Ammonia measurements from space with the Cross-track Infrared Sounder (CrIS): characteristics and applications

Ammonia measurements from space with the Cross-track Infrared Sounder (CrIS): characteristics and applications Mark W. Shephard, Enrico Dammers, Karen E. Cady-Pereira, Shailesh K. Kharol, Jesse Thompson, Yonatan GainariuMatz, Junhua Zhang, Chris A. McLinden, Andrew Kovachik, Michael Moran, Shabtai Bittman, Christopher Sioris, Debora Griffin, Matthew J. Alvarado, Chantelle Lonsdale, Verica Savic-Jovcic, and Qiong Zheng. 5

near the surface along with global coverage. Presented here is a current survey of the CrIS NH3 Fast Physical Retrieval (CFPR) product characteristics with some example applications.

CrIS satellite retrievals
The CrIS instrument is a Fourier Transform Spectrometer Here we focus only on the longer time series of data provided by the CrIS instrument flown on S-NPP. The S-NPP satellite is in a sun-synchronous low earth orbit with overpass times of ~1:30 and 13:30 mean local time. CrIS is a hyperspectral infrared instrument with a spectral resolution of 0.625 cm -1 . The main advantage of CrIS is the combination of dense global coverage and the improved sensitivity in the boundary layer due to the low spectral noise of ~0.04 K at 280K in the NH3 spectral region 10 (Zavyalov et al., 2013) and the early afternoon overpass which coincides with high thermal contrast (difference between the surface and air temperature) when the infrared instrument is more sensitive. A detailed description of the CrIS Fast Physical Retrieval (CFPR) algorithm for deriving ammonia applied to both simulated spectra and initial sample observations was provided by Shephard and Cady-Pereira (2015). Since then the CFPR algorithm has been applied globally to CrIS spectra from May 2012 onwards. The input atmospheric state required for the radiative transfer forward model calculations are 15 obtained from the Level 2 Cross-tTrack Infrared and Microwave Sounding Suite (CrIMSS) Atmospheric Vertical Profile Environmental Data Record (product ID: REDRO) (Divakarla et al., 2014) product from May 1, 2012 to April 7, 2014, after that the retrieved Level 2 NESDIS-unique CrIS-ATMS product system (NUCAPS) (Liu et al., 2014) are used. The CPFPR retrieves the surface temperature and emissivity for each observation (field-of-view) prior to the ammonia retrieval. Ammonia profiles are retrieved at 14 profile levels to capture the vertical sensitivity of ammonia that varying from profile-to-profile 20 depending on the atmospheric conditions. The CrIS satellite ammonia observations do not have equal sensitivity in the vertical, and have coarse vertical resolution (e.g. ~1 to 3-km). Hence, surface level values and total column values are both highly correlated with the boundary layer values where the satellite typically has peak vertical sensitivity. Note that atmospheric ammonia is typically short-lived so that higher concentrations are generally close to the sources, which are generally near the surface. This is demonstrated later in Section 3.2 with model emissions and corresponding simulated surface concentrations. 25 An update from the initial Shephard and Cady-Pereira (2015) analysis is that under favourable conditions CrIS detects NH3 near surface concentrations down to ~0.3-0.5 ppbv (e.g. Kharol et al., 2018), which is less than half of the more conservative estimate of ~1 ppbv previously reported using an Observation System Simulation Experiment (OSSE). This is mainly due to the better than specified noise capabilities in the observed CrIS NH3 spectra, and the limited number of sampling conditions used in the original OSSE experiment. 30 Since the CFPR uses a mathematically robust physics-based optimal estimation framework (Rodgers, 2000) it provides the vertical sensitivity and the measurement information content (obtained from the averaging kernels), and an estimate of the retrieval errors (error covariance matrices), for each observation. The output sensitivity and error parameter characterization are key for utilizing CrIS observations in air quality model applications such as data assimilation, data fusion, and model based emission inversions (e.g. Li et al., 2019). It is also important that, as done first in the TES NH3 retrieval (Shephard et al., 2011), the CFPR algorithm uses only three a priori ammonia profiles. These a priori profiles represent unpolluted, moderate, and polluted conditions with no prescribed latitudinal or seasonal dependence. For each retrieval one of these three a priori 5 profiles is selected based on the estimated ammonia spectral signal, as there is little known about ammonia globally (i.e. there is no spatial climatology field used for the a priori as is commonly done for retrievals of better known species). The retrieval quality flags are described in Appendix A. 10 Figure 1. CrIS retrieved NH3 profiles from a 50km radius around Cabauw, Netherlands from April-September 2016. The retrieved profile level values below 600 hPa are shown in the left panel and coloured according to the surface value, and the corresponding rows of the averaging kernels are shown in the right panel. The box-and-whiskers showing the statistics (e.g. median, percentiles, and outliers (circles)) of the rows of the averaging kernel values at each retrieval level are also provided on the averaging kernel plot. 15 As shown in Figure 1Figure 1 the peak sensitivity is generally in the boundary layer below ~700 hPa (3-km). The lower instrument noise with similar spectral resolution offered by CrIS allows for greater sensitivity near the surface with less dependency on the thermal contrast from an operational meteorological sensor. This also follows from simulation studies performed by Clarisse et al. (2010) that show with even twice the CrIS noise level there would be a significant reduction in 20 the dependence on thermal contrast for sensitivity in the daytime boundary layer. As there is generally only ~1 degree-offreedom for signal (DOFS) (e.g. 0.95 average in Figure 1Figure 1) with coarse