Large sub-regional differences of ammonia seasonal patterns over India reveal inventory discrepancies

Ammonia (NH3) is a key precursor of haze particles and fine particulate matter (PM2.5) and its spatiotemporal variabilities are poorly constrained. In this study, we present measurements of NH3 over the Indian subcontinent region from the Infrared Atmospheric Sounder Interferometer (IASI) and Cross-track Infrared Sounder (CrIS) satellite instruments. This region exhibits a complex emission profile due to the number of varied sources, including crop burning, fossil fuel combustion, fertilizer application, livestock and industrial sources. Observations from the CrIS and IASI instruments are oversampled to a resolution of 0.02° × 0.02°. Five regions with distinct spatiotemporal NH3 profiles are determined using k-means clustering. Maximum NH3 columns are seen in July over the western India with column densities of 6.2 × 1017 mol cm−2 and 7.2 × 1017 mol cm−2 respectively for IASI and CrIS. The seasonality of measured NH3 columns show annual maxima occurring in spring in Eastern India and Bangladesh and in mid-summer for the western Indo-Gangetic plain. Our observational constraints suggest that the impact of local farming practices on NH3 emissions is not well captured in emission inventories such as Coupled Model Intercomparison Project Phase 6 (CMIP6), which exhibits peaks in the late spring and autumn. The spatial variability in the seasonal patterns of NH3 is also not captured by the single emissions profile used in CMIP6 for India. The high-resolution maps obtained from these measurements can be used to improve NH3 emission inventories in order to understand its sources for more accurate predictions of air quality in the Indian subcontinent. Our study points to the need for regionally specific emissions inventories for short-lived species such as NH3 that have heterogeneous emissions profiles due to specific agricultural practices and other emission source characteristics.


Introduction
and is a key precursor to fine particulate matter (PM) (Gong et al 2013). Deposition of NH 3 into ecosystems can lead to eutrophication and acidification of the soils and waterways leading to decreased biodiversity (Leip et al 2015). NH 3 has a short lifetime of 0.5-5 days (Walker et al 2000) and exhibits a highly heterogeneous spatial distribution. Ammonia readily reacts with atmospheric acids, particularly nitric acid and sulfuric acid to form ammonium nitrate and ammonium sulfate respectively (Adams et al 1999). The short lifetime of NH 3 and the dependence of both formation and sinks on climactic conditions, such as atmospheric temperature and humidity, results in large uncertainties on both emissions and atmospheric concentrations (Hellsten et al 2018).
India experiences some of the worst air pollution in the world (World Health Organization 2016), and air quality concerns extend throughout much of the continent (Hammer et al 2020, Ravishankara et al 2020). Ammoniated aerosols over the Indian subcontinent are significant components of unhealthy PM (Kumar and Sarin 2010, Sudheer and Rengarajan 2015, Saraswati et al 2019, Ojha et al 2020, Acharja et al 2022. The Indo-Gangetic Plain (IGP) is a highly polluted area consisting of parts of Pakistan, northern India, southern Nepal and most of Bangladesh. Sources of aerosols in this region include biomass burning (including waste), vehicular emissions, industrial activity and desert dust . In particular, the IGP is a major emissions hotspot for NH 3 (Parashar et al 1998, Van Damme et al 2018. The region shows strong seasonal variability in PM (Awasthi et al 2011, Mahapatra et al 2018 due to unique meteorology, orography and seasonal emissions, such as crop burning. Crop clearance via burning is especially common in India (Jain et al 2014) where plant residues from common crops, such as wheat, barley and rice, are removed.
