of the Tibetan Plateau showed a moderateincrease trend at the rate of 0–1 µg m−3. However, the 8 hO3concentration in Shigatse and Shannan even displayed a decreasing trend in 2005–2018.
Besides this, the 8 hO3 concentrations on the Tibetan Plateau displayedsignificantly seasonal discrepancy. The estimated 8 hO3 level on the Tibetan Plateau followed the order of spring (75.00±8.56 µg m−3) > summer (71.05±11.13 µg m−3) > winter (56.39±7.42 µg m−3) > autumn(56.13±8.27 µg m−3) (Fig. 9 and Table 4). The 8 hO3concentrations in most of prefecture-level cities showed similarseasonal characteristics, with overall seasonal variation on the Tibetan Plateau. Based on the result summarised in Table S2, it was found that thekey precursors of ambientO3 generally displayed higher emissionsin winter compared with other seasons. However, the seasonal distribution ofambientO3 concentration was not in accordance with the precursoremissions, suggesting that the meteorological factors might play moreimportant roles in ambientO3 concentration. It was well known that thehigher air temperatures in spring and summer were closely related to the lowsp and high sund, both of which promotedO3 formation (Sitnov et al.,2017). Although summer showed the highest air temperature and the longestsunshine duration, the higher rainfall amount in summer decreased theambientO3 concentration via wet deposition (Li et al., 2017a, 2019b). Moreover, the highest blh occurred in spring, which wasfavourable to the strong stratosphere–troposphere exchange process on the Tibetan Plateau (Skerlak et al., 2014). Therefore, the 8 hO3 concentrations insummer and winter were lower than that in spring. Nonetheless,the 8 hO3 levels in Diqing, Shannan, and Nyingchi displayed the highestvalues in spring (56.38±7.87,73.90±5.97, and73.22±2.77 µg m−3), followed by winter (45.88±7.05,61.71±4.32, and62.24±3.63 µg m−3) and summer (44.35±5.90,61.00±5.86, and59.60±2.33 µg m−3), and the lowestones in autumn (37.45±5.76,54.70±3.13, and53.84±2.06 µg m−3). The lowerO3 level in summer than winter was mainlyattributable to the higher precipitation observed in the summer of thesecities (Fig. S11). In addition, it should be noted that theNOx andVOC emissions of the southern Tibetan Plateau (e.g. Shannan) exhibited highervalues in winter compared with other seasons.
3.5 The nonattainment days over the Tibetan Plateau during 2005–2018
The annual mean nonattainment days in the 19 prefecture-level cities overthe Tibetan Plateau are summarised in Table 2. The value of 100 µg m−3 was regardedas the critical value for the 8 hO3 level by the World Health Organization(WHO). Nonattainment days refer to total days with the 8 hO3concentration higher than 100 µg m−3. Although the annual mean8 hO3 concentrations in all of the cities over the Tibetan Plateau did notexceed the critical value, not all of the regions experienced excellent airquality in the long term (2005–2018). Some cities of Qinghai Province,including Huangnan, Haidong, and Guoluo, suffered from 45, 40, and 40nonattainment days each year (Fig. 10 and Table 5). Besides this, some cities onthe southern Tibetan Plateau, such as Shigatse and Shannan, also experienced morethan 40 nonattainment days each year, suggesting that the Tibetan Plateaustill faced the risk ofO3 pollution. Fortunately, some remotecities, such as Ali, Ngari, and Qamdo, did not experience excessiveO3 pollution all the time, which was ascribed to low precursoremissions and appropriate meteorological conditions. It should be noted thatthe nonattainment days in regions with highO3 concentration showedsignificant seasonal difference, whereas the seasonal difference wasnot remarkable in cities with lowO3 pollution. As shown in Table 2,it should be noted that nearly all of the nonattainment days could bedetected in spring and summer, which was in good agreement with theO3levels in different seasons, indicating that theO3 pollution issueshould be given more attention in spring and summer.
Table 5The mean nonattainment days (8 hO3 level >100 µg m−3) in 19 prefecture-level cities over the Tibetan Plateau each year.
The determination of nonattainment days showed some uncertainties, owing tothe predictive error of modelledO3 concentration. First of all,meteorological data used in RF–GAM model were collected from reanalysis data,and these gridded data often showed some uncertainties, which could increasethe uncertainty ofO3 estimation. Second, theO3 column amountused in the present study reflected the verticalO3 column amount ratherthan the surfaceO3 concentration. Thus, it could decrease the predictiveperformance of the surfaceO3 level.
