Farm-level nutritional factors associated with milk production and milking behavior on Canadian farms with automated milking systems

The objective of this study was to describe nutritional strategies utilized on Canadian dairy farms with automated milking systems (AMS), both at the feed bunk and the concentrate offered at the AMS, as well as to determine what dietary components and nutrients, as formulated, were associated with milk production and milking behaviors on those farms. Formulated diets, including ingredients and nutrient content, and AMS data were collected from April 1, 2019, until September 30, 2020, on 160 AMS farms (Eastern Canada [East] = 8, Ontario [ON] = 76, Quebec [QC] = 22, and Western Canada [West] = 54). Both partial mixed ration (PMR) and AMS concentrate samples were collected from May 1 to September 30, 2019, on 169 farms (East = 12, ON = 63, QC = 42, West = 52). AMS milking data were collected for 154 herds. For each farm (n = 160), milk recording data were collected and summarized by farm to calculate average milk yield and components. Multivariable regression models were used to associate herd-level formulated nutrient composition and feeding management practices with milk production and milking behavior. Milk yield (37.0 ± 0.3 kg/d) was positively associated with the PMR ether extract (EE) concentration (PMR % EE; +0.97 kg/d per percentage point ( p.p .) increase) and with farms that fed barley silage as their major forage source on farm (n = 16; +2.18 kg/d) compared with haylage (n = 42), while farms that fed corn silage (n = 96; +1.23 kg/d) tended to produce more milk than farms that fed haylage. Greater milk fat content (4.09 ± 0.28%) was associated with greater PMR-to-AMS concentrate ratio (+0.02 p.p. per unit increase) and total diet net energy for lactation (+0.046 p.p. per 0.1 Mcal/kg increase), but lesser


INTRODUCTION
Automated milking systems (AMS) continue to be implemented on dairy farms to assist with labor shortages, improve farmers' personal lives, and improve cow health and production (Tse et al., 2017;Tse et al., 2018).Researchers have identified various housing and management factors associated with successful transition and adaptation to AMS.Greater milk yield has been associated with: increased feed availability (Deming et al., 2013); a new barn and fewer cows per AMS (Tremblay et al., 2016); increased feed push-up frequency and sand bedding (Matson et al., 2021); as well as lower herd prevalence of lameness (Matson et al., 2022).Researchers have also reported that a lower ratio of cows to AMS units, improved herd locomotion score, and free-flow cow traffic systems are associated with greater milking frequency (King et al., 2016;Deming et al., 2013;Matson et al., 2022).
In addition to housing and management factors, nutrition is a key component to the success of farms using AMS.For farms to realize the full benefits of an AMS, cows are required to voluntarily visit the AMS unit to be milked.Prescott et al. (1998) demonstrated that cows had a stronger motivation to milk when incentivized with concentrate offered at the AMS compared with the urge to empty their udder.The amount of concentrate offered at the AMS may vary, depending on nutritional strategies, grouping, DIM, and parity (Bach and Cabrera, 2017).However, feeding large amounts of concentrate in the AMS often resulted in great variability in milk production and milking behaviors (Bach and Cabrera, 2017).Some researchers have demonstrated that increasing AMS concentrate allocation resulted in increased milkings, milk yield, AMS visits, and total DMI (Halachmi et al., 2005;Schwanke et al., 2019;Henriksen et al., 2018), while others demonstrated no improvements or reductions in milking performance (Milgliorati et al., 2005;Tremblay et al., 2016;Henriksen et al., 2019).Aside from affecting milking behavior and production, increasing supplementation of concentrate in the AMS has resulted in the reduction of 0.54 to 1.58 kg of PMR intake for every 1-kg increase in AMS concentrate, depending on the energy density of the PMR (Schwanke et al., 2022;Hare et al., 2018).Increased nutrient density or large concentrate allowances offered in the AMS are often paired with a low nutrient-dense PMR (Bach and Cabrera, 2017); this substitution effect of a lower energy-dense PMR for AMS concentrate has resulted in similar (Bach et al., 2007) or greater total DMI (Schwanke et al., 2019).Despite these studies, little is known about the optimal PMR composition, either in ingredient or nutrient composition.
Not only is AMS concentrate allocation important, but also the ingredients and nutrient composition of that concentrate.Cows have shown a preference for pelleted AMS concentrate compared with mash and texturized, while also preferring the primary ingredient to be a barley and oat mixture or wheat over just corn or barley-based AMS concentrate (Rodenburg et al., 2004;Madsen et al., 2010).Feeding high-starch concentrates can affect milk production and composition through its effects on appetite, NDF digestibility, and ruminal pH, in addition to increasing the risk of lameness (Oba and Wertz-Lutz, 2011).However, the negative impacts of feeding high starch concentrate on milk yield, milk composition, and the number of AMS visits were not observed when concentrate allowance was limited to 3 kg/d (Halachmi et al., 2006) compared with 8 kg/d (Miron et al., 2004a).
To our knowledge, no farm-level studies have been conducted to assess the variation in nutritional strategies, including diet composition and formulation, on commercial AMS farms.The objective of this study was to describe the variation in nutrient composition and dietary ingredients of both AMS concentrates and PMR offered and formulated on Canadian dairy farms milking with an AMS, as well as to determine what formulated dietary components and nutrients were associated with milk production and milking behavior on those farms.Data for this study were collected as part of a larger study (Matson et al., 2021) in which the herd-level housing and management strategies of AMS farms across Canada were benchmarked, and associations of those herd-level housing factors and management practices were associated with milk production and quality.

