Using satellite measurements and mesoscale modelling to 1 understand the contribution to an extreme air pollution 2 event in India 3 4

1 Several ambient air quality records corroborate severe and persistent degradation of air quality 2 over North India during the winter months with evidence of a continued increasing trend of 3 pollution across the Indo-Gangetic Plain (IGP) over the past decade. A combination of 4 atmospheric dynamics and uncertain emissions, including the post-monsoon agricultural 5 stubble burning, make it challenging to resolve the role of each individual factor. Here we 6 demonstrate the potential use of an atmospheric transport model, the Weather Research and 7 Forecasting model coupled with chemistry (WRF-Chem) to identify and quantify the role of 8 transport mechanisms and emissions on the occurrence of the pollution events. The 9 investigation is based on the use of CO observations from TROPOspheric Monitoring 10 Instrument (TROPOMI), onboard the Sentinel 5-Precursor satellite, and the surface 11 measurement network as well as WRF-Chem simulations to investigate the factors contributing 12 to CO enhancement over India during November 2018. We show that the simulated column13 averaged dry air mole fraction (XCO) is largely consistent with TROPOMI observations with a 14 spatial correlation coefficient of 0.87. The surface-level CO concentrations show larger 15 sensitivities to boundary layer dynamics, wind speed, and diverging source regions, leading to 16 a complex concentration pattern and reducing the observation-model agreement with a 17 correlation coefficient ranging from 0.41 to 0.60 for measurement locations across the IGP. We 18 find that daily satellite observations can provide a first-order inference of the CO transport 19 pathways during the enhanced burning period, and this transport pattern is reproduced well in 20 the model. By using the observations and employing the model at a comparable resolution, we 21 confirm the significant role of atmospheric dynamics as well as residential, industrial and 22 commercial emissions in the production of the exorbitant level of air pollutants in North India. 23 We find that biomass burning plays only a minimal role in both column and surface 24 enhancements of CO, except for in the state of Punjab during the high pollution episodes. 25 While the model reproduces observations reasonably well, a better understanding of the factors 26 controlling the model uncertainties is essential to relate the observed concentrations to the 27 underlying emissions. Overall, our study emphasizes the importance of undertaking rigorous 28 policy measures, mainly focusing on reducing residential, commercial and industrial emissions 29 in addition to actions already underway in the agricultural sectors. 30 31 32 33 2 https://doi.org/10.5194/acp-2020-1034 Preprint. Discussion started: 26 October 2020 c © Author(s) 2020. CC BY 4.0 License.


Abstract 1
Several ambient air quality records corroborate severe and persistent degradation of air quality 2 over North India during the winter months with evidence of a continued increasing trend of 3 pollution across the Indo-Gangetic Plain (IGP) over the past decade. A combination of 4 atmospheric dynamics and uncertain emissions, including the post-monsoon agricultural 5 stubble burning, make it challenging to resolve the role of each individual factor. Here we 6 demonstrate the potential use of an atmospheric transport model, the Weather Research and 7 Forecasting model coupled with chemistry (WRF-Chem) to identify and quantify the role of 8 transport mechanisms and emissions on the occurrence of the pollution events. The 9 investigation is based on the use of CO observations from TROPOspheric Monitoring 10 Instrument (TROPOMI), onboard the Sentinel 5-Precursor satellite, and the surface 11 measurement network as well as WRF-Chem simulations to investigate the factors contributing 12 to CO enhancement over India during November 2018. We show that the simulated column-13 averaged dry air mole fraction (XCO) is largely consistent with TROPOMI observations with a 14 spatial correlation coefficient of 0.87. The surface-level CO concentrations show larger 15 sensitivities to boundary layer dynamics, wind speed, and diverging source regions, leading to 16 a complex concentration pattern and reducing the observation-model agreement with a 17 correlation coefficient ranging from 0.41 to 0.60 for measurement locations across the IGP. We 18 find that daily satellite observations can provide a first-order inference of the CO transport 19 pathways during the enhanced burning period, and this transport pattern is reproduced well in 20 the model. By using the observations and employing the model at a comparable resolution, we 21 confirm the significant role of atmospheric dynamics as well as residential, industrial and 22 commercial emissions in the production of the exorbitant level of air pollutants in North India. 23 We find that biomass burning plays only a minimal role in both column and surface 24 enhancements of CO, except for in the state of Punjab during the high pollution episodes. 25 While the model reproduces observations reasonably well, a better understanding of the factors 26 controlling the model uncertainties is essential to relate the observed concentrations to the 27 underlying emissions. Overall, our study emphasizes the importance of undertaking rigorous 28 policy measures, mainly focusing on reducing residential, commercial and industrial emissions 29 in addition to actions already underway in the agricultural sectors. 30 31 32 33