In the present study, we developed a novel hybrid model (RF–GAM) based onmultiple explanatory variables to estimate the surface 8 hO3concentration across the remote Tibetan Plateau. The 10-foldcross-validation method demonstrated that RF–GAM achieved excellentperformance, with the highestR2 value (0.76) and lowest root-mean-square error (RMSE) (14.41 µg m−3), compared with other models,including the RF–STK, RF, BPNN, XGBoost, GRNN, ElmanNN, and ELM models.Moreover, the unlearning ground-level-measuredO3 data validated the fact that theRF–GAM model showed better extrapolation performance (R2=0.67,RMSE =25.68 µg m−3). The result of variable importance suggestedthat time, sund, and sp were key factors for the surface 8 hO3concentration over the Tibetan Plateau. Based on the RF–GAM model, we found thatthe estimated 8 hO3 concentration exhibited notable spatial variation,with higher values in some cities of the northern Tibetan Plateau, such asHuangnan (73.48±4.53 µg m−3) and Hainan (72.24±5.34 µg m−3), and lower values in some cities of the southeastern TibetanPlateau, such as Aba (55.17±12.77 µg m−3). Besides this, we alsofound that theO3 level displayed a slow increase, from64.74±8.30 to66.45±8.67 µg m−3 from 2005 to 2015,while theO3 concentration decreased to65.87±8.52 µg m−3 in 2018. The estimated 8 hO3 level on the Tibetan Plateaushowed significant seasonal discrepancy in the order of spring(75.00±8.56 µg m−3) > summer (71.05±11.13 µg m−3) > winter (56.39±7.42 µg m−3) > autumn (56.13±8.27 µg m−3). Based on thecritical value set by the WHO, most of the cities on the Tibetan Plateau hadexcellent air quality, while several cities (e.g. Huangnan, Haidong,and Guoluo) still suffered from more than 40 nonattainment days each year.
The RF–GAM model forO3 estimation has several limitations. First ofall, theO3 estimation of the northern Tibetan Plateau might show someuncertainties because there are few ground-level monitoring sites, andthus we cannot validate the reliability of predicted values in regionswithout a monitoring site. Secondly, our approach did not include data onthe emission inventory or traffic count because the continuous emissions ofNOx and VOCs were not open access. Lastly, we only focused on thetemporal variation in the surfaceO3 concentration in the past 10 years, andthe short-termO3 data cannot reflect the response ofO3 pollutionto climate change. In the future work, we should combine more explanatoryvariables such as long-termNOx and VOC emissions to retrieve thesurfaceO3 level over the Tibetan Plateau in recent decades.
The dailyO3 column data were collected fromhttps://acdisc.gsfc.nasa.gov/data/Aura_OMI_Level3/OMDOAO3e.003/2005/ (GES DISC, 2020). The ground-observedO3 data were downloaded fromhttps://www.aqistudy.cn/historydata/ (Aqistudy, 2020). All of these data are open access.
The supplement related to this article is available online at: https://doi.org/10.5194/acp-20-6159-2020-supplement.
This study was conceived by RL and HF. Statistical modelling wasperformed by RL, YZ, YM, WZ, and ZZ. RL drafted the paper.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Study of ozone, aerosols and radiation over the Tibetan Plateau (SOAR-TP) (ACP/AMT inter-journal SI)”. It is not associated with a conference.
This work has been supported by the National Natural Science Foundation of China(grant nos. 91744205, 21777025, 21577022, 21177026).
This research has been supported by the National Natural Science Foundation of China (grant no. 91744205).
This paper was edited by Tao Wang and reviewed by two anonymous referees.
Aqistudy: Ground-observed dailyO3 data, available at:https://www.aqistudy.cn/historydata/, last access: 20 May 2020.
Bornman, J. F., Barnes, P. W., Robson, T. M., Robinson, S. A., Jansen, M. A., Ballaré, C. L., and Flint, S. D.: Linkages between stratospheric ozone, UVradiation and climate change and their implications for terrestrialecosystems, Photoch. Photobio. Sci, 18, 681–716,https://doi.org/10.1039/C8PP90061B, 2019.