MATERIALS AND METHODS
Animal use, data collection, and experimental procedures used throughout this trial complied with the guidelines of the Canadian Council of Animal Care (2009) and were approved by the University of Guelph Animal Care Committee (Protocol #3963), University of Guelph Research Ethics Board (REB 19-01-012), University of Calgary Research Ethics Board , and University of Saskatchewan Research Ethics Board (Beh ID 1305).

Farm Recruitment
As part of the larger study (Matson et al., 2021), 177 commercial AMS dairy farms were recruited across Canada (11 in Eastern Canada (East), 27 in Quebec (QC), 82 in Ontario (ON), and 57 in Western Canada (West); Table 1) from the 750 dairy farms that were eligible (23.6% of eligible farms) based on the use of an AMS and enrollment in DHI milk recording as of December, 2018.The regional distribution of the farms included in the study well represent the population and distribution of Canadian AMS herds .The target sample size was calculated through power analysis using WinPepi version 11.65 (Abramson, 2011).Estimates of variation for milk, fat, and protein yields, which were our primary response variables, were based on values from similar herd-level studies previously reported (King et al., 2016;Tremblay et al., 2016).Matson et al. (2021)

Balanced Diets from Nutritionists
With permission from each producer, balanced diet summaries were collected retrospectively from farm nutritionists detailing the PMR (8 in East, 21 in QC, 76 in ON, and 54 in West) and the AMS concentrate (8 in East, 20 in QC, 72 in ON, and 55 in West) for the observational period of April 1, 2019, to September 30, 2020.Of the 177 study farms, 160 supplied nutrient information for PMR, with 5 of those 160 farms failing to provide nutritional information on their AMS concentrate.Formulated nutrient composition, DMI, and feed ingredient proportions were recorded and summarized for each farm individually by calculating a weighted average based on the number of days each formulated diet received was fed during the observational period.From the summarized diets for each farm, the major forage of the PMR, major PMR grain, AMS concentrate form, major ingredient of the AMS concentrate, major fatty acid (FA) of the fat supplement in the PMR, and the major protein source of the AMS concentrate were determined and summarized by region (Tables 2 and 3).
For analyses, the major forage source used on farm was categorized as haylage if alfalfa silage, grass silage, baleage, and mixed alfalfa/grass silage were used.The farms feeding most hay (n = 3) were excluded from the analysis, thus farms were categorized as either corn silage, barley silage, or haylage as the major forage types.Similarly for the AMS concentrate, dry corn and high moisture corn were categorized as corn, while wheat grain, ground wheat, wheat shorts, and wheat middlings were all categorized as wheat.The farms that fed wheat grain did not disclose to what level of processing the wheat underwent.This consolidation resulted in classifying wheat, corn, and barley grains as the major ingredients in the AMS concentrate on farms, while excluding the farms where no specific ingredient information was supplied (n = 32).The nutrient composition of the PMR and AMS concentrate from the formulated diets provided by the nutritionists were also summarized by region (Table 4).Some individual nutrient components were not available for all farms, resulting in a smaller sample size for nutrient composition (Tables 4 and 5) compared with ingredient and The ratio of PMR expected DMI to the expected DMI of the AMS concentrate offered from the formulated diets.

4
The % of DM that was from forage in the PMR offered from the formulated diets.

5
The total % of DM that was from forage in the PMR and the AMS concentrate offered from the formulated diets.
dietary information of the PMR and AMS concentrate (Tables 2 and 3).

Feed Sampling
Fresh feed samples (one PMR and one AMS concentrate) were collected during the herd visits, resulting in 177 PMR samples (11 from East, 27 from QC, 82 from ON, and 57 from West) and 169 AMS concentrate samples (11 from East, 27 from QC, 79 from ON, and 52 from West).Feed samples were frozen and shipped to a central collection point where they were stored at −20°C upon collection and later thawed for 24 h before being oven-dried at 60°C for 48 h for DM analysis.Dried PMR and AMS concentrates were ground to pass through a 1-mm screen (Model 4 Wiley Laboratory Mill, Thomas Scientific, Swedesboro, NJ).Ground samples were sent to A & L Laboratory Services Inc. (London, ON, Canada) for analysis of ash (550°C; AOAC International, 2000, method 942.05),ADF (AOAC International, 2000, method 973.18),NDF with heat-stable α-amylase and sodium sulfite (AOAC International, 2000, method 2002.04), CP (N × 6.25;AOAC International, 2000, method 990.03;Leco FP-628 Nitrogen Analyzer, Leco Corp., St. Joseph, MI), starch (heat-stable amylase and amyloglucosidase; AOAC International, 2000, method 996.11), sugar (AOAC International, 2000, method 968.28), crude fat (EE, using pet ether; AOAC International, 2000, method 920.39), minerals (using aquaregia digestion inductively coupled plasma atomic emission spectroscopy), and calculation of net energy (using NRC, 2001 equations).Nutrient composition of the sampled AMS concentrates and PMR were summarized by region (Supplemental Table S1).