Although NH 3 is critical in the formation of PM and atmospheric aerosols, anthropogenic emissions are not widely regulated (Felix et al 2017 and NH 3 is not considered a criteria pollutant in the United States (Gilliam and Hall 2016), though is regulated in the European Union. The Central Pollution Control Board of India has a limited number of NH 3 sensors in its network (Pant et al 2019) and additional surface concentrations have been measured at the local scale, particularly in Delhi (Saraswati et al 2018, Kotnala et al 2020. As a result, there is limited spatial coverage in this region and specifically a lack of long-term surface measurements. Current available emissions inventories, both global and local, do not have sufficiently accurate NH 3 emissions for the Indian subcontinent for use with atmospheric models. NH 3 readily reacts with sulfuric acid and nitric acid, leading to the formation of ammonium sulfate and ammonium nitrate. The proportion of PM 2.5 made up of these secondary inorganic aerosols varies by region and season (Park et al 2021), mainly due to the strong dependence on temperature of ammonium nitrate formation (Pinder et al 2008). In order for atmospheric models to accurately predict the distribution of PM in India, more appropriate regional emissions inventories are required.
Improvements in satellite instruments and retrieval algorithms have enabled measurement of atmospheric NH 3 from space. Satellite remote sensing of atmospheric gases allows measurements with spatial coverage beyond that available from groundbased measurements. The first satellite measurements of lower tropospheric NH 3 were achieved with the Tropospheric Emission Spectrometer (TES) (Beer et al 2008, Shephard et al 2011 and subsequently the Infrared Atmospheric Sounder Interferometer (IASI) , the Cross-track Infrared Sounder (CrIS) (Shephard and Cady-Pereira 2015), the Atmospheric Infrared Sounder (AIRS) (Warner et al 2017), and Greenhouse Gases Observing Satellite (Someya et al 2020) have added to the suite of instruments measuring NH 3 from space. Satellite remote sensing has previously been used to study NH 3 in India, including annual trends (Van Damme et al 2021). Spatiotemporal analyses have also been carried out over India, including the IGP, on a monthly basis at IASI native resolution (Kuttippurath et al 2020) and at coarser resolutions for comparison with models (Pawar et al 2021). These analyses use preexisting political or geographical boundaries to study regional trends.
The work presented here follows a data-driven approach to determine distinct regions of NH 3 seasonality at high resolution achieved from oversampling many years of data from two independent satellite platforms, IASI and CrIS. The purpose of this study is to use satellite remote sensing to measure the spatiotemporal distribution of NH 3 in the Indian subcontinent and explore sub-regions where further, particularly bottom up, studies should be focused. Using the multi-year NH 3 products from the IASI and CrIS instruments we present spatial and seasonal variations in NH 3 columns over India, Pakistan and Bangladesh. In support of using IASI and CrIS measurements, a detailed analysis of the NH 3 seasonal cycle over France was performed using both IASI and CrIS (Viatte et al 2020), and IASI and CrIS showed similar spatiotemporal NH 3 patterns despite different overpass times, instruments and retrieval algorithms. A k-means clustering technique is employed on the oversampled satellite data to classify regions with approximately similar seasonality in this study area based on monthly means.

Method
The IASI instrument measures the N-H stretching mode of NH 3 at around 950 cm −1 . Three such instruments are in orbit on the Metop-A, -B and -C satellites in sun-synchronous orbits, each having two overpasses per day at a local time of approximately 0930 and 2130. As a result, IASI achieves global coverage two times per day with pixel size 12 km × 12 km at nadir (up to 20 km × 39 km at the edge of the swath) and a swath width of 2400 km. As thermal contrast is an important factor in instrument sensitivity and therefore retrieval quality, only the measurements from the morning overpasses of Metop-A and -B were used for this study due to limited thermal contrast for the nighttime overpasses. Measurements with more than 10% cloud contamination were not included. Version 3 of the IASI NH 3 retrievals have been used for this analysis for the years 2008(Franco et al 2018, Van Damme et al 2021. CrIS is a Fourier Transform spectrometer deployed on board the Suomi-NPP satellite, launched on October 2011. Suomi-NPP is in a sunsynchronous orbit with two overpasses per day at approximately 0130 and 1330, providing global coverage with a swath width of approximately 2200 km. Only measurements from the ascending node in the afternoon have been considered here. The instrument has a high spectral resolution of 0.625 cm −1 (Tobin et al 2013) and circular 3 × 3 pixels, with a 14 km diameter at nadir. CrIS has a similar spatial coverage to IASI (Shephard and Cady-Pereira 2015) but has decreased spectral noise in the 950 cm −1 spectral region (Zavyalov et al 2013). The detection limit of CrIS is approximately 0.9 ppbv (Shephard and Cady-Pereira 2015) and can reach down to 0.3 ppbv under ideal conditions (Kharol et al 2018). The CrIS product version 1.5 data were used for this analysis for the period 2013-2017 (Shephard et al 2020).
