Urban ozone variability using automated machine learning: inference from different feature importance schemes.

Tropospheric ozone is an air pollutant at the ground level and a greenhouse gas which significantly contributes to the global warming. Strong anthropogenic emissions in and around urban environments enhance surface ozone pollution impacting the human health and vegetation adversely. However, observations are often scarce and the factors driving ozone variability remain uncertain in the developing regions of the world. In this regard, here, we conducted machine learning (ML) simulations of ozone variability and comprehensively examined the governing factors over a major urban environment (Ahmedabad) in western India. Ozone precursors (NO2, NO, CO, C5H8 and CH2O) from the CAMS (Copernicus Atmosphere Monitoring Service) reanalysis and meteorological parameters from the ERA5 (European Centre for Medium-Range Weather Forecast's (ECMWF) fifth-generation reanalysis) were included as features in the ML models. Automated ML (AutoML) fitted the deep learning model optimally and simulated the daily ozone with root mean square error (RMSE) of ~2 ppbv reproducing 84-88% of variability. The model performance achieved here is comparable to widely used ML models (RF-Random Forest and XGBoost-eXtreme Gradient Boosting). Explainability of the models is discussed through different schemes of feature importance, including SAGE (Shapley Additive Global importancE) and permutation importance. The leading features are found to be different from different feature importance schemes. We show that urban ozone could be simulated well (RMSE = 2.5 ppbv and R2 = 0.78) by considering first four leading features, from different schemes, which are consistent with ozone photochemistry. Our study underscores the need to conduct science-informed analysis of feature importance from multiple schemes to infer the roles of input variables in ozone variability. AutoML-based studies, exploiting potentials of long-term observations, can strongly complement the conventional chemistry-transport modelling and can also help in accurate simulation and forecast of urban ozone.

Photochemical evolution of air in a tropical urban environment of India: A model-based study.

The photochemical processes over tropical Indian region impact the atmospheric composition and air quality over local to global scales; nevertheless, studies on detailed atmospheric chemistry remain sparse in this region. In this study, we investigate the photochemical evolution of air in the downwind of a tropical semi-arid urban environment (Ahmedabad) in India using the Master Mechanism model. The 5-days long chemical evolution has been simulated for the winter conditions - when this region experiences strong ozone build up. Model environment has been set up by including the meteorological conditions, overhead ozone, and aerosol loading, etc. Nitrogen oxides (NOx), carbon monoxide (CO), ozone (O3), and several volatile organic compounds (VOCs) have been initialized in the model based on the wintertime observations. The model predicts large O3 production (∼115 ppbv) in the downwind regions, followed by a gradual decrease from the 3rd day onwards. Additionally, significant amounts of the secondary inorganics, e.g. nitric acid (∼17 ppbv), hydrogen peroxide (∼9 ppbv), and organics, e.g. ketones (∼11 ppbv), are also simulated. The noontime maximum levels of hydroxyl (OH) and hydroperoxyl (HO2) radicals are simulated to be 0.3 and 44 pptv, respectively. While the production of OH is dominated by the reaction of NO with HO2 on the first day, photolysis of O3 dominates subsequently with reduction in NOx levels. VOCs are the major OH sink during day 1, however contribution of CO is greater on further days. The air mass trajectory analysis suggests the outflow of ozone-rich air over the rural areas and the Arabian Sea, in agreement with measurements and a global model. Our study highlights the strong impact of the urban outflows on the regional atmospheric composition. The continuous measurements of VOCs and radicals are needed over tropical regions to complement the models and further improve the understanding of air chemistry.

Frequency distribution of pollutant concentrations over Indian megacities impacted by the COVID-19 lockdown.