DHI Milk Recording Data and Data Recorded by AMS
The DHI (Lactanet, Sainte-Anne-de-Bellevue, QC, Canada) data (collected as per: https: / / lactanet .ca/en/ robotic -services/ ) on milk fat and protein yields, milk yield, and milk fat and protein content (%) were available for 161 farms (8 in East, 21 in QC, 77 in ON, and 55 in West).Data were collected from all successful milking tests from April 1, until September 30, 2020 (6 to18 tests per farm; mean ± SD = 15.5 ± 0.9).Data were summarized by test day and by farm to calculate, for that time period: the daily mean milk, milk fat, and milk protein yield, as well as milk fat and protein content per cow.Herd-level milk fat and protein yields were calculated by averaging the product of the cow-level milk fat or protein concentration (%) and respective cow-level milk production (kg/d) data for each milk test day.Mean DIM and parity for each farm were also calculated from the DHI data, summarized first by test day level for each farm and then across the 18-mo study period for each farm.
Milking activity data were extracted from the AMS computer for Lely (Time-for-Cows, Lely Industries N.V, Maassluis, the Netherlands; n = 100) and DeLaval (Delpro, DeLaval International AB, Tumba, Sweden; n = 40) systems through the use of TeamViewer (Team-Viewer, Goppingen, Germany).Information about date, time, duration, AMS concentrate delivered, and outcome of each AMS visit (failure, refusal, or successful milking) were recorded for 154 farms (4 in East, 23 in QC, 74 in ON, and 53 in West) across the 18-mo study period.Data were summarized by cow for each day and then by farm for the observational period to determine the mean daily milking frequency (times/d), failures (times/d), refusals (times/d), visit duration (min/visit), milking duration (min/milking), milking interval (h), and AMS concentrate delivered (kg as fed/milking).Data from farms with GEA AMS were obtained online (FarmView, GEA Farm Technologies, Bonen, Germany; n = 14).The data retrieved for GEA already was summarized daily per farm, but was further summarized to a farm level for the study period.

Management and Housing Survey
During the farm visits in April to September 2019, trained research personnel conducted an interview with the farm manager or owner regarding overall farm characteristics, farm management, housing, and nutritional management, as outlined by Matson et al. (2021).A follow up survey was sent out to all farms to consider any changes made over the trial period; these surveys were collected from October 2020 to February 2021 (n = 162 respondents).Data involving feed management and basic herd demographics, as described by Matson et al. (2021), from the farm surveys were selected for  Lines with superscripts occur when there is a significant regional effect and different superscripts in the same line indicate significant differences (P < 0.05) between the 4 regions within PMR or AMS concentrate (pairwise comparisons with Tukey adjustment). 1 This P-value represents the regional effect for either the PMR or the AMS concentrate (AMS).
2 Highest standard error across regions. 3 The % of crude fat in the PMR and AMS concentrated offered from the formulated rations.
inclusion in the present study.In short, the number of AMS units per farm, breed of cow, cow traffic system, frequency of mixing PMR (times/d), frequency of PMR delivery (times/d), and frequency of PMR pushup (times/d) were recorded.Breed was categorized for each herd as either Holstein (>90% of the total milking cows were Holstein) or non-Holstein herds.Cow traffic system was categorized into free (n = 146) or guided flow traffic systems, where guided systems included 6 milk first, 4 feed first, and 6 modified guided (guided cow traffic system with lying stalls included within the feed alley area) flow systems.

Statistical Analysis
Statistical analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC) with farm as the experimental unit for all models and all data averaged at the farm level.Before analyses, all data were screened for normality using the UNIVARIATE procedure in SAS.All data were determined to be normally distributed.The nutrient compositions of the formulated and sampled diets were analyzed using a mixed-effects model using the MIXED procedure in SAS to test for regional differences.The fixed effect of region was tested for all the nutrient components of the formulated and sampled diets; when significant, the differences were investigated using the PDIFF procedure in the LSMEANS statement, with Tukey-Kramer adjustment.
The MIXED procedure in SAS was used to model the association of various explanatory variables (herd demographics, nutrient and ingredient composition of formulated diets, and feed management) with milk yield, milk fat yield, milk protein yield, milk fat concentration, milk protein concentration, milking frequency, and milking refusals per cow.Possible explanatory variables were included as continuous (parity, DIM, frequency of mixing fresh feed, frequency of feed delivery, frequency of feed push-up, expected AMS concentrate DMI, expected PMR DMI, total DMI, % forage of the PMR, % forage of the total ration, ratio of PMR-to-AMS concentrate, and the nutrient composition of the formulated diets) or categorical (breed, region, cow traffic system, major forage in the PMR, and major ingredient in the AMS concentrate) variables based on their distributions.The nutrient composition from the formulated diets used as potential explanatory variables included the dry matter (DM, %), crude protein (CP, %), ADF (%), NFC (%), sugar (%), starch (%), NE L (Mcal/kg DM), and crude fat (EE, %) for the AMS concentrate, the PMR, and the total diet (DM basis).Possible explanatory variables were first screened in univariable models, where only variables with P < 0.25 in the univariable models were retained to be tested in the multivariable models (Dohoo et al., 2009).The CORR This P-value represents the regional effect for either the PMR or the AMS concentrate (AMS). 3 The % of crude fat in the total ration offered from the formulated diets.
procedure in SAS was used to examine correlations between continuous variables before inclusion in and running of multivariable models.Similarly, for categorical variables, the FREQ procedure in SAS was used to test for associations between variables before running the multivariable models.For correlated variables (r >0.6), the most biologically plausible variable was considered for the multivariable model.In multivariable models that included breed and any DMI variables, the DMI variable was removed due to its correlation with breed.
Similarly, if a nutrient component was significant in the univariable model for AMS concentrate, PMR, and the total diet; only the total diet nutrient component was offered to the multivariable model.Backward elimination was used to remove any remaining explanatory variables in the multivariable model one at a time starting with the greatest P-value, that was P > 0.1, until all remaining variables in the model had a P ≤ 0.1.Two-way interactions that were biologically plausible were examined in the remaining model.Confounding variables (nonintervening variables whose removal resulted in a > 25% change in the coefficients of retained variables in the final model) would have been kept in the model; however, none were observed.For all multivariable final models, when significant categorical variables were retained, the PDIFF procedure in the LSMEANS statement, with Tukey-Kramer adjustment, was used to investigate the differences.Due to the interrelationship between nutrition, milk production, and milking behaviors, separate mixedeffects univariable models were run to detect associations of milking behaviors with milk, milk fat, and milk protein yield as well as milk fat and milk protein content.Only those significant univariable models are presented.Statistical significance was denoted at P ≤ 0.05 and tendencies were declared as 0.05 < P ≤ 0.10 for all models.