Introduction 1
Biomass burning (BB) has been recognized as the second-largest source of radiatively and 2 chemically active trace gases (e.g. CO, CO 2 , and SO 2 ) and aerosols (e.g. PM10, and PM2.5) in 3 the global atmosphere, which has significant implications for climatic change and human 4 health (Andreae, 2001;Bond, 2004 there would be a reduction of about 40 ppm CO 2 from the current atmospheric concentration 10 level, indicating the importance of fire activities for the global carbon budget. 11 In India, emissions from open-biomass burning include significant contributions from 12 agricultural crop residue burning in addition to forest and grassland fires and play an essential 13 role in terms of releasing total carbon content to the atmosphere. Agricultural stubble burning 14 during the post-harvesting period is one of the main kinds of biomass burning practices used in 15 India to clear the land to make it suitable for the next crop (Tai-Yi, 2012; Zha et al., 2013). 16 According to previous estimates, crop waste open burning, which includes its use in residential 17 heating and cooking, is responsible for 78-83% (116-289 Tg yr -1 ) of the total biomass burned 18 in India during the year 2001 while rest of the contributions are from forest fires 19 (Venkataraman et al., 2006 (Yadav et al., 2018). Since agricultural stubble burning is a practice prohibited by law in 1 India, official surveys conducted to estimate the extent of fire emission are not reliable. There 2 is, therefore, a critical need to improve the current knowledge base to help to make future 3 policies and implement mitigation strategies. observations with unprecedented data density at high spatial and temporal resolution paves the 14 more direct way for a detailed study on the origin, distribution and extent of trace gas levels 15 over a vast domain on a monthly to daily basis. Carbon monoxide (CO) is one of the major 16 gases emitted from biomass burning and incomplete fossil fuel combustion. The major sink of 17 CO is reaction with the hydroxyl radical (OH) to form CO 2 and precursor tropospheric ozone.

18
The lifetime of CO in the atmosphere is between several weeks and several months and varies 19 with the location and season depending on the oxidizing capacity of the environment (Jaffe, 20 1968). Compared to CO 2 and CH 4, the short lifetime of CO makes it easier to detect from the 21 background concentration level and thus it can be a good tracer of pollution transport (Dekker 22 et al., 2017). Therefore, CO can be used as a proxy for the anthropogenic emissions of other 23 pollutants, for example, emissions of important GHGs such as carbon dioxide (Gamnitzer et 24 al., 2006 In this study, we make use of carbon monoxide (CO) observations from TROPOMI (see Sect. 36 2.1) and the surface measurement network to investigate different regional sources of CO in 37 terms of their contribution to the total column and surface-level concentrations during high 38 pollution episodes in the winter season. By comparing CO measurements with high-resolution 39 model simulations generated by WRF-Chem-GHG, we aim to understand the contribution of 40 different sources to the observed CO enhancement. In particular, we focus on CO enhancement 41 caused by the emissions from both biomass burning and anthropogenic activities and their 42 relative roles in the severe air pollution of major cities nearby. This paper aims to address the 1 following questions: 1) How large is the CO enhancement over northern India detected by 2 TROPOMI during the agricultural stubble burning period? 2) What is the regional contribution 3 of CO emissions over India  two observational products over India. The SICOR and WFMD-DOAS algorithms differ in 40 many aspects including radiative transfer models, inversion schemes and the quality filtering 41 method used. Whereas WFMD retrievals are limited to cloud free scenes, SICOR aims to 1 retrieve CO columns for cloudy ground pixels also. A global comparison between these two 2 datasets from December 2018 (Schneising et al., 2019) shows a very similar spatial CO pattern 3 for both algorithms with a high correlation coefficient of 0.98 and a regression factor close to 4 the 1:1 line, confirming good agreement between the two datasets. An overview of the 5 TROPOMI datasets used in this study is provided in Table 1 and additional details are provided 6 in the following two sub-sections. 7