vertical resolution (half-width-at-half-maximum of the rows of the averaging kernels) of ~1 to 3-km, the retrieved surface level concentrations are highly correlated with the Formatted: Font: Not Bold Formatted: Font: Not Bold retrieved levels at higher elevations in the boundary layer. The retrieved profiles in Figure 1Figure 1 still tend to have distributions that are grouped around the three a priori profiles and so, future updates to the retrieval will investigate various refinements to the a priori profiles and constraints. Figure 2Figure 2 shows a sample single day scene of NH3 retrievals on May 15, 2016 during the Fort McMurray fires (Adams et al., 2019) ranging from low background values of < 1 ppbv to elevated values up to 30 ppbv. Corresponding Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) 5 (Winker et al., 2003) lidar measurements on this day shows the smoke plume reaching altitudes above the ground of ~ 2.0 to 3.0 km (~800 to 700 hPa) (for reference see Figure A  The estimated random errors for both the observation (consisting of only measurement errors here as no cross-state 15 errors are estimated) and the total error (includes the measurement and representative (or smoothing) error) are computed for both the individual retrieval profile levels and the integrated total column (see Shephard and Cady-Pereira, 2015). Observation errors can be used if the vertical resolution of the satellite observations are already taken into consideration (e.g. satellite observation operator is applied to the comparison dataset), whereas the total error should be used if the satellite retrieved value is to represent the discrete observation resolution (e.g. individual profile level value, or vertical column value were the vertical 20 sensitivity is not considered). Errors from a single day of global retrievals as a function of concentration amounts are provided in Figure 3Figure 3 for total column, and in Figure 4Figure 4 for values from each profile level in the boundary layer below 700 hPa (~3-km). For the total column amounts, the measurement errors are typically in the 10 to 15% range, whereas the total errors are ~30%. The individual profile retrieval levels have measurement errors of ~10%, except for low concentrations 6 with amounts < 1 ppbv where the error rises to ~30%. When the smoothing error component is included for the individual profile levels, the profile level total random error increase to the range of 60% to 100%. This is expected given that current ammonia nadir infrared retrievals have limited vertical information (resolution), leading to significant smoothing of the retrieved profile. 5 Figure 3. Estimated CrIS retrieved NH3 total column measurement and total errors statistics from global values for May 28, 2017. Both absolute error estimates (top panels) and corresponding relative (fractional) errors (bottom panels) are shown. The diamonds are the mean values for each box range. Only retreivals over land with a quality flag of 5 (therefore DOFS ≥ 0.1) where included. The N and M are the number of points and the median value for each box, respectively.
10 Figure 4. Errors for the individual for all profile levels below 700 hPa (~3-km) using the same plotting criteria as in Figure  3Figure 3.
The CFPR (Version 1.3) was also validated by Dammers et al., (2017) against ground-based FTIR observations to 5 determine the actual errors (as opposed to the estimated retrieval errors). These initial results show a good overall comparison results with a correlation of r~0.8 with a slope of 1.02. For retrievals with total column values > 1.0 x 10 16 molecules cm -2 (ranging from moderate to high levels) the relative bias difference is < 5%, with a standard deviation ranging from 25 to 50%.
For total column comparisons for smaller values < 1.0 x 10 16 molecules cm -2 there are larger differences with a CrIS higher than the FTIR by ~30% with a standard deviation of ~40%. Initial CrIS comparisons of the surface level retrievals with in-10 situ surface observations from the Ammonia Monitoring Network (AMoN) over North America show a correlation of 0.76 and an overall mean CrIS -AMoN difference of ~+15% . An extension of this analysis over more surface locations and longer time-periods are currently being performed. These CrIS retrieved ammonia profiles (Level 2 products) are used to generate gridded averaged (Level 3) products over various spatial grids and time-periods (e.g. monthly average on 0.1 x 0.1 degree latitude and longitude grid). To reduce discontinuities between adjacent grid points and increase the effective 15 resolution of observations that are averaged over extended time-periods (e.g. monthly), oversampling with weighting (e.g.
Gaussian weighting based on distance from the centre of the grid) (e.g. Fioletov et al., 2011;Pommier et al., 2013) is used in the generation of the Level 3 gridded averaged products. The output Level 3 product grid size and level of oversampling are flexible and based on the purpose and number of observations (e.g monthly, annual, multi-year).   The distribution of the CrIS degrees-of-freedom for signal (DOFS) for a 5-year mean (2013-2017) corresponding to Figure 5 is provided in Figure 6. As noted previously, for satellite nadir NH3 observations there is often limited independent pieces of information with DOFS values of ~1 or less. This information distribution depends on the atmospheric state (i.e. abundance of NH3, temperature, vertical thermal contrast in temperature, etc.). Thus, warmer climates with more persistent NH3 sources will tend to have more DOFS. Higher latitudes where there is both little agricultural sources of NH3 and cooler 5 temperatures there is often less DOFS on average; the exception is often under conditions where there are large episodic biomass burning events. What is averaged out of a 5-year mean plot is that single pixel retrieval can reach maximum DOFS of ~2 is large forest fire plumes. 10 Figure 6. CrIS 5-year (2013CrIS 5-year ( -2017 global mean of DOFS corresponding to the retrievals in Figure 5.