Herd Demographics
Similar to other farm-level AMS studies in Canada (King et al., 2016) and USA (Siewert et al., 2018), herds in this observational study averaged 126 ± 117 lactating cows, milked on 2.7 ± 2.2 AMS units per farm (Table 1).This resulted in 47.1 lactating cows being milked per AMS unit (Table 1) which was similar to King et al. (2016;49.4(Jacobs and Siegford, 2012).Similar to previous Canadian AMS herd studies (Deming et al., 2013;King et al., 2016), farms in this study had a mean parity of 2.4 ± 0.3 and averaged 170.7 ± 12.8 DIM for their lactating cows (Table 1).

Formulated Diets and Feed Samples
Based on the diet formulations received from the study farms' nutritionists, we observed that 60% of the study farms nationally utilized corn silage as their main source of forage in their PMR(Table 2).Different types of haylage were the second most common major forage source.Most of the farms utilizing haylage as their major forage source resided in the East, ON, and QC.However, the West's second most common major forage was barley silage (Table 2).Three farms in total (1.9% nationally), used hay as their major forage source in their PMR (Table 2).The differences observed in forage type used on farm can be attributed to regional and environmental factors, such as the shorter growing season and less heat units in portions of the West, which are more suited for growing barley silage.Thus, farms across Canada may be limited in their major forage type based on what is optimal for their regional growing conditions.
The primary concentrate ingredient used in the formulated PMR was corn (Table 2).Similar to the barley silage distribution, 14.4% of farms nationally had barley grain as their primary concentrate source, but of those farms 91% were in the West.
Within the PMR, 36 farms did not include enough information to determine the major FA content of the fat supplement in the PMR.The majority (61%) of farms fed palm fat (Table 2).One limitation of this study was lacking the ability to determine the concentration of individual FA, as various products were used with varying, undisclosed FA profiles.A limitation of classifying farms based on summarized, formulated diets is some farm's major forage, grain, or fat supplement may not remain the greatest inclusion during the whole observational period.
A large majority of the farms on this study feed pelleted concentrate at the robot with the remaining farms feeding a texturized feed (Table 3).Nationally, 45.2% of herds incorporated wheat (primarily wheat shorts) as their major ingredient in their AMS concentrate formulation, which varied by region (Table 3).The second most common major AMS concentrate ingredient was corn.Barley grain was fed as the primary ingredient in their AMS concentrate on 12.9% of farms; all but one of those were in the West (1 in East).It has been previously observed that cows have increased visits to the AMS, increased intake, and less fetching required when fed wheat or a barley and oat mix based AMS concentrate compared with corn (Madsen et al., 2010).Crop availability and price may be a more prominent factor in the decision for ingredient choice rather than cow preference.Alternatively, pelleting barley compared with steam flaked has been shown to increase voluntary AMS visits, decrease holding pen time, increase box time, and shorten the milking interval without affecting milk production or composition in a feed-first AMS (Johnson et al., 2022).
Soybean meal was the most common major AMS protein source nationally with canola meal and dry distillers' grains (Table 3) as the next most common.Researchers previously observed that dairy cows had a greater preference with canola meal over soybean meal (Bertilsson et al., 1994;Sporndly and Asberg, 2006), resulting in increased rate of consumption and less orts.This could be one plausible explanation for some farms choosing canola meal as their major protein source; however, regional availability is most likely a contributing factor as well, as most of the farms feeding canola meal as their major protein source were from the West.The remaining farms fed wheat middlings, corn gluten feed, or peas as the major source of protein of their AMS concentrate.
Based on the formulated diets, farms in the East and ON fed for the greatest PMR DMI, and the least amount of AMS concentrate.Conversely, farms in the West and QC formulated for lower PMR DMI and higher DMI at the AMS (Table 4).No regional differences were observed for total formulated DMI (Table 5).Researchers have previously demonstrated that if AMS concentrate allowance is altered, cows adjust their PMR (Hare et al., 2018;Schwanke et al., 2019).This adjustment in PMR DMI has been termed a substitution effect (Bach et al., 2007) and ranges in substitution have been observed anywhere from 0.54 to 1.58 kg/d reduction in PMR DMI for every 1 kg-increase in AMS concentrate consumed (Hare et al., 2018;Schwanke et al., 2022).The average predicted AMS concentrate DMI, based on the formulated diet, for all regions was 4.3 kg/d (Table 4).Previous observational studies have reported AMS concentrate allowances of 5.01 (Siewert et al., 2018) and ~5.43 kg/cow/d (Salfer et al., 2018).Researchers have previously observed that cows consume pellets at a rate of 250 to 400 g/min (Kertz et al., 1981) during the first 4 to 8 min of a feeding event, that milking frequency averages 2.2 to 2.8 milkings/d (Bach et al., 2009;Deming et al., 2013), and that each milking lasting approximately 7 min (Castro et al., 2012).This results in approximately 3.85 to 7.84 kg/d (as fed) of AMS concentrate that a cow can theoretically consume each day.
To our knowledge, no previous herd-level observational study has reported the nutrient composition of the formulated PMR or AMS concentrate.Diets in the East were formulated to contain the lowest ADF, ash, and K in their PMR, while farms in QC had PMR diets formulated to contain, on average, the highest CP, ADF, Ca, and Cl content and the lowest starch, crude fat (EE), and NE L (Table 4).Formulated PMR for ON herds had the highest predicted NFC and starch levels, while containing the lowest average CP, sugar, and Cl content (Table 4).PMR diets on West farms had the highest formulated average sugar, EE, ash, and NE L levels, with the lowest NFC and Ca levels (Table 4).No regional differences were detected in the formulated PMR for the formulated DM, NDF, P, Mg, or Na content, with the national averages recorded in Table 4.
The sampled and analyzed PMR (see Supplemental Table S1) was quite similar to the formulated PMR nationally (Table 4), with the exception of sampled PMR having a higher % DM and lower starch, sugar, and EE content.Numerical variation between sampled and formulated diets within region were observed; however, differences between regions were consistent.Differences could be expected, as samples were taken at one time point during the study period and, thus, we could not account for daily, seasonal, ingredient, and other changes on farm that may not be reflected in the formulated diets (St-Pierre and Weiss, 2015).
No regional differences in the formulated AMS concentrates were detected for DM, ADF, NDF, ash, or NE L levels (Table 4).East farms fed formulated AMS concentrates with the highest NFC and starch, with the lowest CP content.Formulated AMS concentrates in QC contained the highest predicted EE, but the lowest starch and sugar content.Farms in ON had higher predicted CP regionally, while having the lowest NFC and EE in their formulated AMS concentrate.Compared with the formulated AMS concentrate national average (Table 4), the sampled concentrates (see Appendix 1) were higher in DM, NDF, sugar, ash, and NE L content, while having lower levels of starch and EE.Similar to the PMR, large numerical variations were observed between formulated and the sampled AMS concentrate; however, the between regional differences for individual nutrient components were consistent.While we would expect less variation between sampled and formulated AMS concentrate than PMR offered on farms, these data are based on a single sample from those farms, presenting a study limitation.
It should be noted that summarizing formulated diets from various nutritionists and nutrition companies adds potential error in how each nutrient component value was measured or calculated.In addition, nutrition companies may not use the same feed analysis methods or nutritional models, to estimate values such as NFC and NE L , resulting in values that may add bias to our results and need to be considered when drawing conclusions.Another limitation of these data is that they are based on formulated diets and not actual DMI of PMR and AMS concentrate.The diets submitted were formulated for the average cow and did not consider variation in the AMS allocation (Henriksen et al., 2019;Paddick et al., 2019) or day to-day mixing on farm.Furthermore, changes made on the farm, relative to that formulated, may not have always been recorded and would not be reflected in these data.Similarly, if the nutritionist left out any diet changes when reporting all formulated ration changes to the research team, this may also affect the potential bias observed in the data set.

