Agreement between milk fat, protein, and lactose observations collected from the Dairy Herd Improvement Association (DHIA) and a real-time milk analyzer

Abstract

The objective of this study was to quantify the agreement between AfiLab real-time milk analyzer (Afimilk, Kibbutz Afikim, Israel) measures for fat, protein, and lactose based on near-infrared spectrum light scattering, and those collected on Dairy Herd Improvement Association (DHIA) test days and measured with the Bentley 2000 analyzer (Bentley Instruments Inc., Chaska, MN), which uses mid-infrared spectrum light. The AfiLab data were collected twice daily for each milking cow in the herd at 12-h intervals from the double-12 parlor at the University of Florida Dairy Unit (Hague, FL) from January 2010 to December 2011. Bentley data for the 23 DHIA test days in 2010 and 2011 were also obtained. Approximately 450 cows were tested each month. Tested milk was collected during 1 milking each month, alternating monthly between morning and evening milkings. AfiLab data were matched with Bentley fat and protein (n=10,273; 23 test days) and lactose (n=6,741; 16 test days). Overall means ± standard deviations (SD) of monthly mean Bentley fat, protein, and lactose were 3.74 ± 0.80%, 3.06 ± 0.37%, and 4.76 ± 0.30%, respectively. Overall means ± SD of monthly mean AfiLab minus Bentley observations were -0.08 ± 0.12 percentage points (PP) for fat (n=23), 0.02 ± 0.11 PP for protein (n=23), and -0.02 ± 0.08 PP for lactose (n=16). Overall means ± SD of monthly within-test-day SD of AfiLab minus Bentley observations were 0.66 ± 0.11 PP for fat, 0.27 ± 0.03 PP for protein, and 0.26 ± 0.03 PP for lactose. Overall means ± SD of the corresponding monthly correlations were 0.59 ± 0.09 for fat, 0.67 ± 0.04 for protein, and 0.46 ± 0.08 for lactose. Averaging the AfiLab observations from up to 6 milkings before and after the test-day milking improved the agreement for protein and lactose but not for fat. Averaging the 13 protein observations improved the mean difference to 0.01 ± 0.10 PP and the SD of the difference to 0.23 ± 0.03 PP. The correlation increased to 0.78 ± 0.04. Averaging the 13 lactose observations improved the SD of the difference to 0.23 ± 0.02 PP, but the mean of the difference decreased to -0.03 ± 0.09 PP. The correlation for lactose increased to 0.55 ± 0.05. Generally, AfiLab slightly overestimated low Bentley components and underestimated high Bentley components. We found some evidence for a systematic cow effect on lack of agreement for lactose, but not for fat and protein. The agreement between AfiLab and Bentley observations was better for protein and lactose than that for fat. Combinations of AfiLab observations from various milkings improved the agreement for protein and lactose. AfiLab real-time milk analyzers may be helpful to estimate DHIA observations.

Visible and near-infrared bulk optical properties of raw milk

“…The industry has recently adopted this technology and implemented it into milking systems to measure the major milk components online (Katz et al, 2011(Katz et al, , 2003Pinsky et al, 2013). However, despite the continuous recalibration, their accuracy and robustness is still not sufficient to support cow health management (Kaniyamattam and De Vries, 2014). This is mainly be-cause the measured spectral signals are, next to absorption, considerably influenced by the physical properties of the milk in terms of the quantity and size of the fat globules and CN micelles.…”

Determination of Lactose in Milk by Raman Spectroscopy

“…In recent years, some convenient, sensitive, and quantitative real-time lactose determination methods were developed, involving enzyme biosensors (Ammam and Fransaer 2010;Safina, Ludwig, and Gorton 2010;Yakovleva et al 2012) and nearinfrared spectroscopy (Melfsen, Hartung, and Haeussermann 2012;Kaniyamattam and De Vries 2014). However, enzyme biosensors may be affected by the presence of galactose and glucose present in milk and milk products (Conzuelo et al 2010).…”

Bayesian integration of sensor information and a multivariate dynamic linear model for prediction of dairy cow mastitis

“…All sensors were obtained from Afimilk. The accuracies of the sensors for fat, protein, and lactose have been evaluated by Kaniyamattam and De Vries (2014). The 7 continuous variables from these sensor data were used in the DLM.…”