Application examples 15
The CrIS NH3 product can be used for many applications such as monitoring, air quality model evaluation (e.g. Whaley et al., 2018;Pleim et al., 2019), dry deposition estimates , and emissions estimates for larger agriculture sources, industrial point sources (Dammers et al., 2018), and wildfires (Adams et al., 2019). Here we provide examples that demonstrate and expand upon these applications.

Monitoring
As previously noted, the operational polar-orbiting satellites (e.g. IASI and CrIS) have the benefit of providing daily global spatial coverage on local to regional (e.g. 10's of km) scales over many decades that can help fill in gaps in current monitoring networks. Provided here are examples of daily, seasonal, and annual observations of ammonia by the CrIS satellite. While not currently done so, it is possible to derive global daily ammonia products in near real time. 5

Daily
Ammonia in general is relatively short-lived in the boundary layer so its day-to-day atmospheric concentration levels over a region can vary greatly depending mainly on the meteorology (e.g. windspeed, temperature) and episodic events (e.g. biomass burning, spreading of fertilizer) and concentration of reactant acid gases. However, ammonia in the free-troposphere above the boundary layer, for example during forest fires, is not quickly scavenged or deposited and hence has a longer lifetime and 10 travels over large distances . In addition, ammonia deposited on certain surfaces (e.g. vegetation) can be re-released into the atmosphere later (bi-directional flow) depending on the ammonia balance between the air and leaf apoplastic concentrations (compensation point) (do that there is bi-directional flow) (e.g. Massad et al., 2010;Bash et al., 2013;Pleim et al., 2019). In Version 1.5 of the CFPR NH3 retrievals the observations with insufficient ammonia signal in the measured spectrum, mainly due to ammonia concentrations below the sensor detection limit (< 0.3-1.0 ppbv) or clouds blocking the ammonia signal, are not currently being processed. Thus, cloud filtering is presently achieved implicitly through the threshold of the ammonia signal in the spectra (i.e. no ammonia spectral signal through clouds) and the use of a surface brightness temperature threshold derived from global seasonal climatological cloud top temperatures (based on International Satellite 25 Cloud Climatology Project (ISCCP) maps; https://isccp.giss.nasa.gov/products/browsed2.html). This upfront cloud prescreening also improves the data processing rate as it reduces the number of potential number of retrievals globally by an average of 35%, with this rate varying depending on location and season. Comparing the MODIS true-colour imagery with the CrIS NH3 observations in Figure 7Figure 6 demonstrates that this technique is very effective for cloud screening. Note that thin clouds (cloud optical depth < 1.0) that are near the surface with cloud-top temperatures close to the surface temperature 30 still impacts the current ammonia retrievals, but in general has a non-significant impact on the overall results as seen in the examples in Figure 7Figure 6. Algorithm refinements such as directly incorporating a newly developed coincident VIIRS

Seasonality
Ammonia concentrations in the atmosphere are influenced by agricultural practices and meteorological conditions.
Ammonia emissions differ over the course of the growing season due to changing farming practices and ambient temperature, leading to a month-to-month variation in concentrations. As an example, the relative spatial seasonal variability in surface concentrations over North America is shown in Figure 8Figure 7. For most of North America, there is often an increase in 5 concentrations during the springtime associated with fertilizer and manure applications, and warming surfaces, at the start of the growing season, which shifts from April in the southern to central parts of the U.S. to May in the northern states and most of Canada. There can also be an increase over some source regions (i.e. U.S. Midwest, Idaho, Washington state) in the summertime associated with increased temperatures and certain farm practices like cleaning corals and manure storages and spreading manure on harvested winter crops or forages in mid to late-summer promoting more volatilization. Also apparent 10 in the plots is the increase in concentrations in the non-agricultural northern latitude regions during the drier summer season associated with wildfires, which can inject ammonia with minimal acid reactants into the free troposphere allowing the transport of ammonia over larger regions (Lutsch et al., 2016;Lutsch et al., 2019). The retrieved elevated concentration values at high-elevation over the Rocky Mountains in the wintertime needs to be further investigated as a potential retrieval issue.
The corresponding plot of total column values provided in Figure A 3 show similar spatial seasonal patterns seen the retrieved 15 surface values. This is generally expected as ammonia is typically short-lived in the boundary layer so higher agricultural hotspots are close to source locations, plus both surface level retrievals and the corresponding integrated total column values are correlated with the profile retrievals in the boundary layer where the satellite typically has maximum sensitivity (as shown in Figure 1). 8 are shown in Figure 10Figure 9. The most salient features in the regional time series reflects the seasonal cycle seen for most regions (e.g. south eastern China, South America), where the ammonia concentrations peak in the warm growing season and are a minimum during the colder season. Some regions also show a double peak in concentration amounts during the 5 growing season (e.g. central U.S.) that can be associated with the large spring time fertilizer or manure application with a second peak in a couple of months later due to increasing temperatures, which can be associated with increased agricultural ammonia volatilization and/or biomass burning. Some regional time series show an increase in peak ammonia concentration amounts with time (e.g. South America), while others show more constant seasonal pattern over the years (e.g. central U.S.).
In addition to agricultural practices, there can also be contributions to the atmospheric ammonia amounts due to biomass 10 burning for some of the regions. When wintertime temperatures are near or below freezing there is a decrease in satellite sensitivity to ammonia, which reduces observation density and can create a small high-bias (e.g. southern Canada in 2016 and 2017).