Milk Yield
For the farms in our study, cows produced a mean of 37.0 kg/d of milk (Table 1), which is higher than previous North American studies in which the yield reported for herds milking, primarily Holstein cows with AMS, averaged 35.1 kg/d (Deming et al., 2013), 33.7 kg/d (King et al., 2016), 33.2 kg/d (Siewert et al., 2018), and 32.0 kg/d (Tremblay et al., 2016).Our model for factors associated with milk yield controlled for the effect of breed and herd average DIM, with Holstein farms producing +7.3 kg/d more milk than non-Holstein herds (Table 6).Across the study farms, the average DIM was 170.7 d (Table 1), and we observed that every 10-d increase in herd-average DIM was associated with −0.63 kg/d less milk per cow/d (Table 6).With both breed and average herd DIM accounted for, a greater formulated crude fat concentration in the PMR (PMR % EE) was associated with greater milk yield (Table 6).Nationally, the PMR % EE was formulated to 4.43% DM (Table 4).Each additional 0.1 percentage point (p.p.) increase in PMR % EE was associated with +0.1 kg/d milk yield (Table 6).Since PMR % EE was positively correlated with PMR NE L , PMR NE L was removed as a potential explanatory variable from this model.Generally, feeding higher levels of fat in the ration resulted in greater milk yield, as about half of the dietary fat is transferred to milk fat, and the rest supplies energy for the cow (Palmquist and Jenkins, 2017).However, the effect fat supplementation has on milk yield is dependent on source, FA profile, and the basal diet being fed (Palmquist and Jenkins, 2017).
The major forage type used on farm also was associated with milk yield (Table 6).Compared with farms feeding haylage as their primary source of forage, farms feeding barley silage had +2.2 kg/d greater milk yield, while farms feeding corn silage tended to have +1.2 kg/d greater milk yield.No differences were detected between the production levels of farms feeding barley or corn silage as their major forage source.Major forage source was confounded with region, as all the farms utilizing barley silage as their major forage were from the West.Thus, further information is needed to determine if another variable may be causing this regional increase in milk yield or if this response can be fully attributed to using barley silage.One possible confounding factor may be FA supplementation, since a higher proportion of West farms fed supplemental palm FA, which is a saturated FA that has less effects on DMI, while improving milk yield compared with unsaturated FA (Mosley et al., 2007).However, once we removed the unreported farms, ON and West both had 78% of farms feeding palm FA supplements (Table 2).Further, we were unable to determine the inclusion rate of these supplemented FA to determine if that was a possible confounding factor.Despite barley silage's lower digestibility compared with other forages, such as corn silage (Ahvenjarvi et al., 2006;Zhang et al., 2010), cereal grain forages may allow for greater DMI and have been shown to serve as a viable partial replacement for haylage without negatively affecting milk production (Phipps et al., 1995;Abdalla et al., 1999).Similar to the current study, Phipps et al. (1992) observed a positive association when substituting corn silage for haylage and milk production.