Direct validation of satellite remote sensed NH 3 has been conducted byground-based or aircraft measurements. For IASI, such validation has been performed in the USA (Guo et al 2021) and Europe, China and Africa (Van Damme et al 2015; Dammers et al 2016). For CrIS, ground-based fourier transform infrared spectroscopy (FTIR) has been used in sites in North America, Europe and Oceania (Dammers et al 2017).
An oversampling algorithm (Sun et al 2018, Van Damme et al 2018 was used to spatially and temporally average both datasets to a 0.02 × 0.02 • grid over the region 5-40 • N and 65-100 • E, covering the totality of India and Bangladesh and most of Pakistan. This method of spatial averaging leads to higher resolution and fewer gaps between observations at the price of temporal resolution. For this analysis, monthly averages for the entire ten and five year IASI and CriS data sets, respectively, were considered, and inter-annual variability was not taken into account. An iterative k-means clustering approach was used to analyze the spatiotemporal variability of NH 3 columns in the study region. k-means clustering techniques have been used for a number of geophysical applications to classify datasets (Wang et al 2021).
This k-means approach groups data into clusters whereby the averaged squared distance between data within a cluster are minimized (Forgy 1965, Arthur andVassilvitskii 2007). Thus, no prior assumed regions, such as political or geographical boundaries, are used to determine regionality of data, rather the clustering is driven by the data itself and minimizes instrument bias. This is particularly important for low column measurements.
Firstly, monthly NH 3 data were standardized to have a mean of 0 and a variance of 1 to avoid observations being grouped by magnitude or differences in mean or variance. The approach then uses k predefined centroids and allocates each value in the set (z n ) based on reducing the sum of squares of each cluster (C), so as to minimize J where, First, 20 seeds were randomly selected (land only) for clustering (i.e. k = 20) and the seasonality of each pixel in the study was compared to each of the seeds and allocated based on minimizing J. When each pixel had been classified to a cluster, the pixels in each cluster were averaged to calculate a new seed value. This process was iterated until convergence, where clusters did not change upon iteration. One issue with a k-means clustering approach is the initial selection of the number of seeds. For this study, k was set to be high (20). At iteration, if a cluster C contained only one pixel (i.e. it matched only with itself), it was removed for the next iteration. Through this method, the number of clusters ended at five for both IASI and CrIS. This process was repeated 10 000 times to test if selection of random seeds at the second step would affect the result. Though certain seed selections had minor affects (for example mountainous pixels, or those overlapping land and ocean), random seed selection did not significantly change the resulting clusters and therefore their seasonalities.

