TY - JOUR
T1 - Technical note
T2 - Method for improving precision of in-parlor milk meters and adjusting milk weights for stall effects
AU - Andreen, D. M.
AU - Salfer, I. J.
AU - Ying, Y.
AU - Reinemann, D. J.
AU - Harvatine, K. J.
N1 - Funding Information:
Funding was provided in part by Agriculture and Food Research Initiative Competitive Grant No. 2015-67015-23358 from the USDA National Institute of Food and Agriculture (Washington, DC) and Penn State University (University Park, PA), including USDA National Institute of Food and Agriculture Federal Appropriations under project number PEN04539 and accession number 1000803. We thank R. Shepardson, B. Bomberger, and C. Matamoros, all of Penn State University, for assistance with data collection. The authors recognize key discussions and contributions from L. Armentano (University of Wisconsin, Madison). The authors have not stated any conflicts of interest.
Publisher Copyright:
© 2020 American Dairy Science Association
PY - 2020/6
Y1 - 2020/6
N2 - Milk yield is a fundamental observation in most dairy experiments and is commonly determined using integrated milk meters that measure milk weight as the cow is being milked. These meters are heavily used in a harsh environment and often are not regularly calibrated, so calibration errors and mechanical problems may create artificial variation in milk weight data. Additionally, direct calibration by collection of milk in a bucket is difficult and imperfect because the use of the bucket may affect yield recorded by the milk meter. The objective of this work was to define a method to easily check parlor meter precision and adjust milk weight values for variation between individual stalls in a parlor. Because most cows are milked in a different stall at each milking, it has been proposed that stall deviations that represent the fixed effect of stall on milk weight could be statistically determined. Individual milk weights from 14 milkings across 7 d from approximately 200 cows were collected from the Penn State dairy farm, which is equipped with a double-10 herringbone parlor with an Afimilk 2000 milking system (S.A.E. Afikim, Afikim, Israel). Milk yield was measured automatically by in-line flow through milk meters (Afi 200; S.A.E. Afikim). The effect of stall on milk weight was modeled using a mixed model that included the fixed effect of stall and the random effects of day, milking time, and cow. First, stall deviations were calculated as the stall least squares means (LSM) minus the average LSM to identify malfunctioning meters requiring service (e.g., deviation exceeding 1 kg). A correction factor for each stall was then generated by dividing the LSM of each stall by the average LSM. Milk yields were then corrected by multiplying the meter weight value by the correction factor. To determine the effect of the correction, raw and corrected meter values were compared with weight of milk collected in a bucket (n = 3/stall). The corrected values had a 5% greater coefficient of determination than raw meter values (0.89 vs. 0.84) and had a lower average percent difference from the bucket milk weight compared with raw meter values (12.6% vs. 13.5%). The method was then used in 3 experiments with 121, 140, and 683 milk yield observations. In all data sets, correcting milk weights slightly improved model fit and had minimal effect on model term standard errors. However, this validation was completed in a parlor where the method was routinely used to identify stalls requiring service; the effect of stall corrections is expected to be larger in parlors without frequent monitoring. Stall deviations are expected to be due predominantly to calibration of the meter but also could be due to differences in pulsation or other stall-specific factors that result in a change in milk yield. It is important to account for these other sources of milk weight variation that are unrelated to treatment. Modeling the effect of stall is a simple, convenient, and low-cost method to monitor and improve milk meter precision and functionality and can be used to reduce artificial variation and experimental error.
AB - Milk yield is a fundamental observation in most dairy experiments and is commonly determined using integrated milk meters that measure milk weight as the cow is being milked. These meters are heavily used in a harsh environment and often are not regularly calibrated, so calibration errors and mechanical problems may create artificial variation in milk weight data. Additionally, direct calibration by collection of milk in a bucket is difficult and imperfect because the use of the bucket may affect yield recorded by the milk meter. The objective of this work was to define a method to easily check parlor meter precision and adjust milk weight values for variation between individual stalls in a parlor. Because most cows are milked in a different stall at each milking, it has been proposed that stall deviations that represent the fixed effect of stall on milk weight could be statistically determined. Individual milk weights from 14 milkings across 7 d from approximately 200 cows were collected from the Penn State dairy farm, which is equipped with a double-10 herringbone parlor with an Afimilk 2000 milking system (S.A.E. Afikim, Afikim, Israel). Milk yield was measured automatically by in-line flow through milk meters (Afi 200; S.A.E. Afikim). The effect of stall on milk weight was modeled using a mixed model that included the fixed effect of stall and the random effects of day, milking time, and cow. First, stall deviations were calculated as the stall least squares means (LSM) minus the average LSM to identify malfunctioning meters requiring service (e.g., deviation exceeding 1 kg). A correction factor for each stall was then generated by dividing the LSM of each stall by the average LSM. Milk yields were then corrected by multiplying the meter weight value by the correction factor. To determine the effect of the correction, raw and corrected meter values were compared with weight of milk collected in a bucket (n = 3/stall). The corrected values had a 5% greater coefficient of determination than raw meter values (0.89 vs. 0.84) and had a lower average percent difference from the bucket milk weight compared with raw meter values (12.6% vs. 13.5%). The method was then used in 3 experiments with 121, 140, and 683 milk yield observations. In all data sets, correcting milk weights slightly improved model fit and had minimal effect on model term standard errors. However, this validation was completed in a parlor where the method was routinely used to identify stalls requiring service; the effect of stall corrections is expected to be larger in parlors without frequent monitoring. Stall deviations are expected to be due predominantly to calibration of the meter but also could be due to differences in pulsation or other stall-specific factors that result in a change in milk yield. It is important to account for these other sources of milk weight variation that are unrelated to treatment. Modeling the effect of stall is a simple, convenient, and low-cost method to monitor and improve milk meter precision and functionality and can be used to reduce artificial variation and experimental error.
UR - http://www.scopus.com/inward/record.url?scp=85083328038&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85083328038&partnerID=8YFLogxK
U2 - 10.3168/jds.2019-17479
DO - 10.3168/jds.2019-17479
M3 - Article
C2 - 32307171
AN - SCOPUS:85083328038
SN - 0022-0302
VL - 103
SP - 5162
EP - 5169
JO - Journal of Dairy Science
JF - Journal of Dairy Science
IS - 6
ER -