Model evaluation
Chemical transport models (CTMs) are used for many ammonia related air quality applications such as estimating acid deposition and secondary particulate matter formation, and scenario runs to inform policy development (e.g. Engardt et. al., 5 2017;Liu et. al., 2019;Makar et al., 2009;Pinder et. al., 2007). Evaluation of the model performance against observations is a key part of air quality modelling validation that ultimately leads to an improved model (e.g. Shephard et al., 2011, Van Damme et al., 2014. An example of using CrIS NH3 observations for CTM evaluation is provided by Whaley et al., (2018).
In that study a newly implemented ammonia bi-direction flux scheme and inclusion of biomass burning into Environment and Climate Change Canada's air quality forecast model, the Global Environmental Multi-scale -Modelling Air quality and 10 CHemistry (GEM-MACH) model (Gong et al., 2015;Makar et al., 2015a,b;Pendlebury et al., 2018), were evaluated over northern Canada using CrIS NH3 observations. CrIS NH3 observations have also recently been used to evaluate improvements to the Community Multiscale Air Quality Model (CMAQ) (Pleim et al., 2019). They demonstrated that CMAQ underestimated NH3 concentrations in the spring, but also that CMAQ and CrIS present the same pattern of high NH3 in the California Central Valley, the Snake River Valley and western High Plains, all regions with high soil pH resulting in high NH3 fluxes, suggesting 15 that CMAQ modeling of soil pH and the fluxes dependent on this parameter are reasonably well modeled.  Figure 12Figure 11a shows the average monthly total NH3 emissions during a two-month summer period (July and August) over the GEM-MACH 25 North American grid. The corresponding average NH3 surface concentrations fields predicted by GEM-MACH during July and August, 2016 is shown in Figure 12Figure 11b. The spatial distribution of model-predicted anthropogenic NH3 surface concentration aggrees well with the spatial distribution of bottom-up model emissions in Figure 12Figure 11a. This is expected since NH3 is a short-lived reactive species and high concentrations of NH3 occur mainly in the areas with high NH3 emissions.
In general, the locations of the elevated ammonia "hot-spot" regions in simulated model surface concentration map in 30 Figure 12Figure 11b match wellspatially co-located with those observed by CrIS (Figure 12Figure 11c). This is seen in the  Figure 12 would indicate that these differences are mainly due to the input emission fields. whereas in the western part of the U.S. the satellite observations tend to be slightly higher over regions with elevated NH3, most notably over the Central Valley and High Plains. Similar results (not shown here) are also seen with other chemical transport models (e.g. 5 GEOS-Chem) using the same U.S. EPA emissions inventory. whereas in the In addition, the western part of the U.S. the satellite observations tend to be slightly higher over regions with elevated NH3, most notably over the Central Valley and High Plains. There are also several elevated regions in the satellite observations in Mexico that appear to be underreported in the emission inventories. One other potential difference between the model simulations and the satellite observations is the contribution of forest fires to the NH3 concentration amounts, especially at higher latitudes where there are limited agricultural 10 sources. The reason for this potential difference is that the fire emissions were not considered in this GEM-MACH simulation.

