Search

Nitrogen oxides (NOx = NO + NO2) are of great concern due to their impact on human health and the environment. In recent years, machine learning (ML) techniques have been widely used for surface NO2 estimation with rapid developments in computational power and big data. However, the uncertainties inherent to such retrievals are rarely studied. In this study, a novel ML framework has been developed, enhanced with uncertainty quantification techniques, to estimate surface NO2 and provide corresponding data‐induced uncertainty. We apply the Boosting Ensemble Conformal Quantile Estimator (BEnCQE) model to infer surface NO2 concentrations over Western Europe at the daily scale and 1 km spatial resolution from May 2018 to December 2021. High NO2 mainly appears in urban areas, industrial areas, and roads. The space‐based cross‐validation shows that our model achieves accurate point estimates (r = 0.8, R2 = 0.64, root mean square error = 8.08 μg/m3) and reliable prediction intervals (coverage probability, PI‐50%: 51.0%, PI‐90%: 90.5%). Also, the model result agrees with the Copernicus Atmosphere Monitoring Service (CAMS) model. The quantile regression in our model enables us to understand the importance of predictors for different NO2 level estimations. Additionally, the uncertainty information reveals the extra potential exceedance of the World Health Organization (WHO) 2021 limit in some locations, which is undetectable by only point estimates. Meanwhile, the uncertainty quantification allows assessment of the model's robustness outside existing in‐situ station measurements. It reveals challenges of NO2 estimation over urban and mountainous areas where NO2 is highly variable and heterogeneously distributed. Plain Language Summary: Inferring surface NO2 concentrations is an effective way to monitor and mitigate NOx pollution which is of great concern due to its impact on human health and the environment. Machine learning (ML) techniques have been widely used for surface NO2 estimation with rapid developments in computational power and big data. However, such estimations can be uncertain due to inherent errors in the data, and this uncertainty is rarely studied. We develop a novel ML framework to estimate surface NO2 concentrations and provide corresponding uncertainty information. We infer surface NO2 levels over Western Europe at the daily scale and 1 km spatial resolution from May 2018 to December 2021. Our model's performance is reliable as verified by in‐situ station measurements and an independent physics‐based model. We observe NO2 hotspots over urban areas, industrial areas, and major roads. The uncertainty quantification (UQ) techniques allow us to analyze the influence of different input data on estimating different NO2 levels. The UQ also helps to identify potential NO2 exceedances of the WHO 2021 limit, which have not been observed in previous research. Additionally, we assess the model's robustness outside of in‐situ stations and witness the challenge of NO2 estimation over urban and mountainous areas. Key Points: A novel and reliable machine learning model with uncertainty quantification is applied to infer surface NO2 levels over Western EuropeOur work uncovers how predictors impact model inference of various surface NO2 levels differentlyOur approach identifies areas of high uncertainty in surface NO2 mapping and potential environmental risks from overlooking uncertainty [ABSTRACT FROM AUTHOR]

Airborne imaging differential optical absorption spectroscopy (DOAS), ground-based stationary DOAS, and car DOAS measurements were conducted during the S5P-VAL-DE-Ruhr campaign in September 2020. The campaign area is located in the Rhine-Ruhr region of North Rhine-Westphalia, western Germany, which is a pollution hotspot in Europe comprising urban and large industrial sources. The DOAS measurements are used to validate spaceborne NO 2 tropospheric vertical column density (VCD) data products from the Sentinel-5 Precursor (S5P) TROPOspheric Monitoring Instrument (TROPOMI). Seven flights were performed with the airborne imaging DOAS instrument for measurements of atmospheric pollution (AirMAP), providing measurements that were used to create continuous maps of NO 2 in the layer below the aircraft. These flights cover many S5P ground pixels within an area of 30 km × 35 km and were accompanied by ground-based stationary measurements and three mobile car DOAS instruments. Stationary measurements were conducted by two Pandora, two Zenith-DOAS, and two MAX-DOAS instruments. Ground-based stationary and car DOAS measurements are used to evaluate the AirMAP tropospheric NO 2 VCDs and show high Pearson