Calfapietra, C., Fares, S., Manes, F., Morani, A., Sgrigna, G., and Loreto, F.:Role of Biogenic Volatile Organic Compounds (BVOC) emitted by urban trees onozone concentration in cities: A review, Environ. Pollut., 183, 71–80,https://doi.org/10.1016/j.envpol.2013.03.012, 2013.
Chan, C., Wong, K., Li, Y., Chan, L., and Zheng, X.: The effects of SoutheastAsia fire activities on tropospheric ozone, trace gases and aerosols at aremote site over the Tibetan Plateau of Southwest China, Tellus B, 58,310–318,https://doi.org/10.1111/j.1600-0889.2006.00187.x,2006.
Chen, G., Knibbs, L. D., Zhang, W., Li, S., Cao, W., Guo, J., Ren, H., Wang,B., Wang, H., and Williams, G.: Estimating spatiotemporal distribution ofPM1 concentrations in China with satellite remote sensing, meteorology,and land use information, Environ. Pollut., 233, 1086–1094,https://doi.org/10.1016/j.envpol.2017.10.011, 2018a.
Chen, G., Li, S., Knibbs, L. D., Hamm, N., Cao, W., Li, T., Guo, J., Ren, H.,Abramson, M. J., and Guo, Y.: A machine learning method to estimate PM2.5concentrations across China with remote sensing, meteorological and land useinformation, Sci. Total Environ. 636, 52–60,https://doi.org/10.1016/j.scitotenv.2018.04.251, 2018b.
Chen, G., Morawska, L., Zhang, W., Li, S., Cao, W., Ren, H., Wang, B., Wang,H., Knibbs, L. D., and Williams, G.: Spatiotemporal variation of PM1pollution in China, Atmos. Environ., 178, 198–205,https://doi.org/10.1016/j.atmosenv.2018.01.053, 2018c.
Chen, P., Yang, J., Pu, T., Li, C., Guo, J., Tripathee, L., and Kang, S.:Spatial and temporal variations of gaseous and particulate pollutants in sixsites in Tibet, China, during 2016–2017, Aerosol Air Qual. Res., 19, 516–527,https://doi.org/10.4209/aaqr.2018.10.0360, 2019.
Emberson, L. D., Pleijel, H., Ainsworth, E. A., Berg, M. V. D., Ren, W., Osborne, S., Mills, G., Pandey, D., Dentener, F., Buker, P., Ewert, F., Koeble, R., and Dingenen, R. V.: Ozone effects on crops and consideration in crop models, Eur. J. Agron., 100, 19–34,https://doi.org/10.1016/j.eja.2018.06.002, 2018.
Fang, K., Makkonen, R., Guo, Z., Zhao, Y., and Seppä, H.: An increase in thebiogenic aerosol concentration as a contributing factor to the recentwetting trend in Tibetan Plateau, Sci. Rep., 5, 14628,https://doi.org/10.1038/srep14628, 2015.
Feng, Z., Hu, E., Wang, X., Jiang, L., and Liu, X.: Ground-levelO3pollution and its impacts on food crops in China: a review, Environ. Pollut., 199, 42–48,https://doi.org/10.1016/j.envpol.2015.01.016, 2015.
Feng, Z., De Marco, A., Anav, A., Gualtieri, M., Sicard, P., Tian, H.,Fornasier, F., Tao, F., Guo, A., and Paoletti, E.: Economic losses due to ozoneimpacts on human health, forest productivity and crop yield across China, Environ. Interna, 131, 104966,https://doi.org/10.1016/j.envint.2019.104966, 2019.
Fioletov, V., Bodeker, G., Miller, A., McPeters, R., and Stolarski, R.: Globaland zonal total ozone variations estimated from ground-based and satellitemeasurements: 1964–2000, J. Geophys. Res., 107, ACH 21-21–ACH 21-14,https://doi.org/10.1029/2001JD001350, 2002.
Fu, Y., Liao, H., and Yang, Y.: Interannual and decadal changes in troposphericozone in China and the associated chemistry-climate interactions: A review, Adv. Atmos. Sci., 36, 975–993,https://doi.org/10.1007/s00376-019-8216-9, 2019.
Gao, Z., Shao, X., Jiang, P., Cao, L., Zhou, Q., Yue, C., Liu, Y., and Wang, C.:Parameters optimization of hybrid fiber laser-arc butt welding on 316Lstainless steel using Kriging model and GA, Opt. Laser Technol., 83,153–162,https://doi.org/10.1016/j.optlastec.2016.04.001, 2016.
