Distinct seasonality in vertical variations of tropospheric ozone over coastal regions of southern China

The surface ozone pollution is strongly coupled with ozone variations above the ground. Using sufficient airborne ozone profiles during 2012-2018, this study reveals the tropospheric ozone distributions over four cities located in coastal regions of southern China. The 7-year mean tropospheric ozone profiles in the four cities consistently show a double-maxima profile, with a local maximum at 1 km altitude and the other in the middle-to-upper troposphere. Seasonally, springtime ozone is larger than the annual mean throughout the troposphere, while ozone in summer is high in the middle-to-upper troposphere, leading to largest vertical variations among seasons. Ozone in the middle-to-upper troposphere is lower in autumn than in spring and summer. The winter ozone is characterized with a minimum in the lower troposphere, and low values in the middle-to-upper troposphere, leading to least vertical variations among seasons. We untangle the causes for these complicated vertical ozone variations using the GEOS-Chem model. The tropospheric ozone over southern China is partitioned into locally produced ozone, regionally transported native ozone, imported ozone from outside of China (foreign ozone) and natural stratospheric ozone. The results suggest that the springtime ozone abundance is due to the enhanced import of foreign and stratospheric ozone and the intensified regional transport processes of native ozone. In summer, local ozone production is enhanced and regional transport of ozone in the middle-to-upper troposphere is strengthened due to upward air motions, while such transport becomes weaker in autumn leaving low ozone in the middle-to-upper troposphere. In winter, the intensive westerly jets promote foreign and stratospheric ozone again in the middle-to-upper troposphere, but the local ozone production and regional transport are sharply reduced, resulting in low ozone near the surface. This study provides new insights into regional ozone profiles and reveals the significance of vertical ozone variations on surface ozone prevention strategy.

The Role of Convective Up- and Downdrafts in the Transport of Trace Gases in the Amazon

Deep convective clouds can redistribute gaseous species and particulate matter among different layers of the troposphere with important implications for atmospheric chemistry and climate. The large number of atmospheric trace gases of different volatility makes it challenging to predict their partitioning between hydrometeors and gas phase inside highly dynamic deep convective clouds. In this study, we use an ensemble of 51,200 trajectories simulated with a cloud-resolving model to characterize up- and downdrafts within Amazonian deep convective clouds. We also estimate the transport of a set of hypothetical non-reactive gases of different volatility, within the up- and downdrafts. We find that convective air parcels originating from the boundary layer (i.e., originating at 0.5 km altitude), can transport up to 25% of an intermediate volatility gas species (e.g., methyl hydrogen peroxide) and up to 60% of high volatility gas species (e.g., n-butane) to the cloud outflow above 10 km through the mean convective updraft. At the same time, the same type of gases can be transported to the boundary layer from the middle troposphere (i.e., originating at 5 km) within the mean convective downdraft with an efficiency close to 100%. Low volatility gases (e.g., nitric acid) are not efficiently transported, neither by the updrafts nor downdrafts, if the gas is assumed to be fully retained in a droplet upon freezing. The derived properties of the mean up- and downdraft can be used in future studies for investigating convective transport of a larger set of reactive trace gases.

Enhancement of carbon monoxide concentration in atmosphere due to large scale forest fire of Uttarakhand

The richly forested Indian state of Uttarakhand experienced widespread forest fires in April to May 2016. The current study examines dispersion of carbon monoxide (CO) from the source regions of forest fire to distant places, using the Lagrangian particle dispersion model, FLEXPART. Atmospheric Infrared Sounder (AIRS) observations revealed that CO columnar concentrations had increased by almost 28 percentage during 24 April to 02 May 2016 with respect to the previous non-burning period of April 2016 at Uttarakhand. It is also seen that there is considerable enhancement of 45 percentage in average columnar concentration of CO during the burning period, compared to that in the previous 5 years as observed by AIRS. In the present study, concentrations of CO at different pressure levels and columnar CO over Uttarakhand during the forest fire event have been simulated using FLEXPART. The area averaged profile of model derived CO has been compared with the profile from AIRS onboard Aqua. Comparison between model derived columnar CO and satellite observations shows good agreement with coefficient of correlation (r) approximately 0.91 over the burnt areas. Further analysis using FLEXPART reveals that the transport of pollutants is towards north-eastern and eastern regions from the locations of forest fire events. Model derived vertical distribution of CO over Tibet, which is situated at the north-east of Uttarakhand, shows significant increase of CO concentration at higher altitudes around 3 km from the mean sea level during the fire event. FLEXPART results show that the emissions from the Uttarakhand fires were transported to Tibet during the study period.