Scientific TROPOMI WFMD CO product 8
The WFM-DOAS retrieval algorithm was initially developed for the SCIAMACHY instrument

Operational TROPOMI/SICOR CO product 20
The operational TROPOMI/SICOR CO product (referred to as SICOR hereafter) is retrieved found with CAMS with a standard deviation of 6% and a Pearson correlation coefficient of 0.9 26 (Borsdorff, 2018a). As per the validation of SICOR with ground-based total column 27 measurements of TCCON, a mean bias of 6 ppb with a standard deviation of 3.9 ppb and 2.4 28 ppb has been found for clear and cloudy skies respectively (Borsdorff, 2018a). 29

Ground-level observations 30
To assess the model performance against the surface level measurements, we use 31 measurements from ground-based air quality monitoring network maintained by the Central 32 Pollution Control Board (CPCB) of India. The measurements of CO are performed using CO 33 analysers based on non-dispersive infrared spectroscopy, and the data are provided as 6-hour 34 averages via a publicly-accessible online portal (https://app.cpcbccr.com/ccr/#/caaqm-35 dashboard-all/caaqm-landing/data). Though we have analysed CO measurements available 36 from all stations for the period of 3-20 November 2018, measurement stations that are too close 37 to local emissions sources showing extremely large and ambiguous variations in which stability 38 of the analyser may be questioned, were excluded for the evaluation. All the stations used for 39 this evaluation are listed in Table 2.

WRF-Chem-GHG model 1
We utilize a high-resolution modelling framework based on a WRF-Chem-GHG (version 2 3.9.1.1, hereafter referred to as WRF) for simulating CO concentrations at a spatial resolution 3 of 10 km × 10 km) and a temporal resolution of 1 hour. The model solves the compressible 4 Euler non-hydrostatic equations and uses a terrain-following hydrostatic pressure coordinate 5 system in the vertical direction (Skamarock et al., 2008). In our case, simulations have 39 6 vertical levels extending from the surface to 50 hPa (~20 km) and the model domain describes 7 a region with a spatial extent of 3500 km × 2500 km, covering the Indian domain and some 8 parts of Bangladesh, China, Nepal and Pakistan. 9 For meteorological initial and boundary conditions, we have taken ECMWF ERA5 data on an 10 hourly basis with a horizontal resolution of 0.25°× 0.25°. For CO concentration fields, initial 11 and boundary conditions are prescribed from the Copernicus Atmosphere Monitoring Service 12 (CAMS re-analysis data). CAMS provides the estimated mixing ratios of CO with a spatial of CO and concluded that the CO enhancement pattern is hardly affected by VOCs and OH 10 oxidation. 11