Results
The monthly oversampled high-resolution maps of NH 3 over the study area are presented in figure 1 for IASI and figure 2 for CrIS. A comparison between the NH 3 column densities of the two instruments is not the intention of this study, although a monthwise comparison is shown in figure S1 and the seasonality correlation is shown in figure S2. In general, the correlation is very strong from October to April and in regions with high column amounts. The correlation is weakest in the summer months, caused by regions of low NH 3 (southern India) and missing data in the monsoon season. The short lifetime of NH 3 and the different overpass times of the satellites makes such an assessment difficult and perhaps misleading; however, both independent datasets show the same general spatial and temporal variability of NH 3 over the region. The average column measurements over the highly polluted   area of the western IGP in its maximum during July are 4.07 × 10 16 and 3.94 × 10 16 molecules cm −2 and for the eastern IGP in its maximum during April are 1.76 × 10 16 and 1.60 × 10 16 molecules cm −2 for IASI and CrIS, respectively. In addition to the overpass time, sensor differences, instrument sensitivities and different retrieval approaches may result in different measured NH 3 columns. However, in general, the agreement in spatiotemporal distribution between the two sensors in this region is strong, similar to IASI/CrIS analyses in France (Viatte et al 2020).
Maximum NH 3 columns are seen in July over the western IGP with column densities of 6.2 × 10 17 molecules cm −2 for IASI and 7.2 × 10 17 molecules cm −2 for CrIS. These results agree spatially with annual average maps of satellite remote sensing of NH 3 over southern Asia (Van Damme et al 2014, Kuttippurath et al 2020, Someya et al 2020. This area is highly populated and contains large areas of intensive agricultural activities including livestock and crop production, most notably, wheat. In contrast, over the eastern IGP, the maximum seasonal columns are in March and April, a local seasonality distinct from the regional average. Figure 3 demonstrates the large difference between monthly ammonia emissions using the Coupled Model Intercomparison Project Phase 6 (CMIP6) with CrIS and IASI total column measurements. The emissions use seasonality data from the Community Emissions Data System (CEDS) developed for CMIP6 (Hoesly et al 2018). The CMIP6 seasonality exhibits a maximum in May and a smaller peak in September. In contrast, satellite observations show that NH 3 column peaks in July, consistent with emissions calculated based on seasonality from observations from AIRS and TES (Warner et al 2017, Paulot et al 2018). The larger range of IASI (higher columns in summer and lower in winter, relative to CrIS) result in lower winter contributions to total annual column averages relative to CrIS (and conversely, higher summer contributions). Satellite data from figure 3 show a more accurate seasonality for the greater India region than by using an assumed profile from other regions such as Europe. In CEDS, agricultural NH 3 emissions, the dominant source of NH 3 in India, are distributed across months according to the ECLIPSE v5 model following European practices (Friedrich andReis 2004, Giglio et al 2006). The peaks in May and September correspond to fertilizer application before planting and after harvest. Our results show that this approach does not capture Indian emissions well. The large differences in column strengths and emission sources over different parts of this region, combined with satellite biases, make such a regional averaging approach inaccurate, although improved over using European-based parameterizations. To this end, a k-means clustering approach, as applied below, reduces the effect of instrument bias and can discern sub-regional seasonality in more detail.

Sub-regional seasonality
The high spatial resolution of NH 3 obtained from oversampling extensive satellite datasets have been analyzed to ascertain sub-regional seasonality of NH 3 over India, Pakistan and Bangladesh. Figures 1 and  2 show how the spatial distribution changes on a monthly basis, but the results of the k-means clustering analysis in figure 4 more clearly show the different seasonal distributions of NH 3 in South Asia. The maps have been sub-divided into regions which contain pixels with similar seasonality to each other.  Figures (b) and (d) show the proportion of ammonia observed per month of total annual NH3, in each region, for both satellite instruments. The regions with the highest NH3 columns, the Western IGP has a summer peak, and the Ganges Delta region shows a springtime peak.
The clustering analysis results in five distinct regions for both IASI and CrIS. The spatial patterns of the clusters are strikingly similar between the two datasets from these independent measurements, demonstrating the robustness with our analysis. The regions with the most distinct seasonality are northwest India and eastern Pakistan (shown in orange and green), and southern and eastern India and Bangladesh (shown in pale blue and red).
