Ammonia (NH3) is an important atmospheric pollutant with several adverse effects: deposition of NH3 can lead to eutrophication and acidification, which has negative effects on biodiversity (Sheppard et al., 2011). NH3 is a precursor of atmospheric aerosols thereby influencing the global radiation budget as well as having a negative influence on human health (Aneja et al., 2001; Baek et al., 2004). In addition, microbial oxidation of NH3 results in secondary production of nitrous oxide (N2O) and nitric oxide (NO) (Zhu et al., 2013). N2O is a very potent greenhouse gas (GHG) and NO is involved in atmospheric reactions producing tropospheric ozone (Aneja et al., 2001).

Agricultural activities, mainly manure application and management, account for around 94 % of NH3 emissions in Europe (Nielsen et al., 2017). Reliable measurements in the agricultural sector are highly important to give accurate estimates of NH3 emissions in order to reduce emissions, for example, by validation of technological improvements within the agricultural sector.

The ventilation air from animal production facilities contains a variety of chemical compounds, relatively high water content and high densities of particulate matter. This includes a great number of volatile organic compounds (VOCs) that are emitted in agricultural buildings from, e.g., silage, manure and the animals themselves (Feilberg et al., 2010; Hafner et al., 2013; Hansen et al., 2012; Ngwabie et al., 2008; Shaw et al., 2007; Yuan et al., 2017). Several NH3 analyzers are sensitive to water vapor and other gasses (Bobrutzki et al., 2010; Huszár et al., 2008; Ni and Heber, 2008; Rom and Zhang, 2010; Rosenstock et al., 2013). Such interferences can introduce errors depending on the instrument used. Furthermore, it is challenging to measure NH3 concentrations correctly due to its high water solubility and polarity, which causes adsorption on surfaces in the sampling system and within the instrument (Rom and Zhang, 2010; Shah et al., 2006; Vaittinen et al., 2014). This “sticky nature” of NH3 causes delays in the measurements, giving longer response times (Rom and Zhang, 2010; Shah et al., 2006; Vaittinen et al., 2014). It is an additional challenge to measure NH3 in livestock buildings, where dust and particles provide large surface areas for adsorption in particulate filters used to protect measuring instruments.

In many agricultural emission studies, photoacoustic spectroscopy (PAS) (Poissant et al., 2005; Rom and Zhang, 2010; Saha et al., 2010; Wu et al., 2012; Zhang et al., 2005; Zong et al., 2014) has been used, but findings by Rosenstock et al. (2013) and Liu et al. (2019) show high interference on NH3 measurements from a variety of organic compounds including carboxylic acids and alcohols. Hassouna et al. (2013) reported nonconstant bias in the results from PAS measurements on NH3 and N2O caused by organic compounds often present at agricultural sites, which makes the PAS measurements unreliable in an agricultural setting. Another issue with PAS measurements is the long response time of up to 25 min (Rom and Zhang, 2010), which lowers the time resolution of emission measurements under dynamic conditions. NH3 is underestimated by approximately 14 % and 2 % after 12.5 and 25 min, respectively (Rom and Zhang, 2010). Typical measurements in cattle barns take place at multiple points. For example, Rong et al. (2014) measured at 7 points in a dairy cow building and Ngwabie et al. (2009) measured at 11 points in a dairy cow barn. The cycle time for a typical setup in a barn would therefore be on the order of 3–5 h, which makes it impossible to monitor temporal variations with only 4–8 measurements per measurement point per day. Thus, the response time is a key parameter for equipment measuring at multiple points as done in livestock buildings.

In a few recent studies, cavity ring-down spectroscopy (CRDS) has been used to measure ammonia concentrations (Maasikmets et al., 2015; Sintermann et al., 2011) in agricultural settings. The CRDS provides measurements in real time with high sensitivity, high selectivity and a fast response time. The CRDS analyzer maintains high linearity, precision and accuracy over changing environmental conditions without the need for frequent calibration. The work of Martin et al. (2016) led to an improvement of water vapor interference calculations on Picarro's CRDS for NH3 measurements. The scaling factor error they discovered is approximately 2 % of the absolute water concentration.