Comparison of WRF simulations with satellite column observations 13
To evaluate the performance of WRF, we have performed a comparison study on a daily and 14 monthly basis using WFMD column CO (XCO) data during the period 1-30 November 2018 15 over the Indian domain. The WFMD dataset also provides the column averaging kernel vector 16 (AK), describing the vertical sensitivity of the retrieved CO column to the partial column at 17 different vertical levels (Schneising et al., 2019). In order to compare the satellite data with 18 model simulations quantitatively, we have to use the AK to take into account the vertical 19 sensitivity of the instrument. In the dataset, the elements of the AK mostly have values close to 20 1, meaning that the instrument is sensitive to the full column of CO. As such, the prior 21 estimates have a negligible contribution to the retrieved columns. To compare the simulated 22 concentration fields with the satellite observations, the simulated pressure-weighted column-23 averaged dry air mole fraction after applying the averaging kernel, !"#$ is calculated as 24 follows: 25 In this equation, l is the index of the vertical layer and n is the number of vertical layers, and ! 28 the corresponding column-averaging kernel of the WFMD algorithm. is the pressure-29 weighted column averaged dry air mole fraction calculated from model simulations. ! is the a 30 priori dry air mole fraction profile used by the WFMD retrieval algorithm, which is also 31 provided in the data product, and is the model simulation. ! is the mass of dry air for the 32 corresponding layer and ! is the total mass of dry air. For the comparison, we used only 33 WRF simulations that correspond to the satellite sampling time. For a fair comparison between 34 the satellite observations and model simulations, the averaging kernel matrix and a priori 35 profile for each retrieval have been applied to the corresponding model output as explained in 36 Eq. 1. For the ease of the statistical analysis, the observations, though comparable to the model 37 resolution, are gridded to the WRF spatial resolution of 10 km × 10 km. Both WFMD and 38 WRF averaged data for the month of November and a period of 6-9 November (enhanced 39 biomass burning period as per the GFAS data) are utilized in this study to investigate the 40 column enhancement by fire CO and their distribution over the study domain. During the 1 enhanced biomass-burning period, a definite enhancement in XCO is found over the biomass 2 burning hotspot. The monthly averaged map shows decreased concentration levels over these 3 hotspots, which is attributed to the CO concentration dispersion resulted by changing weather 4 conditions. 5

Comparison of WRF simulations with ground-level observations 6
To evaluate the model performance at surface level, we have performed a comparison study 7 with the CO in situ measurements obtained from the ground-level pollution measurement 8 stations. We use the data collected from 20 measurement stations within the IGP region and 9 evaluation is done against each station data. In order to see overall agreement for different 10 regions in the IGP, we have averaged the data temporally using only the stations within the 11 corresponding regions (Delhi, Punjab, and the IGP). The entire month is not used here due to 12 the existence of data gaps from several stations. In order to avoid very localised influence and 13 noise in the observed data, the 1-hourly datasets are temporally averaged to 6-hourly 14 resolution. 15

Regional and seasonal variation of fire CO emission 17
In order to examine the spatio-temporal variations of the monthly fire CO emission, we have 18 divided the entire region into five sub-regions as shown in Fig. 1 shows fewer emissions during the whole year. However, emission spikes are seen in the IGP 32 during the October-November (post-monsoon) period. Over the IGP, the fire CO emissions 33 show evident monthly variations with a higher emission during the post-monsoon time 34 compared to the pre-monsoon period. About 73% of the country's total fire CO emissions 35 during the post-monsoon period are from the IGP region. Of these IGP post-monsoon 36 emissions, 70% come from the northwest states of the IGP: Punjab and Haryana. Over this 37 region, 25% of the total fire CO emissions happened within a short period during 6-9 38 November, which accounts for about 18% of the country's post-monsoon total fire CO 39 emissions. During the monsoon time, all regions are found to have fewer fire emissions, which 40 can be attributed to the fact that rainfall leads to suppressed fire activity. In addition to the 1 minimal possibility of fire activities during the rainy season, note that MODIS has only a 2 limited capability to detect fire emissions over a cloudy scene. during the short period of 6-9 November. 16 showing higher values during the biomass burning period than the monthly average. A distinct 22 enhancement in XCO can be observed during the biomass-burning period specifically over the 23 state of Punjab and Haryana, with a distribution plume towards the southeast direction 24