Dry deposition of reactive nitrogen
Deposition of basic ammonia and ammonium-containing aerosols on land surfaces leads to acidification of the soil, when ammonium is oxidized (nitrified) to nitrate (NH4 + + 2O2 → NO3 -+ 2H + + H2O) (Goulding, 2016). The protons generated from this reaction cause the acidification. Excessive The atmospheric deposition of NH3 addscontributes excessive reactive nitrogen 10 into water that can contributes to eutrophication. Kharol et al., (2018) first demonstrated the utility of using CrIS NH3 observations with modelled dry deposition velocities to compute estimates of dry deposition of reactive nitrogen (Nr) from ammonia for the 2013 warm (growing, April to September) season over North America. van der Graaf et al., (2018) followed a similar approach over Europe using the LOTOS-EUROS model to IASI total column NH3 observations into surface estimates. Here wWe expanded upon the CrIS dry deposition their seasonal analysis to compute annual estimates of the relative Nr dry deposition flux from NH3 and nitrogen dioxide (NO2) for the years 2013-2017. As described in Kharol et al., 5 (2018), the Ozone Monitoring Instrument (OMI) was used for the NO2 deposition estimates. The spatial patterns of the annual 2013 NH3 dry deposition flux shown in Figure 13Figure 12 closely resemble the warm season results shown by Kharol et al., (2018) as ammonia has a short lifetime in the atmosphere so the majority of the ammonia deposition occurs on leafy vegetation during the growing season close to agricultural sources. In contrast, the dry deposition of NO2 is generally associated with emissions from urbanized areas and industrial sources year round, changing only slightly during the warm season. This results 10 in total dry deposition in Canada (excluding Territories) and the U.S. in 2013 from NH3 (NO2) of ~0.75 (0.17) Tg N year -1 and ~1.09 (0.91) Tg N year -1 , respectively, of which ~0.5 (0.1) Tg N warm season -1 and ~0.9 (0.4) Tg N warm season -1 , were deposited in the growing seasons as in Kharol et al., (2018). The 2013 annual ratio maps show NH3 having larger proportion of the (NH3 + NO2) (~82% and ~55 % over Canada and the U.S.). As shown in Figure 13Figure 12, 31 out of the 50 U.S.
states are (mostly located in central and western U.S.) show greater dry deposition rate from NH3 compared to NO2. In contrast, 15 the industrial northeastern states indicate higher dry deposition rate from NO2 than NH3. NH3 is expected to continue to be the dominate source of the reactive nitrogen dry deposition flux over most regions in North America as NH3 emissions are projected to increase in the future (e.g. Bauer et al., 2016;Ellis et al., 2013;Paulot et al., 2013), and because of declining trends in NO2 emissions (e.g. Kharol et al., 2015;Krotkov et al., 2016;Lamsal et al., 2015).  The five-year annual mean NH3 dry deposition flux for the period of 2013-2017 over North America are shown in Figure 14Figure 13(a). The year-to-year variability in the NH3 dry deposition flux over North America are shown in Figure   14Figure 13(b-f). The hotspots in the northern latitudes during 2014 and 2015 are mainly associated with large forest fires that may lead a 2-3 fold local increase relative to background value . The NH3 dry deposition flux hotspots evident in the agricultural states of central U.S., and the Canadian provinces of Alberta, British Columbia and Saskatchewan 5 during 2017 (Figure 14Figure 13f) are mainly due to the combined effect of forest fires and a warmer than average summers as shown in the CrIS NH3 concentrations maps in Figure 11Figure 10. The annual average and variability in ammonia dry deposition of reactive nitrogen over Canada and the U.S. is ~0.8 ± 0.08 Tg N year -1 and ~1.23 ± 0.09 Tg N year -1 , respectively.
Note that there will be a significant contribution and variability from large forest fires in northern latitudes as seen in Figure   14Figure 13. 10