Quantifying stratosphere-troposphere transport of ozone using balloon-borne ozonesondes, radar windprofilers and trajectory models

In a series of 10-day campaigns in Ontario and Quebec, Canada, between 2005 and 2007, ozonesondes were launched twice daily in conjunction with continuous high-resolution wind-profiling radar measurements. Windprofilers can measure rapid changes in the height of the tropopause, and in some cases follow stratospheric intrusions. Observed stratospheric intrusions were studied with the aid of a Lagrangian particle dispersion model and the Canadian operational weather forecast system. Definite stratosphere-troposphere transport (STT) events occurred approximately every 2-3 days during the spring and summer campaigns, whereas during autumn and winter, the frequency was reduced to every 4-5 days. Although most events reached the lower troposphere, only three events appear to have significantly contributed to ozone amounts in the surface boundary layer. Detailed calculations find that STT, while highly variable, is responsible for an average, over the seven campaigns, of 3.1% of boundary layer ozone (1.2 ppb), but 13% (5.4 ppb) in the lower troposphere and 34% (22 ppb) in the middle and upper troposphere, where these layers are defined as 0-1 km, 1-3 km, and 3-8 km respectively. Estimates based on counting laminae in ozonesonde profiles, with judicious choices of ozone and relative humidity thresholds, compare moderately well, on average, with these values. The lamina detection algorithm is then applied to a large dataset from four summer ozonesonde campaigns at 18 North American sites between 2006 and 2011. The results show some site-to-site and year-to-year variability, but stratospheric ozone contributions average 4.6% (boundary layer), 15% (lower troposphere) and 26% (middle/upper troposphere). Calculations were also performed based on the TOST global 3D trajectory-mapped ozone data product. Maps of STT in the same three layers of the troposphere suggest that the STT ozone flux is greater over the North American continent than Europe, and much greater in winter and spring than in summer or fall. When averaged over all seasons, magnitudes over North America show similar ratios between levels to the previous calculations, but are overall 3-4 times smaller. This may be because of limitations (trajectory length and vertical resolution) to the current TOST-based calculation.

Forecasting air quality time series using deep learning

This paper presents one of the first applications of deep learning (DL) techniques to predict air pollution time series. Air quality management relies extensively on time series data captured at air monitoring stations as the basis of identifying population exposure to airborne pollutants and determining compliance with local ambient air standards. In this paper, 8 hr averaged surface ozone (O₃) concentrations were predicted using deep learning consisting of a recurrent neural network (RNN) with long short-term memory (LSTM). Hourly air quality and meteorological data were used to train and forecast values up to 72 hours with low error rates. The LSTM was able to forecast the duration of continuous O₃ exceedances as well. Prior to training the network, the dataset was reviewed for missing data and outliers. Missing data were imputed using a novel technique that averaged gaps less than eight time steps with incremental steps based on first-order differences of neighboring time periods. Data were then used to train decision trees to evaluate input feature importance over different time prediction horizons. The number of features used to train the LSTM model was reduced from 25 features to 5 features, resulting in improved accuracy as measured by Mean Absolute Error (MAE). Parameter sensitivity analysis identified look-back nodes associated with the RNN proved to be a significant source of error if not aligned with the prediction horizon. Overall, MAE's less than 2 were calculated for predictions out to 72 hours.

IMPLICATIONS

Novel deep learning techniques were used to train an 8-hour averaged ozone forecast model. Missing data and outliers within the captured data set were replaced using a new imputation method that generated calculated values closer to the expected value based on the time and season. Decision trees were used to identify input variables with the greatest importance. The methods presented in this paper allow air managers to forecast long range air pollution concentration while only monitoring key parameters and without transforming the data set in its entirety, thus allowing real time inputs and continuous prediction.