Repeatability and Reproducibility of Measures of Bovine Methane Emissions Recorded Using a Laser Detector.

Methane (CH4) emissions represent a worldwide problem due to their direct involvement in atmospheric warming and climate change. Ruminants are among the major players in the global scenario of CH4 emissions, and CH4 emissions are a problem for feed efficiency since enteric CH4 is eructed to the detriment of milk and meat production. The collection of CH4 phenotypes at the population level is still hampered by costly and time-demanding techniques. In the present study, a laser methane detector was used to assess repeatability and reproducibility of CH4 phenotypes, including mean and aggregate of CH4 records, slope of the linear equation modelling the aggregate function, and mean and number of CH4 peak records. Five repeated measurements were performed in a commercial farm on three Simmental heifers, and the same protocol was repeated over a period of three days. Methane emission phenotypes expressed as parts per million per linear meter (ppm × m) were not normally distributed and, thus, they were log-transformed to reach normality. Repeatability and reproducibility were calculated as the relative standard deviation of five measurements within the same day and 15 measurements across three days, respectively. All phenotypes showed higher repeatability and reproducibility for log-transformed data compared with data expressed as ppm × m. The linear equation modelling the aggregate function highlighted a very high coefficient of determination (≥0.99), which suggests that daily CH4 emissions might be derived using this approach. The number of CH4 peaks resulted as particularly diverse across animals and therefore it is a potential candidate to discriminate between high and low emitting animals. Results of this study suggest that laser methane detector is a promising tool to measure bovine CH4 emissions in field conditions.


Introduction
Greenhouse gases (GHG) represent a worldwide problem for their impact on global warming and climate change. Anthropic activities related to the primary and secondary sectors are responsible for the majority of GHG released in the atmosphere. Methane (CH 4 ) is considered one of the most detrimental GHG, with a global warming potential 28-fold greater than that of carbon dioxide [1]. The livestock sector is estimated to account for 18% of global anthropogenic GHG emissions, and ruminants are the main emitters of atmospheric CH 4 [2]. Ruminants produce from 250 to 500 L of CH 4 per head per day and are estimated to account for 8% to 10% of global warming in the next century [3]. Besides environmental and ecological issues, CH 4 emissions in ruminants represent a problem in terms of feed and production efficiency, since 2% to 12% of cattle gross energy intake is lost through CH 4 eructation [4].
For decades, the scientific community has focused on different strategies and approaches aiming at reducing ruminants CH 4 emissions, which, at least partially, succeeded in lowering environmental impact and increasing production efficiency at the same time. Nevertheless, large-scale collection of individual phenotypes related to CH 4 emissions is still hampered by high recording costs and time-demanding techniques, including respiration chambers, GreenFeed (C-Lock Inc., Rapid City, SD, USA), and sniffer-based systems [5]. For these reasons, the development of alternative, cost-effective, and rapid tools for CH 4 measurement or prediction is an emerging research field of global interest. Until now, mid-infrared spectroscopy [6], archaeol quantification [7], sulphur hexafluoride tracer, and CO 2 to CH 4 ratio [8] have been investigated as indirect proxies for CH 4 production in cattle. Laser methane detector (LMD) has been proposed as an alternative instrument to directly measure CH 4 emissions. The sensitivity and accuracy of LMD have been assessed in controlled conditions, by comparing data acquired through LMD with those measured through the respiration chamber (coefficient of determination, R 2 from 0.64 to 0.97) [9,10] and GreenFeed (R 2 = 0.64) [11]. However, it is still unclear if LMD may be appropriate in terms of repeatability and reproducibility to accurately measure CH 4 emissions in field conditions. This research question was investigated by assessing repeatability and reproducibility of different CH 4 -related phenotypes, including mean and aggregate of CH 4 records, slope of the linear equation modelling the aggregate function, and mean and number of CH 4 peak records. For this purpose, CH 4 emissions were measured on Simmental heifers as a case study.
Methane emissions were measured through Laser Methane Mini (Crowcon, Abingdon, UK) and CH 4 was expressed as parts per million per linear meter (ppm × m). Each measurement was performed by pointing the laser to the nostril of a single animal for 5 min, at a distance of 3 m, according to the protocol proposed by Chagunda et al. [12]. Each heifer was restrained in a single pen but it could perform natural activities such as standing, lying, eating, and ruminating. The LMD was set to detect one record of CH 4 emission every 0.5 s, for a total of 600 records for each measurement (5 min).
Every single measurement was forwarded via Bluetooth from the laser device to a Lenovo Tab E7 (Lenovo, Hong Kong, China) tablet, equipped with an Android operating system and Gas Viewer (Tokyo Gas Engineering, Tokyo, Japan) application, saved as .csv file, sent to a dedicated e-mail box, and downloaded in a computer workstation to allow for the local storage of data. The methanogram plot resulting from a single measurement is depicted in Figure 1a. Five consecutive measurements were performed within a day for each animal, which achieved a total of 15 measurements and 9000 records per day. All measurements were performed at the same time of the day (between 8:00 a.m. and 9:30 a.m.), and in the same order for the three animals involved in the study. The protocol was repeated for three consecutive days. As a result, 15 measurements were performed across three days for each heifer, achieving a final dataset of 45 measurements and 27,000 records.