correlation coefficients of 0.88 and 0.89 and slopes of 0.90 ± 0.09 and 0.89 ± 0.02 for the stationary and car DOAS, respectively. Having a spatial resolution of about 100 m × 30 m, the AirMAP tropospheric NO 2 VCD data create a link between the ground-based and the TROPOMI measurements with a nadir resolution of 3.5 km × 5.5 km and are therefore well suited to validate the TROPOMI tropospheric NO 2 VCD. The observations on the 7 flight days show strong NO 2 variability, which is dependent on the three target areas, the day of the week, and the meteorological conditions. The AirMAP campaign data set is compared to the TROPOMI NO 2 operational offline (OFFL) V01.03.02 data product, the reprocessed NO 2 data using the V02.03.01 of the official level-2 processor provided by the Product Algorithm Laboratory (PAL), and several scientific TROPOMI NO 2 data products. The AirMAP and TROPOMI OFFL V01.03.02 data are highly correlated (r=0.87) but show an underestimation of the TROPOMI data with a slope of 0.38 ± 0.02 and a median relative difference of - 9 %. With the modifications in the NO 2 retrieval implemented in the PAL V02.03.01 product, the slope and median relative difference increased to 0.83 ± 0.06 and + 20 %. However, the modifications resulted in larger scatter and the correlation decreased significantly to r=0.72. The results can be improved by not applying a cloud correction for the TROPOMI data in conditions with high aerosol load and when cloud pressures are retrieved close to the surface. The influence of spatially more highly resolved a priori NO 2 vertical profiles and surface reflectivity are investigated using scientific TROPOMI tropospheric NO 2 VCD data products. The comparison of the AirMAP campaign data set to the scientific data products shows that the choice of surface reflectivity database has a minor impact on the tropospheric NO 2 VCD retrieval in the campaign region and season. In comparison, the replacement of the a priori NO 2 profile in combination with the improvements in the retrieval of the PAL V02.03.01 product regarding cloud heights can further increase the tropospheric NO 2 VCDs. This study demonstrates that the underestimation of the TROPOMI tropospheric NO 2 VCD product with respect to the validation data set has been and can be further significantly improved. [ABSTRACT FROM AUTHOR]

Airborne imaging differential optical absorption spectroscopy (DOAS), ground-based stationary and car DOAS measurements were conducted during the S5P-VAL-DE-Ruhr campaign in September 2020. The campaign area is located in the Rhine-Ruhr region of North Rhine-Westphalia,Western Germany, which is a pollution hotspot in Europe comprising urban and large industrial emitters. The measurements are used to validate space-borne NO2 tropospheric vertical column density data products from the Sentinel-5 Precursor (S5P) TROPOspheric Monitoring Instrument (TROPOMI). Seven flights were performed with the airborne imaging DOAS instrument formeasurements of atmospheric pollution (AirMAP), providing measurements which were used to create continuous maps of NO2 in the layer below the aircraft. These flights cover many S5P ground pixels within an area of 30 kmx 35 km and were accompanied by ground-based stationary measurements and three mobile car DOAS instruments. Stationary measurements were conducted by two Pandora, two zenith-sky and two MAX-DOAS instruments distributed over three target areas. Ground-based stationary and car DOAS measurements are used to evaluate the AirMAP tropospheric NO2 vertical column densities and show high Pearson correlation coefficients of 0.87 and 0.89 and slopes of 0.93±0.09 and 0.98±0.02 for the stationary and car DOAS, respectively. Having a spatial resolution of about 100mx 30m, the AirMAP tropospheric NO2 vertical column density (VCD) data creates a link between the ground-based and the TROPOMI measurements with a resolution of 3.5 kmx 5.5 km and is therefore well suited to validate the TROPOMI tropospheric NO2 VCD. The measurements on the seven flight days show strong NO2 variability, which is dependent on the different target areas, the weekday, and the meteorological conditions. The AirMAP campaign dataset is compared to the TROPOMI NO2 operational off-line (OFFL) V01.03.02 data product, the reprocessed NO2 data, using the V02.03.01 of the official L2 processor, provided by the Product Algorithm Laboratory (PAL), and several scientific TROPOMI NO2 data products. The TROPOMI data products and the AirMAP data are highly correlated with correlation coefficients between 0.72 and 0.87, and slopes of 0.38±0.02 to 1.02±0.07. On average, TROPOMI tropospheric NO2 VCDs are