Geng, G., Murray, N. L., Chang, H. H., and Liu, Y.: The sensitivity ofsatellite-based PM2.5 estimates to its inputs: Implications to modeldevelopment in data-poor regions, Environ. Interna., 121, 550–560,https://doi.org/10.1016/j.envint.2018.09.051, 2018.
GES DISC: DailyO3 column data, available at:https://acdisc.gsfc.nasa.gov/data/Aura_OMI_Level3/OMDOAO3e.003/2005/, last access: 20 May 2020.
Ghude, S. D., Chate, D., Jena, C., Beig, G., Kumar, R., Barth, M., Pfister,G., Fadnavis, S., and Pithani, P.: Premature mortality in India due toPM2.5 and ozone exposure, Geophys. Res. Lett., 43, 4650–4658,https://doi.org/10.1002/2016GL068949, 2016.
He, X., Pang, S., Ma, J., Zhang, Y.: Influence of relative humidity on heterogeneous reactions ofO3 andO3∕SO2 with soot particles: potential for environmental and health effects, Atmos. Environ., 165,198–206,https://doi.org/10.1016/j.atmosenv.2017.06.049, 2017.
He, X. and Zhang, Y.-H.: Influence of relative humidity onSO2 oxidation byO3 andNO2 on the surface of TiO2 particles: Potential for formation of secondary sulfate aerosol, Spectrochim. Acta A, 219, 121–128,https://doi.org/10.1016/j.saa.2019.04.046, 2019.
Huang, G., Huang, G.-B., Song, S., and You, K.: Trends in extreme learningmachines: A review, Neural Networks, 61, 32–48,https://doi.org/10.1016/j.neunet.2014.10.001, 2015.
Kim, S.-Y., Bechle, M., Hankey, S., Sheppard, E. L. A., Szpiro, A. A.,and Marshall, J. D.: Concentrations of criteria pollutants in the contiguous US,1979–2015: Role of model parsimony in integrated empirical geographicregression, UW Biostatistics Working Paper Series, 2018.
Li, C. H. and Park, S. C.: Combination of modified BPNN algorithms and anefficient feature selection method for text categorization, Inform. ProcessManag., 45, 329–340,https://doi.org/10.1016/j.ipm.2008.09.004,2009.
Li, H., Liu, T., Wang, M., Zhao, D., Qiao, A., Wang, X., Gu, J., Li, Z.,and Zhu, B.: Design optimization of stent and its dilatation balloon usingkriging surrogate model, Biomed. Eng. Online, 16, 13,https://doi.org/10.1186/s12938-016-0307-6, 2017a.
Li, J., Mao, J., Fiore, A. M., Cohen, R. C., Crounse, J. D., Teng, A. P., Wennberg, P. O., Lee, B. H., Lopez-Hilfiker, F. D., Thornton, J. A., Peischl, J., Pollack, I. B., Ryerson, T. B., Veres, P., Roberts, J. M., Neuman, J. A., Nowak, J. B., Wolfe, G. M., Hanisco, T. F., Fried, A., Singh, H. B., Dibb, J., Paulot, F., and Horowitz, L. W.: Decadal changes in summertime reactive oxidized nitrogen and surface ozone over the Southeast United States, Atmos. Chem. Phys., 18, 2341–2361,https://doi.org/10.5194/acp-18-2341-2018, 2018.
Li, R., Cui, L., Li, J., Zhao, A., Fu, H., Wu, Y., Zhang, L., Kong, L.,and Chen, J.: Spatial and temporal variation of particulate matter and gaseouspollutants in China during 2014–2016, Atmos. Environ, 161, 235–246,https://doi.org/10.1016/j.atmosenv.2017.05.008, 2017b.
Li, R., Cui, L., Meng, Y., Zhao, Y., and Fu, H.: Satellite-based prediction ofdailySO2 exposure across China using a high-quality randomforest-spatiotemporal Kriging (RF-STK) model for health risk assessment, Atmos. Environ, 208, 10–19,https://doi.org/10.1016/j.atmosenv.2019.03.029, 2019a.
Li, R., Wang, Z., Cui, L., Fu, H., Zhang, L., Kong, L., Chen, W., and Chen, J.:Air pollution characteristics in China during 2015–2016: Spatiotemporalvariations and key meteorological factors, Sci. Total Environ., 648, 902–915,https://doi.org/10.1016/j.scitotenv.2018.08.181, 2019b.