Other regions in central and southern India show seasonal variation in observed NH 3 column density, although the total NH 3 signal from these areas are relatively low. Seasonal variations in NH 3 emissions may be present in these areas, however column measurements from satellite are closer to the detection limit. In areas of NH 3 columns around or below the sensitivity threshold, temporal variations are expected to be difficult to quantify. Figures 4(b)-(d) show the seasonality profiles of each distinct region from IASI and CrIS, respectively. In figure 4(b), the purple and green traces, representing mountainous areas from the Himalaya and Sulaiman Range show maximum seasonal signal in the mid-summer. These areas however represent a small fraction of total NH 3 in the study region as shown in figure 5, with between 3% and 6% of total NH 3 in July coming from these areas. Figure 5 shows significant similarity between the two instruments, with the pale blue sub-region of eastern India dominating NH 3 columns in December-February and the western IGP (orange) dominating in the mid-summer.

Discussion
The primary source of NH 3 in the IGP is from agricultural practices, through volatilization of waste from livestock and nitrogen-based fertilizers for crops. The IGP has two main growing seasons, Rabi (October-March) and Kharif (July-November). Figure 6 shows the seasonalities of NH 3 column in the western and eastern part of the IGP, along with shaded regions showing the approximate planting dates for the Kharif (June-July) and Rabi (November-February) crops. Most nitrogen-based  fertilizer is applied approximately one month after planting, which agrees very well for the Kharif crop in the western IGP, but no NH 3 peak is seen for the Rabi season planting dates. High NH 3 columns in the western IGP during winter have been observed using IASI nighttime (2130) overpasses (Van Damme et al 2014), however those data are associated with high uncertainties given the lack of thermal contrast at night. In situ measurements in Delhi have also shown high levels of NH 3 in winter (Sharma et al 2010). Fertilizer application is different for rice, with most fertilizers being applied at the time of planting; however, nitrogen-based fertilizer is generally applied 30-90 days after sowing (Craswell et al 1981, Singh et al 2006, Varinderpal-Singh et al 2007. Fertilizer application for rice planted in winter could then explain the NH 3 peak in April for the eastern IGP, although we do not observe a corresponding NH 3 peak for the Kharif crop. The eastern part of the IGP has significantly more rainfall than the western part during the monsoon season (Prakash et al 2018), and NH 3 is known to be scavenged by precipitation events (Black et al 1987, Nowak et al 2006. This could potentially explain the reduced NH 3 columns over eastern India and Bangladesh during the monsoon season, while the western IGP maintains relatively higher concentrations. Biomass burning is also a major source of NH 3 emissions (Bouwman et al 1997). Satellite observations of fire counts from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument were used to monitor biomass burning in the study region. The majority of fires result from cropresidue and stubble burning in the spring and autumn before replanting.  of fires in Eastern India in this month, which could explain the annual maximum NH 3 signal seen in this region in figures 4(a)-(c). There are a small number of fires in Northeastern India and very few in Bangladesh. Although these could contribute to the observed NH 3 columns, it is likely that fertilizer application for rice production is the main cause of the spring seasonal maximum found in this region.
However, maps of fires in India and Pakistan, for example figure 7(b), show no significant biomass burning around the NH 3 peak in the western IGP in July. It is unlikely that biomass burning, while a significant source of atmospheric NH 3 in this region (Kuttippurath et al 2020), is a major factor in the NH 3 seasonal peak in July observed in this study.