Repeatability and Reproducibility of Phenotypes
Phenotypes considered in the present study were: (i) mean of CH4 and lnCH4 records, (ii) aggregate of CH4 and lnCH4 records, (iii) slope of the linear equation modelling the aggregate

Data Editing
The distribution of CH 4 emissions expressed as ppm × m is depicted in Figure 1b. Records exceeding three standard deviations from the mean were discarded from the original raw dataset (ppm × m, n = 27,000), which led to 26,449 records available for subsequent analysis. The original raw data of CH 4 emissions were log e -transformed (lnCH 4 ) to achieve normality and homogeneity of variances. The distribution of the probability function of lnCH 4 is reported in Figure 1c. Additionally, in this case, records exceeding three standard deviations from the mean were deleted from the original dataset (lnCH 4 , n = 27,000), which resulted in 26,856 records available for subsequent analysis.

Repeatability and Reproducibility of Phenotypes
Phenotypes considered in the present study were: (i) mean of CH 4 and lnCH 4 records, (ii) aggregate of CH 4 and lnCH 4 records, (iii) slope of the linear equation modelling the aggregate function, (iv) mean of CH 4 and lnCH 4 peak records, calculated on the last decile of the distribution, and (v) number of CH 4 and lnCH 4 peak records. Repeatability of the previously mentioned phenotypes was calculated as the relative standard deviation (RSD r ) of five consecutive measurements carried out within the same day and within the same animal. Similarly, reproducibility of phenotypes was calculated as the relative standard deviation (RSD R ) of 15 measurements collected across three days of analyses and within the same animal, as proposed by Chagunda et al. [12], Franzoi et al. [13], and Niero et al. [14].  (Table 1). Average CH 4 emissions of the present study was about half the mean value reported by Chagunda et al. [12]. In terms of variability, the standard deviation in our study was lower than that obtained by Chagunda et al. [12]. These differences are likely due to the different experimental design adopted by Chagunda et al. [12] who measured CH 4 in lactating cows. Methane emissions in the present study were lower even when compared with Sorg et al. [10], and this difference is likely due to (i) the diverse conditions of measurements, (ii) different categories of animals, and (iii) the different feed administered to animals in the two studies. In the present study, measurements were carried out on Simmental heifers housed in an aerated barn, whereas Sorg et al. [10] measured CH 4 exhaled from Holstein Friesian lactating cows in the spent air of the respiration chamber. Overall mean, mode, and median of CH 4 expressed as ppm × m were rather different from each other and thus, skewness and kurtosis were relatively far from zero (Table 1). Visual inspection of data distribution (Figure 1) and Shapiro Wilk's test suggested that CH 4 expressed as ppm × m was not normally distributed (p < 0.05). The log e -transformation of CH 4 produced a much more normal trait (lnCH 4 ), as reported in Figure 1. Methane emissions averaged 4.45 and 4.46 in the pre-edited and post-edited dataset, respectively. Mode and median were 4.16 and 4.48, and Shapiro Wilk's test was not statistically significant (p > 0.05), both in the pre-edited and in the post-edited datasets. Skewness and kurtosis were close to zero in the pre-edited and post-edited dataset. Logarithmic transformations were proposed also by Ali and Shook [15] and Benedet et al. [16] to achieve normal distributions and homogeneity of variances for the milk somatic cell count and blood β-hydroxybutyrate, respectively.