Emissions estimate for a concentrated agricultural region 5
Another application of CrIS ammonia observations are emission estimates. Emission inventories are traditionally built from the bottom up, using emission factors and source locations to construct a complete inventory (e.g. see Appendix D). This process is very labor intensive, which means that the inventories are often released somewhat infrequently, with gaps of a few years between releases. Furthermore, the inventories can be incomplete or inaccurate due to a lack of knowledge on source locations, magnitudes, and temporal variations. This is particularly true for farm based emissions that require complex data about farm activities, which can only be obtained with complex farm surveys typically conducted sporadically. Top-down satellite observations can be used to provide another source of emissions information and to supplement the inventories with more detailed information on hotspot locations and temporal variations (e.g. seasonal and inter-annual variations) (e.g. Van  However, emissions in this region are expected to vary, especially on a monthly/seasonal basis. The 2013 Canadian Ammonia 5 Emissions from Agriculture Indicator (AEAI) monthly emission inventory (Sheppard and Bittman, 2016) states 30% of all emissions are assume to take place in the cold season (October-March) and 70% in the warm season (April-September).
Therefore, the annual emission total has to be adjusted by a factor of 1/0.7 = 1.43, which makes the adjusted emission total of 53.4±9.0 kt yr -1 . Similarly, the diurnal emission profile can be approximated by using the diurnal emission profile for livestock that is used to prepare emissions for the GEM-MACH model. Most of the emissions around Lethbridge are from livestock, 10 which has a peak in the morning to middle of the day following cattle activity including feeding and excretion and the increasing surface temperatures (Denmead et al., 2014;Van Haarlem et al., 2008), and are thus emitted before the satellite overpass time (~1:30 local time). Thus, the inventory value from an hour before the overpass (12:30 local time) is used to adjust the emissions to daily averages by a factor of 1/1.44 leading to a final annual emission total of 37.1 ± 6.3 kt yr -1 . Figure   16Figure  20 Figure 1615. The bars show the Lethbridge regional emissions as in the AEAI (blue), HTAPv2 (red), and APEI (green) inventories summed over a box with a bottom-left corner of (49.35º, -113.31º) and a top-right corner of (50.28º, -111.10º). The orange bar shows the CrIS estimated emissions using the free fitting algorithm and adjusted for diurnal and seasonal variability. The vertical error bar shows the uncertainty of the CrIS estimate that includes the uncertainty in the fit, the total columns, and 25 uncertainties due to the meteorology.
In addition to the annual total, emissions are estimated using 5-year monthly observations and a moving window of 31 days in 1 day increments. This type of analysis can be used to help better constrain the seasonal and daily timing of emissions over this region. Figure 17Figure 16 shows the results of the emission estimates, compared to the AEAI 2013 monthly emission inventory (blue). The orange lines and red dots show the results when applying the plume fitting algorithm to the CrIS monthly dataset and daily sets (31-day moving window), respectively. While some months might have enough 5 information available in the mean total column fields, this is not true for each of the months as indicated by the relatively noisy results outside of the warm season. To improve the stability of the fitted results a lifetime of 2.65 hours and a plume spread of 19 km as obtained from the 5-year analysis above, were used here to estimate the CrIS derived monthly and daily emissions in Figure 17Figure 16. This lifetime and plume spread distance were derived based on 5 years of CrIS data. This type of analysis can be used to help better capture the warm season timing of emissions over this region. The general overall seasonal 10 changes in the spring and summer are similar between the CrIS derived emissions and the AEAI inventory with peak emissions over Lethbridge in May springtime, but CrIS is showing nearly double the amount compared with AEAI. In this example, CrIS is not showing the smaller secondary fall peak (September-October) that is seen in the AEAI inventory.