lower than the AirMAP NO2 results. The slope increased from 0.38±0.02 for the operational OFFL V01.03.02 product to 0.83±0.06 after the improvements in the retrieval of the PAL V02.03.01 product were implemented. Different auxiliary data, such as spatially higher resolved a priori NO2 vertical profiles, surface reflectivity and the cloud treatment, are investigated using scientific TROPOMI tropospheric NO2 VCD data products to evaluate their impact on the operational TROPOMI NO2 VCD data product. The comparison of the AirMAP campaign dataset to the scientific data products shows that the choice of surface reflectivity data base has a minor impact on the tropospheric NO2 VCD retrieval in the campaign region and season. In comparison, the replacement of the a priori NO2 profile in combination with the improvements in the retrieval of the PAL V02.03.01 product regarding cloud heights has a major impact on the tropospheric NO2 VCD retrieval and increases the slope from 0.88±0.06 to 1.00±0.07. This study demonstrates that the underestimation of the TROPOMI tropospheric NO2 VCD product with respect to the validation dataset has been and can be further significantly improved. [ABSTRACT FROM AUTHOR]

The Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) is employed as an intercomparison tool for validating Tropospheric Monitoring Instrument (TROPOMI) satellite NO2 retrievals against high-resolution Airborne Prism EXperiment (APEX) remote sensing observations performed in June 2019 in the region of Antwerp, a major hotspot of NO2 pollution in Europe. The model is first evaluated using meteorological and chemical observations in this area. Sensitivity simulations varying the model planetary layer boundary (PBL) parameterization were conducted for a 3-day period in June 2019, indicating a general good performance of most parameterizations against meteorological data (namely ceilometer, surface meteorology and balloon measurements), except for a moderate overestimation (~1 m s-1) of near-surface wind speed. On average, all but one PBL schemes reproduce fairly well the surface NO2 measurements at stations of the Belgian Interregional Environmental Agency, although surface NO2 is generally underestimated during the day (between -4.3 and -25.1 % on average) and overestimated at night (8.2–77.3 %). This discrepancy in the diurnal evolution arises despite (1) implementing a detailed representation of the diurnal cycle of emissions (Crippa et al., 2020), and (2) correcting the modelled concentrations to account for measurement interferences due to NOy reservoir species, which increases NO2 concentrations by about 20 % during the day. The model is further evaluated by comparing a 15-day simulation with surface NO2, NO, CO and O3 data in the Antwerp region. The modelled daytime NO2 concentrations are more negatively biased during weekdays than during weekends, indicating a misrepresentation of the weekly temporal profile applied to the emissions, obtained from Crippa et al. (2020). Using a mass-balance approach, we determined a new weekly profile of NOx emissions, leading to a homogenization of the relative bias among the different weekdays. The ratio of weekend to weekday emissions is significantly lower in this updated profile (0.6) than in the profile based on Crippa et al. (2020) (0.84). Comparisons with remote sensing observations generally show a good reproduction of the spatial patterns of NO2 columns by the model. Both APEX and TROPOMI columns are underestimated on the 27/6, whereas no significant bias is found on the 29/6. The two datasets are intercompared by using the model as an intermediate platform to account for differences in vertical sensitivity through the application of averaging kernels. The derived bias of TROPOMI v1.3.1 NO2 with respect to APEX is about -10 % for columns between (6–12)x1015 molec. cm-2. The obtained bias for TROPOMI v1.3.1 increases with the NO2 column, following C APEX = 1.217 C v1.3 - 0.783x1015 molec. Cm-2, in line with previous validation campaigns. The bias is slightly lower for the reprocessed TROPOMI v2.3.1, with C APEX = 1.055 C PAL - 0.437x1015 molec. cm-2 (PAL). Finally, a mass balance approach was used to perform a crude inversion of NOx emissions, based on 15-day averaged TROPOMI columns. The emission correction is conducted only in regions with high columns and high sensitivity to emission changes, in order to minimize the errors due to wind transport. The results suggest emissions increases over Brussels-Antwerp (+20 %), Ruhr Valley (13 %), and especially Paris (+39 %), and emission decreases above a cluster of power plants in West Germany. [ABSTRACT FROM AUTHOR]