The megacities experience poor air quality frequently due to stronger anthropogenic emissions. India had one of the longest lockdowns in 2020 to curb the spread of COVID-19, leading to reductions in the emissions from anthropogenic activities. In this article, the frequency distributions of different pollutants have been analysed over two densely populated megacities: Delhi (28.70° N; 77.10° E) and Kolkata (22.57° N; 88.36° E). In Delhi, the percentage of days with PM2.5 levels exceeding the National Ambient Air Quality Standards (NAAQS) between 25 March and 17 June dropped from 98% in 2019 to 61% in 2020. The lockdown phase 1 brought down the PM10 (particulate matter having an aerodynamic diameter ≤ 10 µm) levels below the daily NAAQS limit over Delhi and Kolkata. However, PM10 exceeded the limit of 100 µgm-3 during phases 2-5 of lockdown over Delhi due to lower temperature, weaker winds, increased relative humidity and commencement of limited traffic movement. The PM2.5 levels exhibit a regressive trend in the highest range from the year 2019 to 2020 in Delhi. The daily mean value for PM2.5 concentrations dropped from 85-90 µgm-3 to 40-45 µgm-3 bin, whereas the PM10 levels witnessed a reduction from 160-180 µgm-3 to 100-120 µgm-3 bin due to the lockdown. Kolkata also experienced a shift in the peak of PM10 distribution from 80-100 µgm-3 in 2019 to 20-40 µgm-3 during the lockdown. The PM2.5 levels in peak frequency distribution were recorded in the 35-40 µgm-3 bin in 2019 which dropped to 15-20 µgm-3 in 2020. In line with particulate matter, other primary gaseous pollutants (NOx, CO, SO2, NH3) also showed decline. However, changes in O3 showed mixed trends with enhancements in some of the phases and reductions in other phases. In contrast to daily mean O3, 8-h maximum O3 showed a reduction over Delhi during lockdown phases except for phase 3. Interestingly, the time of daily maximum was observed to be delayed by ~ 2 h over Delhi (from 1300 to 1500 h) and ~ 1 h over Kolkata (from 1300 to 1400 h) almost coinciding with the time of maximum temperature, highlighting the role of meteorology versus precursors. Emission reductions weakened the chemical sink of O3 leading to enhancement (120%; 11 ppbv) in night-time O3 over Delhi during phases 1-3.

Exploring the potential of machine learning for simulations of urban ozone variability.

Machine learning (ML) has emerged as a powerful technique in the Earth system science, nevertheless, its potential to model complex atmospheric chemistry remains largely unexplored. Here, we applied ML to simulate the variability in urban ozone (O3) over Doon valley of the Himalaya. The ML model, trained with past variations in O3 and meteorological conditions, successfully reproduced the independent O3 data (r2 ~ 0.7). Model performance is found to be similar when the variation in major precursors (CO and NOx) were included in the model, instead of the meteorology. Further the inclusion of both precursors and meteorology improved the performance significantly (r2 = 0.86) and the model could also capture the outliers, which are crucial for air quality assessments. We suggest that in absence of high-resolution measurements, ML modeling has profound implications for unraveling the feedback between pollution and meteorology in the fragile Himalayan ecosystem.

On the widespread enhancement in fine particulate matter across the Indo-Gangetic Plain towards winter.

Fine particulate matter (PM2.5, aerodynamic diameter ≤2.5 µm) impacts the climate, reduces visibility and severely influences human health. The Indo-Gangetic Plain (IGP), home to about one-seventh of the world's total population and a hotspot of aerosol loading, observes strong enhancements in the PM2.5 concentrations towards winter. We performed high-resolution (12 km × 12 km) atmospheric chemical transport modeling (WRF-Chem) for the post-monsoon to winter transition to unravel the underlying dynamics and influences of regional emissions over the region. Model, capturing the observed variations to an extent, reveals that the spatial distribution of PM2.5 having patches of enhanced concentrations (≥100 µgm-3) during post-monsoon, evolves dramatically into a widespread enhancement across the IGP region during winter. A sensitivity simulation, supported by satellite observations of fires, shows that biomass-burning emissions over the northwest IGP play a crucial role during post-monsoon. Whereas, in contrast, towards winter, a large-scale decline in the air temperature, significantly shallower atmospheric boundary layer, and weaker winds lead to stagnant conditions (ventilation coefficient lower by a factor of ~4) thereby confining the anthropogenic influences closer to the surface. Such changes in the controlling processes from post-monsoon to winter transition profoundly affect the composition of the fine aerosols over the IGP region. The study highlights the need to critically consider the distinct meteorological processes of west-to-east IGP and changes in dominant sources from post-monsoon to winter in the formulation of future pollution mitigation policies.