The satellite NH 3 seasonal trends show mixed agreement with limited in-situ measurements. Kumar et al (2004) observed higher NH 3 in summer than winter at Agra in the IGP, in agreement with the satellite seasonal trends. In Ahmedabad in western India, Sudheer and Rengarajan (2015) showed late June/early July measurements were 10% lower than those in December/early January for a one-year period. IASI/CrIS indeed show a relatively flatter seasonal trend in this region compared to the IGP, yet the satellites have a summer maxima. Sharma et al (2010) reported higher NH 3 in winter than summer and the monsoon months at Dehli, in opposite of satellite trends. More Saraswati et al (2018), (2019) measured NH 3 for five years (2011)(2012)(2013)(2014)(2015) in Dehli with a monsoonal minima of 15 ppbv (July-September) and wintertime (November-February) maximum of 25 ppbv. The intrinsic differences between a point-based, surface measurement and column-based satellite measurement may be a key reason for the discrepancy. For example, for the same NH 3 column, a lower boundary layer in winter will yield higher surface concentrations than a higher boundary layer of summer, assuming most NH 3 is in the boundary layer (a reasonable assumption for a surface source and short lifetime species). Indeed, one advantage of column-based NH 3 measurements is the insensitivity to boundary layer effects which can be difficult to model, especially in winter. India trace gas and PM 2.5 concentrations are strongly impacted by the setup of a strong, shallow boundary layer in winter (Brooks et al 2019, Ojha et al 2020. Measurements of boundary layer height and vertical profiles of NH 3 are ultimately needed to compare satellite with surface observations. While more in-situ and long-term datasets of NH 3 are also desired, satellites provide observational NH 3 constraints for improving air quality studies and emission inventories at monthly and subregional scales.

Implications
The spatial and temporal distribution of NH 3 total columns have been measured over India, Pakistan and Bangladesh using observations from two separate satellite instruments, the IASI and the CrIS. Using a k-means clustering technique, the study region has been divided into sub-regions showing distinct seasonality profiles of NH 3 . The seasonalities measured by the two separate satellite instruments show the same spatial patterns and temporal evolution throughout the year. These biases are expected to propagate to the simulation of particular matter with implications for air quality and radiative forcing. For instance, Paulot et al (2018) showed that distributing CEDS NH 3 emissions according to the observed seasonality of NH 3 columns from AIRS (Warner et al 2017) reduced the wintertime increase in clear-sky aerosol radiative effect by 30% over the 2002-2015 period. The observations show two main areas of high NH 3 measurements, the western Indo-Gangetic Plain that has maximum NH 3 measurements in the mid-summer and the eastern Indo-Gangetic Plain which has maximum NH 3 in the spring. Across the IGP, there are two main periods of nitrogen fertilizer application, June-July and November-December (Nishina et al 2017), corresponding to the Kharif (summer) and Rabi (winter) crops. The observed summer peak of NH 3 correlates spatially with fertilizer application for the Kharif wheat crop, although there is no corresponding peak seen for the Rabi crop. Ammonia emissions from fertilizer are highly dependent on temperature (Misselbrook et al 2004, Xu et al 2019 and the IGP has large seasonal temperature changes Norouzi 2020, Narasimha Murthy et al 2021), with temperature maxima in late spring and early summer and minima in the winter. The combination of high fertilizer application and temperatures in summer may account for the observed ammonia peak. The spring maximum over eastern India can be explained by the delayed nitrogen-based fertilizer practices (up to 70 days after sowing) for rice production (Latif et al 2005, Thind et al 2017) and significant fires (figure 7) during this period in eastern-central India. The similarity in spatiotemporal variabilities given by these two independent data sources gives confidence to the patterns described here.
Due to the short lifetime of ammonia, maps of concentrations from satellite should reflect the underlying emissions. The maps shown in figures 1 and 2 are inconsistent with current emissions inventories. This is reflected both on a national scale (figure 3) and significant sub-regional variations (figures 1, 2 and 4). To accurately model ammonia seasonality in India, regionally specific emissions are required. Current emissions models for India use European seasonality data for NH 3 , a region with different meteorology, relative land use and agricultural practices. This work further shows the large spatial gradients that change monthly in this region. NH 3 measured from satellite provides an essential check on emissions inventories over this large sub-continent which has diverse emissions sources and sparse surface measurements. In order to disambiguate the multiple emissions sources of NH 3 , bottom-up measurements are required, particularly during winter in the western IGP.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.