Conclusions
Satellite observations of lower tropospheric ammonia are a relatively new development with the initial proof of concept 10 in the past decade (Beer et al., 2008), thus, there is a great potential for advancements in the retrievals and exploration of new applications. Presented here is an overview of CrIS' ammonia data product highlighting its current capabilities to observe lower tropospheric ammonia with sample applications for monitoring, model evaluation, dry deposition, and emission estimates. The CrIS daily observations demonstrate the influence of meteorology on the spatiotemporal variability of ammonia. These examples show the transport of ammonia concentrations from nearby agriculture sources as well as from fire 15 emissions. Averaging these daily observations over longer time-periods (e.g. monthly, seasonal, and annual) and gridding and oversampling (to yield Level 3 products) illustrates the spatiotemporal variability of ammonia at various timescales. These results demonstrate CrIS' ability to observe regional changes in ammonia concentrations due to agricultural practices, such as spring maximum values over agricultural regions when ammonia is released into the air from the fertilizing of crops. Also shown is the importance of episodic wildfire emissions in the more wildfire active months, especially in regions where there 20 is little or no agriculture sources such as the northern latitudes in North America during July and August.
Initial comparisons of CrIS NH3 satellite observations with GEM-MACH air quality model simulation in summer 2016 show that in some regions there is general agreement on the spatial distribution of the anthropogenic hotspots, while other areas are markedly different and will need further investigation. For this summer period, the model tends to have higher peak values in the eastern U.S., whereas the satellite tends to have larger peak values in the western half of the U.S. As the CTM 25 runs only have anthropogenic emission sources included, we can see that the impact of large summertime wildfires at higher latitudes on the 2-monthly mean concentration levels over large regions can be significant, and can approach the values of agriculture hotspots at lower latitudes.
Expanding on the initial 2013 growing season results from Kharol et al, (2018) we show annual dry deposition rates of nitrogen from ammonia for 5-years from 2013 to 2017 over North America. CrIS satellite derived values show the annual 30 average and variability in the dry deposition of reactive nitrogen from ammonia over Canada and the U.S. of ~0.8 ± 0.08 Tg N year -1 and ~1.23 ± 0.09 Tg N year -1 , respectively. When combining with OMI-derived NO2, the 2013 annual ratio shows NH3 accounting for ~82% and ~55 % of the combined reactive nitrogen dry deposition from these two species over Canada and the U.S. CrIS satellite observations are also used to derived agricultural emissions over the CAFO dominated region of Additional application such as using the CrIS CFPR products for model inversions and data assimilation (Lonsdale et al., 2019;Li et al., 2019) are currently being explored, which take advantage of the averaging kernels and error covariance matrix provided in the CrIS retrieved product (e.g. observation operator), to provide top-down constraints on the ammonia emissions. Additionally, we will continue to refine and validate the CFPR algorithm and product. Some of these potential efforts include: (i) accounting for cloud-free pixels that have no information (no ammonia signal in the spectra) in the CrIS 10 composite (Level 3) products globally for the entire dataset; (ii) investigating retrievals over ocean and elevated concentration values over some deserts and high elevations wintertime conditions (e.g. North American Rockies); (iii) investigating the potential enhancements to the a priori profiles and constraints used in the retrievals; (iv) and validating CrIS NH3 night time observations against available ground-based observations. 15