Enhanced XCO as observed by the satellite 17
including the region of Delhi and Agra. Note that this emission hotspot is also seen in the 25 GFAS inventory during the biomass-burning period (Fig 2). Consistency between the GFAS 26 inventory and satellite observations suggest that the XCO enhancement over the northwest part 27 of the IGP during 6-9 November can be attributed to the crop residue burning that occurred 28 over the Punjab region. The consistency check between two retrieval products (WFMD and 29 SICOR) has resulted in a very similar spatial CO pattern for both algorithms with a high 30 correlation coefficient of 0.97 confirming the robustness of our findings between the two 31 datasets over India (see Table 3). During early winter (November and December), the shallow 32 PBL and low wind speed cause locally-emitted gases to be trapped in the lower atmosphere, 33 which is considered to be the primary cause for high concentrations during this period. For a 34 better understanding of the role of transport and CO emissions from biomass burning to the 35 distribution over the domain, we utilized WRF model simulations and performed a comparison 36 study with the WFMD observations as explained in Sect. 4.1. 37

Agreement with column observations 39
We compared WRF simulations with WFMD observations, averaged over the days of peak 1 burning and over the full month of November 2018. Fig. 3 demonstrates these comparisons. 2 Both satellite and the model show a higher level of column CO over the IGP region than over 3 any other region of the domain. In the monthly averaged plots, the model slightly overestimates 4 (by about 10 ppb) the XCO in most parts of the domain. Between the monthly averaged 5 observations and the simulations, we find a mean difference of 7 ppb with a standard deviation 6 of 8 ppb and a correlation coefficient of 0.87 (Fig. 4). XCO is reported in both observations and simulations over the state of Punjab, starting from 6 13 November and gradually increases in the following days. During this period, the plume is seen 14 to be partly transported in a southeast direction along the region of Delhi and Agra. Over the 15 IGP, there exists an overall slight underestimation by WRF in comparison to TROPOMI during 16 this period with a mean model-to-observation difference of -2.7 ppb. 17 Figure 6 shows the temporal evolution of the CO concentration in three cities (Barnala, New 18 Delhi, and Agra) located along the transport pathway of pollution. The data are averaged in a 19 100 km x 100 km square around the centre of each city. During the biomass-burning period, 20 the XCO over Barnala (Punjab) shows a steady positive increment with time with a peak on 9 21 November with a value of approximately 165 ppb. Both observations and simulations suggest a 22 southeast transport of this plume that increases the CO concentration over Delhi and Agra 23 during 8 and 9 November. Over Delhi, the WFMD XCO reached a maximum on 8 November 24 while modelled CO showed a delay, with a maximum concentration on 9 November. On 9 25 November, observation shows more dispersed XCO over Delhi towards the southeast direction 26 in comparison with model simulations. Over Agra, which is located far away from the 27 pollution hotspot but along the transport pathway, an increase in XCO, which is consistent with 28 that over the other two cities is found. The details in Table 3 confirm the minimal impact of 29 differences in satellite retrieval algorithms on our results. This analysis suggests a promising 30 usage of TROPOMI observations to understand the details of hotspot emissions and the 31 distribution of transport. The model is able to capture many of these spatial and temporal 32 patterns, supporting the potential use of WRF via inverse modelling to infer hotspot emissions 33 using column measurements. 34 Figure 7 shows the model evaluation with ground-level measurements over the regions IGP, 36