Milk Fat Yield and Content
Across the study farms, cows averaged a milk fat yield of 1.51 kg/d and a milk fat content of 4.09% (Table 1), which is similar to the Canadian national average of 4.07% (Agriculture and Agri-Food Canada, 2021).Breed and average herd DIM were also incorporated in the fat yield model to control for expected differences of +0.11 kg/d for Holstein cows compared with non-Holstein herds, as well as −0.01 kg/d for every 10-d increase in herd DIM (Table 6).Holsteins had 0.95-p.p. lower milk fat content than non-Holstein herds (Table 7).Parity was also controlled for in the milk fat content model, with each additional herd-average lactation associated with +0.02 p.p. milk fat content (Table 7).Controlling breed and DIM differences, greater PMR % EE was associated with greater milk fat yield.Every additional 1.0 p.p. increase in the PMR % EE was associated with +0.05 kg/d greater milk fat yield (Table 6).Many researchers have reported that greater fat supplementation results in increased milk fat yield and content (Palmquist et al., 1993;Mcnamara et al., 1995).
The % DM of both the AMS concentrate and the PMR (total % DM; Table 5) was also positively associated with milk fat yield (Table 6).For every 10-p.p. increase in total % DM, farms had +0.07 kg/d milk fat.Both % forage in the diet and the PMR-to-AMS concentrate ratio were not associated with milk fat yield, suggesting the higher % DM may be a pseudo indicator for another variable, such as forage quality (Bolsen and Bolsen, 2004).However, further research is needed to understand this detected association.Milk fat yield was negatively associated (Table 6) with the total concentration of CP (%) offered in the AMS concentrate and in the PMR combined (Table 5); each 1 p.p. increase in total % CP was associated with −0.04 kg/d milk fat yield (Table 6).While this may contradict what has been previously reported (Leonardi et al., 2003;Colmenero and Broderick, 2006), the difference is very small and likely not biologically relevant.Regardless, the observed differences may result from a nutrient imbalance or forage quality differences (Zimmerman et al., 1991;Dinn et al., 1998).
When controlling for the differences in breed and parity, milk fat content was positively associated (Table 7) with the PMR-to-AMS concentrate ratio and the total NE L of the AMS concentrate and PMR (total NE L ; Table 5), while negatively associated (Table 7) with the NFC concentration of the PMR (PMR % NFC; Table 4).Each 2-unit increase in the PMR-to-AMS concentrate ratio was associated with +0.02 p.p. milk fat content (Table 7).Researchers previously observed milk fat content to be lower with greater AMS concentrate allowance (Miron et al., 2004b;Menajovsky et al., 2018).Feeding higher levels of forage has been reported to improve milk fat content, as result of greater buffering of the rumen with the added effective fiber (Kalscheur et al., 1997).Since the forages are within the PMR portion of the ration, increasing the amount of forage in the total diet by increasing the amount of PMR fed is expected to increase the PMR-to-AMS concentrate ratio.
Each 10-p.p. increase in PMR % NFC was associated with −0.2 p.p. milk fat content, while each 0.1 Mcal/ kg of DM increase in total NE L was associated with a +0.05 p.p. milk fat content (Table 7).This suggests that increasing the energy of the total formulated diet, without increasing the NFC content, would improve milk fat content.In support, reducing the NFC content of the diet by increasing % CP or EE of the diet resulted in greater milk fat production (Palmquist et al., 1993;Leonardi et al., 2003;Colmenero and Broderick, 2006).Interestingly, greater NFC, through the replacement of readily available carbohydrate sources with wheat middlings, dried brewer's grains, and soy hulls, resulted in greater milk fat content (Batajoo and Shaver, 1994).

