Sustainability & efficiency

Predicting methane production in dairy cows using milk spectra and genomic data

---

Abstract

Enteric methane (CH4) is a major source of greenhouse gas emissions from dairy farming. Routine recording of CH4 emissions nationwide can be challenging and costly. Fourier-transform mid-infrared (FT-MIR) spectroscopy is a powerful phenotyping tool that offers high-throughput, low-cost, non-invasive, and real-time predictions. We aimed to evaluate the feasibility of predicting CH4 production using multiple data sources, including productive traits, FT-MIR, and genomic data. Data were collected from 1,601 mid-lactation Holstein cows enrolled in 53 trials (8 to 10 weeks) at 11 farms across the U.S. Daily CH4 emissions were measured using GreenFeed systems (C-Lock Inc., Rapid City, SD). Additional data included daily milk yield (MY), milk composition, and milk FT-MIR spectra from morning and afternoon milkings (on a weekly or monthly basis), and body weight (BW) records (every 2-3 weeks). All cows were genotyped with 69,200 SNPs. Milk spectra data consisted of FT-MIR absorbances from different instruments: Foss FT6000 and FT+ (FOSS Analytical, Hilleroed, Denmark) and FTS (Bentley, Chaska, MN). The wavenumbers (wvn) ranged from 5,008 to 925 cm-1 (1,060 wvns) for Foss instruments and from 649.03 to 3,998.59 cm-1 (899 wvns) for Bentley instruments. The FT-MIR spectra were standardized using the piecewise direct standardization method, retaining 535 wvns in the final dataset. Methane production was measured using GreenFeed systems (C-Lock Inc., Rapid City, SD). We evaluated three alternative models to predict CH4 production using a Bayesian approach: model XB with cohort (36 levels combining trial and treatment), days in milk (DIM, 50-200), MY, and BW0.75 as predictors; model S as model XB plus FT-MIR data; and model GS as model S plus genomic data. In addition, we estimated the heritability for all 535 FT-MIR wvns using an animal model including cohort and DIM as fixed effects. Model predictive ability was evaluated using 10-fold, 5-replicate (CV105), leave-one-trial-out (LOTO), and leave-one-farm-out (LOFO) cross-validations. The cohort effect was removed from the models used in LOTO and LOFO analyses. Model GS achieved the highest predictive correlations between the predicted and observed CH4 production (CV105: 0.65, LOTO: 0.57, and LOFO: 0.57), followed by model S (CV105: 0.63, LOTO: 0.53, and LOFO: 0.51) and model XB (CV105: 0.55, LOTO: 0.49, and LOFO: 0.28). Overall, the inclusion of FT-MIR and genomic data improves CH4 predictions. However, the genomic data provided only modest additional improvement, likely due to the moderate to high heritability of the milk FT-MIR (ranging from 0.40±0.05 to 0.66±0.05). Our findings support previous evidence that FT-MIR can predict CH4 emissions in dairy cows, providing a valuable tool for optimizing herd management.

Keywords: 2026

How to Cite:

Appuhamy, R., Baldwin, R., French, E., James, L., Kalscheur, K., Koltes, J., Mantonavi, H., Martinez-Boggio, G., Parker Gaddis, K., Peà±agaricano, F., Santos, J., Sarmikasoglou, E., Tempelman, R., VandeHaar, M., Weigel, K. & White, H., (2026) “Predicting methane production in dairy cows using milk spectra and genomic data”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2283663. doi: https://doi.org/10.31274/wcgalp.23507

Rights: 1