Li, R., Cui, L. L, Fu, H. B., Li, J. L., Zhao, Y. L., and Chen, J. M.:Satellite-based estimation of full-coverage ozone (O3) concentrationand health effect assessment across Hainan Island, J. Clean Prod., 244, 118773,https://doi.org/10.1016/j.jclepro.2019.118773, 2020.
Lin, W., Zhu, T., Song, Y., Zou, H., Tang, M., Tang, X., and Hu, J.: Photolysisof surfaceO3 and production potential of OH radicals in the atmosphereover the Tibetan Plateau, J. Geophys. Res., 113, D02309,https://doi.org/10.1029/2007JD008831, 2008.
Lin, Y. Y., Jiang, F., Zhao, J., Zhu, G., He, X. J., Ma, X. L., Li, S., Sabel,C. E., and Wang, H. K.: Impacts ofO3 on premature mortality and crop yieldloss across China, Atmos. Environ., 194, 41–47,https://doi.org/10.1016/j.atmosenv.2018.09.024, 2018.
Liu, H., Liu, S., Xue, B. R., Lv, Z. F., Meng, Z. H., Yang, X. F., Xue, T., Yu, Q., and He, K. B.: Ground-level ozone pollution and its health impacts in China, Atmos. Environ., 173, 223–230,https://doi.org/10.1016/j.atmosenv.2017.11.014, 2018.
Ma, Z., Hu, X., Sayer, A. M., Levy, R., Zhang, Q., Xue, Y., Tong, S., Bi, J.,Huang, L., and Liu, Y.: Satellite-based spatiotemporal trends in PM2.5concentrations: China, 2004–2013, Environ. Health Per., 124, 184–192,https://doi.org/10.1289/ehp.1409481, 2015.
Malik, A. and Tauler, R.: Exploring the interaction betweenO3 andNOx pollution patterns in the atmosphere of Barcelona, Spain using theMCR-ALS method, Sci. Total Environ., 517, 151–161,https://doi.org/10.1016/j.scitotenv.2015.01.105, 2015.
Marco, D.: Exposure to PM10,NO2, andO3 and impacts on humanhealth, Environ. Sci. Pollut. Res., 24, 2781,https://doi.org/10.1007/s11356-016-8038-6, 2017.
McPeters, R. D., Frith, S., and Labow, G. J.: OMI total column ozone: extending the long-term data record, Atmos. Meas. Tech., 8, 4845–4850,https://doi.org/10.5194/amt-8-4845-2015, 2015.
Ou, J., Zheng, J., Li, R., Huang, X., Zhong, Z., Zhong, L., and Lin, H.:Speciated OVOC and VOC emission inventories and their implications forreactivity-based ozone control strategy in the Pearl River Delta region,China, Sci. Total Environ., 530, 393–402,https://doi.org/10.1016/j.scitotenv.2015.05.062, 2015.
Qian, M., Zhaosheng, W., Rong, W., Mei, H., and Jiali, S.: Assessing the Impactof Ozone Pollution on Summer NDVI based Vegetation Growth in North China, Remote Sensing Technology and Application, 33, 696–702,https://doi.org/10.11873/j.issn.1004-0323.2018.4.0696, 2018.
Santurtún, A., González-Hidalgo, J. C., Sanchez-Lorenzo, A.,and Zarrabeitia, M. T.: Surface ozone concentration trends and its relationshipwith weather types in Spain (2001–2010), Atmos. Environ., 101, 10–22,https://doi.org/10.1016/j.atmosenv.2014.11.005, 2015.
Shao, Z., Er, M. J., and Wang, N.: An effective semi-cross-validation modelselection method for extreme learning machine with ridge regression, Neurocomputing, 151, 933–942,https://doi.org/10.1016/j.neucom.2014.10.002, 2015.
Shen, Z., Cao, J., Zhang, L., Zhao, Z., Dong, J., Wang, L., Wang, Q., Li, G., Liu, S., and Zhang, Q.: Characteristics of surfaceO3 over Qinghai Lakearea in Northeast Tibetan Plateau, China, Sci. Total Environ., 500, 295–301,https://doi.org/10.1016/j.scitotenv.2014.08.104, 2014.