Continental outflow of anthropogenic aerosols over Arabian Sea and Indian Ocean during wintertime: ICARB-2018 campaign.

Chemical characterisation of atmospheric aerosols over Arabian Sea (AS) and Indian Ocean (IO) have been carried out during the winter period (January to February 2018) as part of the Integrated Campaign for Aerosols, gases and Radiation Budget (ICARB-2018). Mass concentrations of organic carbon (OC), elemental carbon (EC), water soluble and insoluble OC (WSOC, WIOC), primary and secondary OC (POC, SOC), water-soluble inorganic ions and trace metals have been estimated with a view to identify and quantify the major anthropogenic pollutants affecting the oceanic environments. Aerosol mass loading was found to exhibit strong spatial heterogeneity (varying from 13 to 84 µg m-3), significantly modulated by the origin of air-mass trajectories. Chemical analysis of aerosols revealed the presence of an intense pollution plume over south-eastern coastal Arabian Sea, near to south-west Indian peninsula (extending from ~ 12°N to 0° at 75°E) with a strong latitudinal gradient (~3 µg m-3/deg. from north to south) dominated by anthropogenic species contributing as high as 73% (38% nss-SO42-, 24.2% carbonaceous aerosols (21% Organic Matter, 3.2% EC) and 10% NH4+). Anthropogenic signature over oceanic environment was also evident from the dominance and high enrichment of elements like Zn, Cu, Mn and Pb in trace metals. Long-range transport of air-masses originating from Indo Gangetic Plains and its outflow regions in Bay of Bengal, has been seen over Arabian Sea during winter, that imparted such strong anthropogenic signatures over this oceanic environment. Comparison with previous cruise studies conducted nearly two decades ago shows a more than two-fold increase in the concentration of nss-SO42-, over the continental outflow region in Arabian Sea.

Distribution of volatile organic compounds over Indian subcontinent during winter: WRF-chem simulation versus observations.

We investigate the distribution of volatile organic compounds (VOCs) over Indian subcontinent during a winter month of January 2011 combining the regional model WRF-Chem (Weather Research and Forecasting model coupled with Chemistry) with ground- and space-based observations and chemical reanalysis. WRF-Chem simulated VOCs are found to be comparable with ground-based observations over contrasting environments of the Indian subcontinent. WRF-Chem results reveal the elevated levels of VOCs (e. g. propane) over the Indo-Gangetic Plain (16 ppbv), followed by the Northeast region (9.1 ppbv) in comparison with other parts of the Indian subcontinent (1.3-8.2 ppbv). Higher relative abundances of propane (27-31%) and ethane (13-17%) are simulated across the Indian subcontinent. WRF-Chem simulated formaldehyde and glyoxal show the western coast, Eastern India and the Indo-Gangetic Plain as the regional hotspots, in a qualitative agreement with the MACC (Monitoring Atmospheric Composition and Climate) reanalysis and satellite-based observations. Lower values of RGF (ratio of glyoxal to formaldehyde <0.04) suggest dominant influences of the anthropogenic emissions on the distribution of VOCs over Indian subcontinent, except the northeastern region where higher RGF (∼0.06) indicates the role of biogenic emissions, in addition to anthropogenic emissions. Analysis of HCHO/NO2 ratio shows a NOx-limited ozone production over India, with a NOx-to-VOC transition regime over central India and IGP. The study highlights a need to initiate in situ observations of VOCs over regional hotspots (Northeast, Central India, and the western coast) based on WRF-Chem results, where different satellite-based observations differ significantly.

Surface ozone in the Doon Valley of the Himalayan foothills during spring.