Bobrutzki et al. (2010) conducted a field intercomparison of 11 atmospheric NH3 measurement techniques at concentrations up to 120 ppb including cavity ring-down spectroscopy (CRDS), and the results show a good overall agreement between the instruments on an hourly basis (R2>0.84).

Our aim is to demonstrate the performance of the CRDS analyzer for measurements of NH3 gas concentration under laboratory and field conditions. This includes identification and quantification of potential interferences by compounds present in livestock buildings by using proton-transfer-reaction mass spectrometry (PTR-MS) to document VOC concentrations in laboratory tests.

Due to the major concerns regarding measurements of NH3 with PAS (Hassouna et al., 2013; Liu et al., 2019; Zhao et al., 2012), it is pertinent that any influence of VOC on CRDS is thoroughly documented, which to our knowledge has not been done previously. The absorptions lines of, for example, acetic acid and methanol found in the HITRAN database (Gordon et al., 2017) are in the same range as the ammonia lines used for measurements in the CRDS; see Fig. 1. This highlights the importance of the current study as the absorption by VOCs in principle may cause similar interference, as reported by Rosenstock et al. (2013) and Liu et al. (2019) for PAS.

This study aims to validate CRDS for measurements in the agricultural industry, thus we test for interference with a number of compounds typically present in pig houses and cattle farms for which NH3 concentration and emission measurements are routinely carried out. Laboratory tests include determination of the response parameters linearity, response time, influence of particulate filters and chemical interference. Field tests include determination of the response parameters linearity, response time and particulate filter effect.

The CRDS analyzer had a linear response during both laboratory tests and field validation (Figs. 2 and 8) in the range from approximately 0.2 to 20 ppm; NH3 concentrations in livestock buildings are normally within this range (Heber et al., 2006; Koerkamp et al., 1998) as seen in Figs. A1 and A2. This is in agreement with the manufacturer specifications that guarantee the range from 0 to 500 ppb and an operational and optional expanded range up to 10 and 50 ppm, respectively (Picarro, 2017). The field calibrations show excellent agreement with the standard gas concentrations. The standard gas had an uncertainty of ±10 %, according to the data sheet from the manufacturer. This may well explain the small offset (< 4 %) from the obtained concentrations seen in Fig. 9, but the system causes minor variations, as the bias is considered constant. The LOD (0.35 ppb) found in the present study is comparable to the manufacturer's specifications for 1 s integration time, as seen in Table 2. The performance of a single point field calibration showed very good agreement with the expected concentrations, as seen in Fig. 9. The potential bias from the zero air might also influence the other laboratory experiments, which can explain some of the variations we see in, for example, filter response (Fig. 4) or water dependency (Fig. 5).

It is a requirement to have fast responding analyzers to understand the dynamic behavior and diurnal variations in NH3 concentrations in animal buildings. Ni and Heber (2008) suggest a response time of less than 2 min to capture temporal NH3 concentration variations. The CRDS shows sufficiently low response times under laboratory (Fig. 3 and Table A2) and field conditions (Fig. 9). These times are also comparable to < 30 s for responses to 3 ppm, as reported by the manufacturer (Picarro, 2017). Furthermore, there are no clear changes in response time without the use of an external particulate matter filter. The tested filters vary by < 10 s in response time with a tendency towards shorter response times at higher concentrations. The manufacturer reports rise- and fall times of approximately 16 s, which is at a minimum twice as much as the present results, Table A2 and Fig. 2. The response to a change from 0 to 1.02 ppm gave response times of 8.1 and 3.2 s for the rise and fall time, respectively. A change from 0 to 10.02 gave response times of 3.6 and 4.8 s for the rise and fall time, respectively (Fig. 3). The rise times with external particulate filters connected were 7.3, 3.0, 8.4 and 5.9 s, for SP1, SP2, SP4 and SP5, respectively. Response times are, in general, faster for higher concentration differences (see Fig. 4), which is ascribed to faster surface saturation. The observed concentrations of NH3 released from particulate filters exposed to air from a livestock house for 2 weeks (Fig. 10) suggest that adsorption of NH3 to the filter material, surfaces and walls is negligible. The levels released over 1 min (< 50 ppb) should be compared to a filter exposure of ammonia of > 100 ppb (ranging in to low ppm levels) over 2 weeks. These results indicate that the use of external filters gives satisfying response times and no problems with adsorption of NH3 on the filter material.