Shen, L., Jacob, D. J., Liu, X., Huang, G., Li, K., Liao, H., and Wang, T.: An evaluation of the ability of the Ozone Monitoring Instrument (OMI) to observe boundary layer ozone pollution across China: application to 2005–2017 ozone trends, Atmos. Chem. Phys., 19, 6551–6560,https://doi.org/10.5194/acp-19-6551-2019, 2019.
Shi, X., Zhao, C., Jiang, J. H., Wang, C., Yang, X., and Yung, Y. L.: SpatialRepresentativeness of PM2.5 Concentrations Obtained Using ObservationsFrom Network Stations, J. Geophys. Res., 123, 3145–3158,https://doi.org/10.1002/2017JD027913, 2018.
Sitnov, S., Mokhov, I., and Lupo, A.: Ozone, water vapor, and temperatureanomalies associated with atmospheric blocking events over Eastern Europe inspring-summer 2010, Atmos. Environ., 164, 180–194,https://doi.org/10.1016/j.atmosenv.2017.06.004, 2017.
Skerlak, B., Sprenger, M., Pfahl, S., Roches, A., Sodemann, H., and Wernli, H.:Rapid exchange between the stratosphere and the planetary boundary layerover the Tibetan Plateau, EGU General Assembly Conference Abstracts, 2014.
Stähle, C., Rieder, H. E., Mayer, M., and Fiore, A. M.: Past and futurechanges in surface ozone pollution in Central Europe: insights fromobservations and chemistry-climate model simulations, EGU General AssemblyConference Abstracts, p. 12677, 2018.
Tan, Z., Lu, K., Jiang, M., Su, R., Dong, H., Zeng, L., Xie, S., Tan, Q.,and Zhang, Y.: Exploring ozone pollution in Chengdu, southwestern China: A casestudy from radical chemistry toO3-VOC-NOx sensitivity, Sci. TotalEnviron., 636, 775–786,https://doi.org/10.1016/j.scitotenv.2018.04.286, 2018.
Vellingiri, K., Kim, K. H., Jeon, J. Y., Brown, R. J., and Jung, M. C.: Changes inNOx andO3 concentrations over a decade at a central urban area ofSeoul, Korea, Atmos. Environ., 112, 116–125,https://doi.org/10.1016/j.atmosenv.2015.04.032, 2015.
Wang, C., Wang, X., Gong, P., and Yao, T.: Polycyclic aromatic hydrocarbons insurface soil across the Tibetan Plateau: Spatial distribution, source andair–soil exchange, Environ. Pollut., 184, 138–144,https://doi.org/10.1016/j.envpol.2013.08.029, 2014a.
Wang, Y., Ying, Q., Hu, J., and Zhang, H.: Spatial and temporal variations ofsix criteria air pollutants in 31 provincial capital cities in China during2013–2014, Environ. Interna., 73, 413–422,https://doi.org/10.1016/j.envint.2014.08.016, 2014b.
Wang, L., Zeng, Y., and Chen, T.: Back propagation neural network with adaptivedifferential evolution algorithm for time series forecasting, Expert Syst.Appl., 42, 855–863,https://doi.org/10.1016/j.eswa.2014.08.018,2015.
Wang, M., Sampson, P. D., Hu, J. L., Kleeman, M., Keller, J. P., Olives, C.,Szpiro, A. A., Vedal, S., and Kaufman, J. D.: Combining land-use regression andchemical transport modelling in a spatiotemporal geostatistical model forozone and PM2.5, Environ. Sci. Tech., 50, 5111–5118,https://doi.org/10.1021/acs.est.5b06001, 2016.
Wang, T., Xue, L., Brimblecombe, P., Lam, Y. F., Li, L., and Zhang, L.: Ozonepollution in China: A review of concentrations, meteorological influences,chemical precursors, and effects, Sci. Total Environ., 575, 1582–1596,https://doi.org/10.1016/j.scitotenv.2016.10.081, 2017.
Wang, N., Lyu, X., Deng, X., Huang, X., Jiang, F., and Ding, A.: AggravatingO3 pollution due toNOx emission control in eastern China, Sci. Total Environ., 677, 732–744,https://doi.org/10.1016/j.scitotenv.2019.04.388, 2019.
Yin, P., Chen, R., Wang, L., Meng, X., Liu, C., Niu, Y., Lin, Z., Liu, Y.,Liu, J., and Qi, J.: Ambient ozone pollution and daily mortality: a nationwidestudy in 272 Chinese cities, Environ. Health Per., 125, 117006,https://doi.org/10.1289/EHP1849, 2017a.