Milk Protein Yield and Content
The cows in our study farms averaged a milk protein yield of 1.25 kg/d and a milk protein content of 3.38% (Table 1), similar to the national average of 3.33% (Agriculture and Agri-Food Canada, 2021).Similar to the milk fat models, Holstein herds had +0.2 kg/d milk protein yield (Table 6), with −0.3-p.p. milk protein content compared with the non-Holstein herds (Table 7).Each 10-d increase in herd-average DIM was associ- Non-Holstein herds included all herds with <90% Holsteins.These included 1 Ayrshire, 3 mixed Ayrshire, 1 Brown Swiss, 7 Jersey, 4 mixed Jerseys, and 1 mixed-breed herd. 3 The % of crude fat in the PMR offered from the formulated diets. 4 The % of DM that was from forage in the PMR offered from the formulated diets. 5 The Mcal/kg of energy in the PMR offered from the formulated diets. 6 The total % of dry matter in the PMR and the AMS concentrate offered from the formulated diets. 7 The total % of crude protein in the PMR and the AMS concentrate offered from the formulated diets.
Greater milk protein yield was positively associated (Table 6) with increased PMR NE L (Table 4) and total % DM while being negatively associated (Table 6) with the percent of forage in the PMR (PMR % forage).Every 10-p.p. increase in forage in the PMR was associated with −0.03 kg/d and every 10-p.p. increase in total % DM was associated with +0.05 kg/d milk protein yield (Table 6).Additionally, each increase of 0.1 Mcal/kg DM in the PMR NE L was associated with +0.02 kg/d milk protein yield (Table 6).
The DM content of the total diet was negatively associated with the % forage in the PMR.Thus, increasing the energy density of the diet by increasing total diet % DM and PMR NE L , while decreasing PMR % forage, could result in greater milk protein yield (Emery, 1978;Cragle et al., 1986) by increasing energy for microbial protein production (Clark et al., 1992).Similarly, reducing the forage content of the diet by increasing concentrate increases the propionic acid production in the rumen (Hernandez-Urdaneta et al., 1976), thus supplying gluconeogenic precursors in the liver and potentially sparing AA for milk protein production in the mammary gland that would otherwise be directed toward hepatic gluconeogenesis (Huhtanen et al., 1998).
Milk protein content was negatively associated (Table 7) with the major forage type used on farm while being positively associated (Table 7) with the PMR % forage and formulated total starch content in the AMS concentrate and PMR (total % starch; Table 5).Compared with farms feeding haylage as their primary source of forage, farms feeding corn silage had −0.10 p.p. milk protein content, while no difference was observed between farms feeding haylage or barley silage (Table 7).Each 10-p.p. increase in PMR % forage was associated with +0.05 p.p. milk protein content (Table 7).The association of both forage type and PMR % forage to protein content can be explained by their associations with milk yield.As previously mentioned, farms with corn silage as their major forage had greater milk yield than farms feeding haylage.Similarly, as the PMR % forage increases, diets would become less nutrient-dense, supply lower levels of NE L, and result in reduced milk yield leading to an elevated milk protein content (Emery, 1978;Cragle et al., 1986).Each 5-p.p. increase in total % starch was associated with +0.03 p.p. milk protein content (Table 7).Increasing dietary Non-Holstein herds included all herds with <90% Holsteins.These included 1 Ayrshire, 3 mixed Ayrshire, 1 Brown Swiss, 7 Jersey, 4 mixed Jerseys, and 1 mixed-breed herd. 3 The ratio of PMR expected DMI to the expected DMI of the AMS concentrate offered from the formulated diets. 4 The % of NFC in the PMR offered from the formulated diets. 5 The % of DM that was from forage in the PMR offered from the formulated diets. 6 The total % starch in the PMR and the AMS concentrate offered from the formulated diets. 7 The total amount of NE L (Mcal/kg DM) in the PMR and the AMS concentrate offered from the formulated diets.
starch content could potentially increase glucose supply, either indirectly by increasing propionate (a major gluconeogenic precursor) absorption from the rumen or directly by increasing glucose absorption when starch is digested in the small intestine.As discussed previously, a greater glucose supply could spare circulating AA from being used for gluconeogenesis in the liver, thus increasing the AA supply to the mammary gland to be used as building blocks for milk protein synthesis (Huhtanen et al., 1998).
The final model for milking frequency included the effects of the flow of cow traffic system used on the farm.Farms utilizing a free-flow cow traffic system (n = 145) had +0.6 milkings/d than farms with guided flow traffic systems (n = 16).Some older, controlled studies reported greater milking frequencies in guided traffic systems (Ketelaar-de Lauwere et al., 1998;Ketelaar-de Lauwere et al., 2000;Bach et al., 2009).Using a subset of herds in Ontario, from the current study, Matson et al. (2022) observed +0.37 milkings/d for free-flow compared with guided flow traffic systems.Other researchers have recorded high levels of average milking frequencies (#/d) in controlled and field studies of 3.2 (Munksgaard et al., 2011), 3.0 (Madsen et al., 2010), 2.7 (Castro et al., 2012), and 2.8 (Deming et al., 2013) for free-flow traffic systems.It should be noted that milking permission settings were not able to be recorded and, thus, considered a limitation of the study.These settings allow or prevent cows from accessing the milking robot more frequently and, thus, would effect the milking frequency on farm.
Milking frequency was also positively associated (Table 8) with feed push-up frequency (Table 1), while being negatively associated with PMR % NFC and total ration % forage (Table 8).Every 5 additional feed push-ups were associated with +0.67 milkings/cow/d (Table 8).Increasing feed availability, through pushing up feed, allows AMS cows to optimize their time for feeding (Deming et al., 2013) rather than searching for feed.Matson et al. (2021) previously reported a posi-tive association between herd-average milk yield and feed push-up frequency.
Each 1-p.p. increase in PMR % NFC was associated with −0.017 milkings/d (Table 8).Schwanke et al. (2019) demonstrated that increasing the NFC of the PMR, by feeding increased concentrate in the PMR, reduced the voluntary AMS visits and milking frequency compared with a higher forage PMR.A reduction in milking frequency may be explained by cows having greater motivation to feed than to milk (Prescott et al., 1998) and by increasing the NFC at the PMR this reduces the motivation to enter the AMS unit to receive the supplemental AMS concentrate.Farms in the current study formulated their total ration to have 58.2 ± 6.63% DM as forage between their AMS concentrate and their PMR (ranging from 41.8 to 74.1% DM as forage; Table 1).Each 10-p.p. increase in total ration % forage tended to be associated with −0.12 milkings/d (Table 8).As demonstrated by Schwanke et al. (2019), increasing the energy density of the PMR by reducing the forage concentration resulted in greater PMR DMI.Another way to reduce the % forage of the total ration would be to increase the AMS concentrate allowance.Regardless, increasing the nutrient density of the diet provides the nutrients for increased milk production (Rabelo et al., 2003;Nielsen et al., 2007), which may be associated with greater milking frequency at the AMS (Tremblay et al., 2016).
The final model for refused visits to the AMS included both parity and breed, where each 1-lactation increase in herd-average parity was associated with +0.9 times/d more refused visits, while Holstein herds had −0.9 times/d fewer refused visits (Table 8).A cow's increase in parity is usually associated with greater milk yield (King et al., 2017), thus greater visits to the AMS (Tremblay et al., 2016).Alternatively, Holsteins have a greater risk of lameness compared with non-Holstein farms (Matson et al., 2022), potentially as a result of breed differences in BCS and foot physiology (Baird et al., 2009;Dippel et al., 2009), and this may result in fewer refused visits (King et al., 2017).
Accounting for the differences of breed and parity, farms with free-flow cow traffic systems had +0.8 milking refusals/d at the AMS compared with guided flow (Table 8).The predetermined restricted flow of guided flow systems may deter excessive refused visits compared with free-flow systems, where cows could re-enter immediately.In addition, selection gates used in some guided flow farms may divert cows to the feeding alley or resting areas without accessing the AMS, thus preventing a recorded refusal.While not recorded in this study, the cow flow traffic system may confound with herd-programed milking allowance settings.Similar to the milking frequency, the amount of refused milkings is also affected by these programmed milking permissions.The amount of concentrate offered at the AMS has previously been associated with increased refusals (Tremblay et al., 2016), which could be an alternative confounding factor.Due to the difference in breed observed, all DMI variables were removed from the refused milking multivariable model, as Holsteins typically consume more feed than other breeds (Kristensen et al., 2015).In the current study, guided flow herds formulated for −0.93 kg/d DM AMS concentrate than free-flow herds on the study (3.44 ± 0.48 kg/d DM and 4.37 ± 1.11 kg/d DM, respectively), while having no difference in PMR (20.9 ± 1.1 kg/d DM and 21.2 ± 1.3 kg/d DM, respectively) or total formulated total DMI (25.5 ± 1.7 kg/d DM and 25.4 ± 1.9 kg/d DM, respectively).Lastly, farms that fed barley silage had the greatest refused AMS visits.Farms feeding barley silage had +0.58 refusals/d than farms feeding haylage as their major forage source (Table 8).As previously discussed, barley silage was fed exclusively in the West.This finding could be confounded with other nonnutritional factors, such as individual AMS settings and other farm-level management practices common on these farms in the West that feed barley silage.