CO2 and CH4 are present in the atmosphere in relatively high concentrations compared to other trace gasses, and animals produce CO2 and CH4, thus elevated concentrations are normal in animal houses. Over a large concentration range, we observed little scatter and no interference of CO2 and CH4 on NH3 measurements, as seen in Fig. 6. The mean concentrations of both compounds are below the LOQ.

For the interference of single VOCs, it was expected that the different dilutions prepared from clean dry air mixed with humid headspace air over a VOC solution gave a correlation between water vapor and VOC concentration. This was also the case for all 10 volatile compounds. Martin et al. (2016) observe an interference from water vapor on NH3 measurements due to spectral line broadening, which the manufacturer corrects for in all models produced after the publication of that work. Our Picarro analyzer from December 2014 does not make this extra correction, and we thus expected a small water dependency for NH3, which is seen in Fig. 5. Figure 5 shows the humidity effect on the CRDS signal generated from pure deionized water and reveals a small dependency for water vapor, which the improvements suggested by Martin et al. (2016) potentially remove. Nonetheless, our results show up to 4.5 ppb NH3 for a nearly water-saturated air stream with an absolute H2O concentration of approximately 1.1 %. Thus, in the extreme case of low NH3 concentrations of 100 pbb and very humid air, a water vapor interference of up to 5 % of the NH3 signal may be present, but under normal conditions this is negligible.

The 10 tested compounds are normally present in sub-ppm levels in agricultural environments (Copeland et al., 2012; Yuan et al., 2017). As seen in Fig. 1, acetic acid and methanol have absorption lines in the wavelength area used for the Picarro CRDS measurements. A concentration range that covers a large dynamic area and exceeds the normal maximum concentration in livestock buildings was used to obtain the potential maximum interference, and we only observed very small water-induced interferences. Figure 7 shows the contribution from the single VOCs that were either corrected for water or uncorrected, and, as can be seen, the interferences are insignificant in general. Overall, the difference between high and low concentration for a single VOC was approximately 1–2 ppb NH3 (Fig. 7), except for acetic acid with a difference of nearly 4 ppb. It should be noted that the water to VOC relation differs for the different compounds. With water correction applied, only 1-propanol and acetic acid have increasing tendencies, where NH3 concentration increases about 0.8 ppb for an increase of 7.5 ppm of 1-propanol and 2.9 ppb for an increase of nearly 1.8 ppm of acetic acid. Acetic acid, 2-propanol and propionic acids were the only compounds with absolute humidity above 1 % as we used higher flow rates over the headspace to obtain the targeted concentrations. The very moist sample of acetic acid had a corrected maximum of 3.7 ppb NH3, which is very low compared to, for example, concentrations in animal buildings, which typically range from < 1 to 20 ppm but can, in extreme cases, reach up to 50 ppm (Heber et al., 2006; Koerkamp et al., 1998). Thus, errors of a few ppb, introduced by humidity effects, would have a small impact on the results. For the given setup, the interferences from water vapor were in the same order of magnitude as the LOQ of 1.15 ppb. For more than half of the VOCs, the NH3 concentration falls below the LOQ for all or most measurement. This demonstrates a very low interference from the investigated VOCs.

The results of this study clearly demonstrate the advantage of CRDS with only few and small interferences, whereas the performance of the photoacoustic analyzers under similar circumstances studied by Liu et al. (2019) showed much more interference from non-NH3 species.

Our tests of the Picarro CRDS showed great linearity during both laboratory and fieldwork. The response times with respect to concentration changes were sufficiently low to measure temporal variations in NH3 concentrations in livestock emissions. Examinations of external particulate filters lead to no clear recommendations for filter material, but all filters gave acceptable response times and only small amounts of NH3 adsorption compared to background levels. Potential interferences were tested for 10 VOCs in known concentrations and the compounds gave negligible interference on CRDS NH3 measurements.