Yin, X., Kang, S., de Foy, B., Cong, Z., Luo, J., Zhang, L., Ma, Y., Zhang, G., Rupakheti, D., and Zhang, Q.: Surface ozone at Nam Co in the inland Tibetan Plateau: variation, synthesis comparison and regional representativeness, Atmos. Chem. Phys., 17, 11293–11311,https://doi.org/10.5194/acp-17-11293-2017, 2017b.
Yoo, J. M., Lee, Y. R., Kim, D., Jeong, M. J., Stockwell, W. R., Kundu, P. K.,Oh, S. M., Shin, D. B., and Lee, S. J.: New indices for wet scavenging of airpollutants (O3, CO,NO2,SO2, and PM10) by summertimerain, Atmos. Environ., 82, 226–237,https://doi.org/10.1016/j.atmosenv.2013.10.022, 2014.
Yu, S. C., Mathur, R., Kang, D., Schere, K., Eder, B., and Plein, J.: Performanceand diagnostic evaluations of a real-time ozone forecast by the Eta–CMAQmodel suite during the 2002 New England Air Quality Study (NEAQS), J. AirWaste Manage., 56, 1459–1471, 2006.
Yu, S. C.: Fog geoengineering to abate local ozone pollution at ground levelby enhancing air moisture, Environ. Chem. Lett., 17, 565–580, 2019.
Zang, L., Mao, F., Guo, J., Gong, W., Wang, W., and Pan, Z.: Estimating hourlyPM1 concentrations from Himawari-8 aerosol optical depth in China, Environ. Pollut., 241, 654–663,https://doi.org/10.1016/j.envpol.2018.05.100, 2018.
Zang, L., Mao, F., Guo, J., Wang, W., Pan, Z., Shen, H., Zhu, B., and Wang, Z.:Estimation of spatiotemporal PM1.0 distributions in China by combiningPM2.5 observations with satellite aerosol optical depth, Sci. TotalEnviron., 658, 1256–1264,https://doi.org/10.1016/j.scitotenv.2018.12.297, 2019.
Zhan, Y., Luo, Y., Deng, X., Grieneisen, M. L., Zhang, M., and Di, B.:Spatiotemporal prediction of daily ambient ozone levels across China usingrandom forest for human exposure assessment, Environ. Pollut., 233, 464–473,https://doi.org/10.1016/j.envpol.2017.10.029, 2018.
Zhang, Q., Streets, D. G., He, K., Wang, Y., Richter, A., Burrows, J. P., Uno,I., Jang, C. J., Chen, D., and Yao, Z.:NOx emission trends for China,1995–2004: The view from the ground and the view from space, J. Geophys. Res., 112, D22306,https://doi.org/10.1029/2007JD008684, 2007.
Zhang, L., Jacob, D. J., Downey, N. V., Wood, D. A., Blewitt, D., Carouge,C. C., van Donkelaar, A., Jones, D. B., Murray, L. T., and Wang, Y.: Improvedestimate of the policy-relevant background ozone in the United States usingthe GEOS-Chem global model with horizontal resolution overNorth America, Atmos. Environ., 45, 6769–6776,https://doi.org/10.1016/j.atmosenv.2011.07.054, 2011.
Zhang, Y. and Wang, Y.: Climate-driven ground-level ozone extreme in the fallover the Southeast United States, P. Natl. Acad. Sci. USA, 113, 10025–10030,https://doi.org/10.1073/pnas.1602563113, 2016.
Zheng, C., Shen, J., Zhang, Y., Huang, W., Zhu, X., Wu, X., Chen, L., Gao,X., and Cen, K.: Quantitative assessment of industrial VOC emissions in China:Historical trend, spatial distribution, uncertainties, and projection, Atmos. Environ., 150, 116–125,https://doi.org/10.1016/j.atmosenv.2016.11.023, 2017.
Zhu, Y., Zhan, Y., Wang, B., Li, Z., and Qin, Y., Zhang, K.: Spatiotemporallymapping of the relationship betweenNO2 pollution and urbanization fora megacity in Southwest China during 2005–2016, Chemosphere, 220, 155–162,https://doi.org/10.1016/j.chemosphere.2018.12.095, 2019.