Mean and Aggregate of CH 4 Emissions
The precision of LMD for determining the mean of CH 4 emissions was assessed through RSD r and RSD R ( Table 2). Methane emissions expressed as ppm × m showed poor repeatability and reproducibility. Repeatability was always greater than 40%, with the minimum value reported for heifer 1 in day 3 (41.57%), and the maximum (74.48%) for heifers 3 and 2 in day 1 and 3, respectively. Reproducibility mirrored the same great variability, ranging from 52.43% (heifer 1) to 56.03% (heifer 2). On the other hand, lnCH 4 highlighted notable improvements in terms of RSD r and RSD R . Repeatability varied from 8.93% (heifer 1 in day 3) to 14.85% (heifer 1 in day 1), whereas reproducibility ranged from 11.98% (heifer 2) to 15.35% (heifer 3). Still, such repeatability and reproducibility values remain greater than values from other studies describing the precision of analytical methods carried out under controlled conditions [13,17]. Overall, the variability observed in the present study was likely due to (i) different physiological activities and behaviours of the animals throughout measurements (e.g., eating, standing, and ruminating) [12], and (ii) the environmental factors inherent to the on-field approach, with particular regard to temperature, wind velocity, proximity of other animals, and humidity [5].
The aggregate value of CH 4 records, the slope, and the coefficient of determination (R 2 ) of the linear equation modelling the aggregate function are reported in Table 3. Aggregate values of CH 4 emissions showed the lowest value for heifer 3 in day 1, being equal to 153,393 ppm × m and 11,212 lnCH 4 . The greatest aggregate value was observed for heifer 1 in day 2 (363,110 ppm × m and 14,313 lnCH 4 ). The slopes of the linear equation mirrored the tendency of aggregate values, being lower and greater concurrently with lower and greater aggregates. Although the agreement between aggregates and slopes may support and reinforce the significance of these traits, it can be argued that the consideration of both phenotypes is redundant since their biological meaning is likely the same. Overall, the R 2 of the aggregate functions of lnCH 4 (0.999) was greater than the R 2 of the aggregate functions of CH 4 (0.989 to 0.997). Such a great accuracy suggests that the linear equation modelling the aggregate function might be used in the future to estimate long-term or daily CH 4 emissions.

Peaks of CH 4 Emissions
The precision of LMD for determining peaks of CH 4 and lnCH 4 emissions was assessed through RSD r and RSD R (Table 4). Repeatability and reproducibility for peaks of CH 4 emissions showed greater precision compared with the same indexes calculated for means of CH 4 emissions. This translated into a relatively low RSD r , ranging from 17.58% to 20.15% for CH 4 , and 4.50% to 5.34% for lnCH 4 , and RSD R , which was always lower than 20% and 5% for CH 4 and lnCH 4 , respectively. The average values for peaks of CH 4 emissions did not vary much across different animals, which suggests that this phenotype might be not adequate to discriminate between high and low CH 4 emitters. Such a low variability was somewhat expected. Peaks of CH 4 emissions were defined as records belonging to the highest decile of both the datasets, which leads to a considerable decrease of variability. For this reason, the number of peaks emitted from every single animal is more informative because it is more differentiated across heifers. A similar approach was adopted by Bobbo et al. [18] in the study of new selection tools for mastitis resistance in dairy cows assuming different alternative somatic cell count traits as indicators of the mastitis event. Table 2. Number of records, mean, repeatability relative standard deviation (RSD r , %) calculated within the day on five measurements, and reproducibility relative standard deviation (RSD R , %) calculated across days on 15 measurements for methane (CH 4 ) and log e -transformed methane (lnCH 4 ) emissions.

Conclusions
The present research assessed the repeatability and the reproducibility of phenotypes related to CH 4 emissions, measured through LMD using Simmental heifers as a key study. The distribution of emission events expressed as ppm × m showed a significant deviation from the normal distribution, but the logarithmic transformation of the data led to normality. Repeatability and reproducibility were much better for lnCH 4 than for CH 4 . The coefficient of determination of the linear equation modelling the aggregate function showed high precision. Such results are promising since these equations might be used to estimate daily or long-term CH 4 emissions. Peaks of CH 4 emissions were rather different across animals in terms of the number of events but were homogeneous in terms of average values. For this reason, the number of peaks may be an interesting phenotype to discriminate between high and low emitting animals. Overall, results of the present study indicate that measures carried out through LMD are fairly repeatable and reproducible. Therefore, in terms of accuracy, LMD may be considered as a promising tool enabling to measure bovine CH 4 emissions in field conditions at relatively low costs. Future studies will focus on the application of LMD for large-scale studies to assess sources of variation of CH 4 emissions.