Elevated ozone (O3) pollution is observed every spring over the Northern Indian region including the Himalayan foothills, with a maximum typically in the month of May. However, studies investigating influences of photochemistry and dynamics in the valleys of Central Himalaya are limited. Here, in situ surface O3 observations conducted at Dehradun (77.99° E, 30.27° N, 600 m above mean sea level) in the Doon Valley during April-July 2018 are presented. These O3 observations reveal the prevalence of an urban environment over Dehradun with enhanced levels during noontime (66.4 ppbv ± 11.0 ppbv in May) and lower levels during night (26.7 ppbv ± 11.5 ppbv). Morning time O3 enhancement rate at Dehradun (7.5 ppbv h-1) is found to be comparable to that at Bode (7.3 ppbv h-1) in another valley of Himalayan foothills (Kathmandu), indicating stronger anthropogenic emissions in the Doon Valley as well. Daily average O3 at Dehradun varied in the range of 13.7-71.3 ppbv with hourly values reaching up to 103.1 ppbv during the study period. Besides the in situ photochemical O3 production, the entrainment of O3-rich air through boundary layer dynamics also contributes in noontime O3 enhancement in the Doon Valley. Monthly average O3 at Dehradun (49.3 ppbv ± 19.9 ppbv) is observed to be significantly higher than that over urban sites in Northern India (35-41 ppbv) and Bode (38.5 ppbv) in the Kathmandu Valley during May. O3 photochemical buildup, estimated to be 30.3 ppbv and 39.7 ppbv during April and May, respectively, is significantly lower in June (21.2 ppbv). Copernicus Atmosphere Monitoring Service (CAMS) model simulations successfully reproduce the observed variability in noontime O3 at Dehradun (r = 0.86); however, absolute O3 levels were typically overestimated. The positive relationship between CAMS O3 and CO (r = 0.65) together with an O3/CO slope of 0.16 is attributed to the influences of biomass burning besides anthropogenic emissions on observed O3 variations in the Doon Valley. O3 observations show an enhancement by 35-56% at Dehradun during a high-fire activity period in May 2018 as compared to a low-fire activity period over the Northern Indian region in agreement with the enhancement found in CAMS O3 fields (10-65%) over the region in the vicinity of Dehradun.

Decadal changes in surface ozone at the tropical station Thiruvananthapuram (8.542° N, 76.858° E), India: effects of anthropogenic activities and meteorological variability.

This paper presents the first observational results from an Indian station on the long-term changes in surface ozone (O3)-a major environmental pollutant and green house gas-over a period of about 40 years. It is based on the in situ measurements carried out during 1973-1975, 1983-1985, 1997-1998 and 2004-2014 at the tropical coastal station, Thiruvananthapuram. From 1973 to 1997, surface O3 shows a slow increase of ~ 0.1 ppb year-1 and a faster increase of 0.4 ppb year-1 afterwards till 2009 after which it showed a levelling off till 2012 followed by a minor decrease. The highest rate of increase is observed during 2005 to 2009 (2 ppb year-1), and the overall increase from 1973 to 2012 is ~ 10 ppb. The increase in day time O3 (peak O3) is estimated as 0.42 ppb year-1 during 1997-2012 and 2.93 ppb year-1 during 2006-2012. Interestingly, the long-term trend in O3 showed seasonal dependence which is more pronounced during O3 peaking seasons (winter/summer). The observed trends were analysed in the light of the changes in NO2, a major outcome of anthropogenic activities and methane which has both natural and anthropogenic sources and also meteorological parameters. Surface O3 and NO x exhibited positive association, but with varying rate of increase of O3 for NO x < 4 and > 4 ppb. Methane, a precursor of O3 also showed increase in tune with O3. Unlike many other high-latitude locations, meteorology plays a significant role in the long-term trends in O3 at this tropical site with water vapour abundance and strong solar irradiance which favour photochemistry. A comparison with the corresponding changes in the satellite-retrieved tropospheric column O3 (TCO) also showed an increase of 0.03 DU year-1 during 1996-2005 which enhanced to 0.12 DU year-1 after 2005. Both surface O3 and satellite-retrieved TCO were positively correlated with daily maximum temperature, increasing at the rate of 1.54 ppb °C-1 and 1.9 DU °C-1, respectively, on yearly basis. Surface O3 is found to be negatively correlated with water vapour content (ρv) at this tropical site, but at higher levels of ρv, O3 shows a positive trend.