Agreement with ground-level observations 35
Delhi and Punjab for a period from 3 to 20 November 2018. The location of ground-level 37 measurement stations used for this study is shown in Fig. 8. The entire month is not used here 38 due to the existence of data gaps from several stations. Taking various ground-based stations 39 over the IGP, Delhi and Punjab, we see an overall good agreement between model and 40 measurements, with a correlation coefficient of 0.6 (for the IGP), 0.6 (Delhi) and 0.41 41 (Punjab). Among these three study regions, a lower correlation is found for the Punjab region 1 in which measurement sites are very close to the biomass burning hotspots, therefore showing a 2 larger variability compared to other stations. These variations are not fully reproduced by the 3 model, resulting in lower correlations over Punjab region. Though the model is able to follow 4 the temporal variation in the surface level CO concentrations, overall underestimations of 9 5 ppb and 54 ppb are found for Punjab and Delhi. For the IGP region, the model underestimates 6 the observed enhancements considerably, resulting in a mean bias of 162 ppb. The observed 7 underestimation of WRF can be attributed to the local source enhancements at the ground-level 8 stations, which are closely located to the cities. For the Punjab region, the model CO surface 9 concentration shows the influence of biomass burning starting from 6 November with a 10 maximum of 800 ppb on 8 November. Unlike the Punjab region, the concentration patterns 11 over Delhi and the IGP show a steadily increasing trend from 6 to 13 November, with a 12 subsequent reduction in mixing ratios for the remaining days. Among these study regions 13 during this period, the lowest and highest surface CO levels are observed over the regions 14 Punjab (mean: 500 ppb) and Delhi (mean: 1500 ppb) respectively. Except for Punjab, we see fires, including biomass burning over India (Cusworth et al., 2018). Overall, the results show 23 that the model simulation at a high spatial resolution is capable of capturing the CO 24 enhancement and reduction pattern at most of the stations, however there is a non-trivial mean 25 bias which can be attributed to issues with simulating transport and PBL dynamics in WRF as 26 well as the variability in emission fluxes which is likely to be not sufficiently well represented 27 in the emission inventories used. 28

Contribution of different sources to the observed concentration 29
To further investigate the contribution of different emission sources to the observations, we use 30 the "tagged-tracer" option in WRF and separate the contributions from different sources as 31 shown in Fig. 6 and 7. Note that the signals contributing to satellite observations are difficult to 32 disentangle without underlying assumptions or the availability of multi-tracers such as CO and 33 NO x and NO y * (NO y * includes NO x , PAN, organic nitrates, HNO 3 , and N 2 O 5 , e.g. Wang et al., 34 2002). The relative contributions of different emission sources and processes to the WRF CO 35 column, as summarized in Table 4, clearly indicate the dominance of anthropogenic signals 36 over biomass burning signals on the XCO enhancements. The significant impact of background 37 signal owing to the advection from the domain boundary throughout the column indicates the 38 influence of far-field fluxes and large-scale transport patterns on column CO (see Fig. 6).

39
During the biomass-burning period, there exists a considerable contribution of biomass burning 40 emissions to the column mixing ratios particularly over the Punjab region (14%). Relatively 41 low contributions of biomass burning signals to the column in Delhi and the IGP compared to 42 Punjab indicates the dominant contribution of surface CO emission to the column in Punjab  1 where the biomass emissions originated. It also suggests the possibility of less dilution of 2 surface emissions during wintertime, enhancing the total column mixing ratios. The effect of 3 advected biomass burning signals in terms of their contribution to the column can be seen over 4 Delhi (12%), however this effect becomes smaller in the IGP (5%) due to further dispersion. 5 The diurnal variation in the surface level CO concentration pattern is due to the diurnal 6 variation in the planetary boundary layer height (PBLH) combined with strong sources of CO 7 at the surface. The contribution from emissions sources over the Delhi, the IGP and Punjab 8 regions for the period of 3-20 and specifically 6-9 November are also summarised in Table 4. 9 For all regions, the influence of background CO concentrations to the observed variability is 10 minimal, as expected (see Fig. 7). The background influence is expected to be smaller for 11 surface CO in urban areas where the CO fraction from local anthropogenic emissions 12 dominates over the background signals. At ground level in Delhi and the IGP, a detectable 13 enhancement in surface CO due to fire CO is found only during 6-9 November. During this 14 period, the average contribution of biomass burning to the ground level concentration is 10%, 15 while the anthropogenic contribution is 79-83%. During 3-20 November over Delhi, however, 16 the average contribution from fire dropped to 4% compared to 85% in the case of the 17 anthropogenic contribution.