Acknowledgements
We would like to acknowledge the NOAA Comprehensive Large Array-Data Stewardship System (CLASS) (Liu et al., 2014), with special thanks to Axel Graumann (NOAA), for providing the CrIS Level 1 and Level 2 CrIS REDRO and NUCAPS input atmospheric state data. We thank Denis Tremblay (Science Data Processing, Inc.) for providing value insights on the performance and characteristics of the CrIS instrument, and Cristen Adams (EMSD, Government of Alberta) for detailed 20 discussions on the fire observations. We are grateful to Leiming Zhang (ECCC) for his helpful discussions on dry deposition.
Karen Cady-Pereira (AER) contribution was supported through funding from NASA grants NNH15CM65C, 80NSSC18K1652, and 80NSSC18K0689. Matthew Alvarado and Chantelle Lonsdale (AER) were supported by NOAA Climate Program Office grants NA13OAR4310060 and NA14OAR4310129 and NASA Applied Science grant 80NSSC19K0190. We would like to thank Nick Krotkov (NASA) for his support with the OMI NO2 product. Competing interests. The authors declare that they have no conflict of interest.

5
Code and data availability. The CrIS NH3 Version 1.5 is produced at Environment and Climate Change Canada (ECCC) and Atmospheric and Environmental Research (AER) using the CFPR algorithm, which is the research version of the near future NASA operational CrIS NH3 retrieval. The CrIS CPFR Version 1.5 ammonia data is currently available from Environment and Climate Change Canada (ECCC) upon request (Mark.Shephard@canada.ca). Python/Matlab code used to create any of the figures is available on request. We use the OMI operational NO2 standard product (SP), version-3 10 (https://disc.gsfc.nasa.gov/datasets/OMNO2_V003/summary).

Appendix/Supplemental
Appendix A: Quality Flags of Version 1.5 Here are the quality flags specified for Version 1.5. The quality flags become more conservative with increasing 5 values, thus, they are applied as greater than or equal to the level you want to use.

Appendix C: Total Column Figures
As a reference, this section contains total column maps that correspond to the mean surface ammonia for multi-year 5 globally, annual and seasonal valuessurface ammonia over North America shown in the main part of the manuscript.   .

5
The AEAI inventory calculation (from agricultural sources) is a multi-stage process involving three sets of information. The inventory is built on very detailed census data on animals for each census district collected annually by Statistics Canada. The data on fertilizer use, including forms of nitrogen, is provided by the fertilizer industry on a provincial basis. Ammonia emissions are strongly influenced by farming practices such as manure handing systems or fertilizer application method. The practices data were acquired by farm surveys, targeted to ammonia with an emphasis on timing of 10 Emission factors for the particular farm practices were obtained from scientific studies conducted in Canada and elsewhere. Some published models for emissions were used, and where possible tested with Canadian data. The emission factors were adjusted for ambient temperatures relating to the practices, the regions and the time of the practices. The manure application emissions were also adjusted for the probability of rainfall. The emission data was granulated to a 50x 50 km grid by ECCC and a finer grid is being contemplated. Note that in some cases, notably where there are few operations, the data is 10 averaged over a larger areas with more operations to ensure confidentiality for the farms.