19
Overall, our findings suggest that the enhanced CO levels during pollution episodes over Delhi 20 and the greater part of IGP are affected by biomass burning. However, a more significant 21 contribution comes from anthropogenic emissions. Unlike the surface CO mixing ratios, the 22 majority of the column CO mixing ratio is contributed by the background signal. A recent 23 study conducted by Dekker et al., (2019) concluded that there exists an underestimation in 24 GFAS fire emission data over the Indian region, which is supported by our findings. However, 25 over the Punjab region, biomass burning played a significant role in determining the ground 26 level CO measurements, especially during 6-9 of November, during which enhanced fire 27 activities occurred. This has contributed considerably to the column mixing ratio that is 28 detected by TROPOMI. On average, for 3-20 of November, 17 % of the total ground level CO 29 concentration over the Punjab region are on account of fire CO emission, whereas for 6-9 30 November, the share is about 38%. 31

Effect of meteorology 32
Usually pollution episodes during winter are the result of meteorological conditions due to low 33 wind speed and shallow boundary layer (PBL height). Figure 9 demonstrates the influence of 34 the PBL height and surface-level wind speed to the observed CO level. We found a negative Lagrangian modelling framework, found a considerable impact of meteorological conditions 12 during November 2017 that contributed to the enhancements of trace gases over Delhi.

13
Together with strong emissions (anthropogenic and biomass burning), they found that these 14 enhancements could be several orders of magnitude higher compared to other seasons. 15

Conclusions 16
The Our analysis shows that daily observations from TROPOMI allow pollution transport from the 5 emissions hotspots to be captured. As an example, we analysed the pollution transport from the 6 fire emissions hotspots over northern India during the enhanced burning period of November 6-7 9. WFMD XCO level started to rise over the fire emission hotspots from 6 November and 8 gradually increased during the following days. Both WFMD and model simulations show the 9 transport of CO polluted air masses towards the northeast part of the IGP along with the capital 10 city Delhi. Due to this pollution transport, the CO concentration level in the cities along the 11 transport pathway shows CO enhancements. A similar transport pattern is also observed in our 12 WRF model simulation. This supports the reliability of WRF transport simulation and suggests 13 the potential of using WRF to estimate CO emission via flux inversions. The good agreement 14 between WFMD and SICOR retrievals over India confirms the robustness of our findings 15 irrespective of the differences in the retrieval algorithm.

17
For the further evaluation of WRF with surface measurements, we used ground level CO 18 measurements from the stations along the IGP for the period of 3-20 November 2018. The 19 comparison shows a good agreement between simulations and observations with a correlation 20 coefficient of 0.6 (for the IGP), 0.6 (Delhi) and 0.41 (Punjab). Over these regions, the surface 21 CO showed a steady increasing trend from 6 to 13 November, followed by a reduction in 22 mixing ratio in the following days. Among these study regions, the lowest and highest surface 23 CO level was observed over the regions Punjab and Delhi respectively.

25
Overall, our results imply a minimal role of biomass burning in terms of its contribution to 26 both column and surface enhancements compared to other anthropogenic sources, except for 27 the state of Punjab during the high pollution episodes. This is also consistent with Dekker et al.

28
(2019), which concluded that the low wind speeds and shallow atmospheric boundary layers 29 were the most likely causes for the temporal accumulation and subsequent dispersion of CO reduce residential and commercial emissions in addition to measures already being taken in the 10 agricultural sectors (e.g. the implementation of second-generation direct-seeders, such as the 11 Happy Seeder, which facilitate sowing under heavy stubble conditions, thereby avoiding the 12 need for residue burning, NAAS, 2017). The future task involves the implementation of 13 appropriate inverse techniques suitable for flux inversion of spatially resolved sources of CO 14 emissions over India.

16
Sahu, L. K., Sheel, V., Pandey, K., Yadav, R., Saxena, P. and Gunthe, S.: Regional biomass 23 burning trends in India: Analysis of satellite fire data, J. Earth Syst. Sci., 124 (7) Table 3. Comparison between WFMD and SICOR products over India during the burning 1 period and the full month of November 2018. Abbreviations N, MB, SD, and R correspond to 2 the number of observations, mean bias, standard deviation of differences, and correlation 3 coefficient respectively.