Associations of Milking Behaviors and Production
Since nutrition, milking behavior, and milk production can all influence each other, milking behaviors were left out of the previously discussed multivariable models, which accounted for the various measured nutritional and management factors.We observed a positive association between milking frequency and milk, fat, and protein yields (Table 9).Each additional successful milking/d was associated with +3.4 kg/d milk, +0.15 kg/d fat, and +0.11 kg/d protein yield (Table 9).Milking frequency is known to be associated with greater production (Erdman and Varner, 1995;Smith et al., 2002;Siewert et al., 2018).Milking refusals were negatively associated with milk yield and positively associated with milk fat and protein content (Table 9).Each additional milking refusal per day tended to be associated with −0.79 kg/d milk yield, +0.11 p.p. milk fat content, and +0.05 p.p. milk protein content (Table 9).While maintaining refused AMS visits above 1/d is suggested (Kozlowska et al., 2013), excessive milking refusals can negatively affect the cow's ensuing behavior resulting in less DMI and lying down compared with a successful milking (Stefanowska et al., 2000).The reduced milk yield observed from increasing refused milkings, would then result in greater milk fat and protein content, as it becomes less diluted.Alternatively, AMS settings complicate this relationship; unfortunately  individual herd and cow settings were not able to be recorded.
Across study farms, cows had average milking durations of 4.7 ± 0.47 min/milking and a milking visit duration of 7.1 ± 0.59 min/visit (Table 1).Greater milking duration was negatively associated with milk fat content, while milking visit duration was negatively associated with both milk fat and protein content.Each 1-min increase in milking duration was associated with −0.12 p.p. milk fat content (Table 9).Similarly, each 1-min increase in visit duration was associated with −0.12 p.p. milk fat content and −0.06 p.p. milk protein content (Table 9).Tremblay et al. (2016) observed that increasing the time cows spent in the AMS during each visit (boxtime) was associated with greater milk production.While we did not observe an association between milking or visit duration with milk yield, it can be assumed the milk fat and protein content would be diluted as result of greater milk yield.

CONCLUSION
In this study we described the nutritional strategies, including the formulated diets and the major ingredients that Canadian AMS dairy herds implemented in their AMS concentrates and PMR, and explored their association with milk yield and milk components as well as milking behaviors.Nutritional and farm management factors, including dietary crude fat, forage % of the diet, energy density, cow traffic flow, and major forage fed on the farm, were associated with greater milking frequency and milk production.Greater milking frequency was associated with greater milk, fat, and protein yields on these farms milking with AMS.Overall, benchmarking the diets formulated and nutritional strategies used on AMS farms, as done in this study, can help us understand how diet variability may be associated with milk production and milking behavior, and how this information can be used to improve diet formulation for AMS herds.
targeted a sample size of 200 herds, 177 were included in the present study, which would allow the Van Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS detection of a 7% difference in the response variables between categorical predictors.Producers' consent was obtained before farms being initially visited between April 2019 and September 2019.

Van
Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS

Van
Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS

Van
Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS

Van
Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS

Van
Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS

Table 1 .
Van Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS Descriptive statistics of the demographics and feeding management of study herds using automated milking systems (AMS) 2ECM is 4% Energy corrected milk yield. 3

Table 2 .
Van Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS The major forage, grain, and fatty acid supplement used by each study herd in their partially mixed ration (PMR) as recorded by their nutritionist (summarized by region)

Table 3 .
Concentrate type, major ingredients, and protein sources in the concentrate offered at the automated milking system (AMS) as recorded by nutritionist on each study farm (summarized by region)

Table 4 .
Regional differences in nutritional feeding strategies of the partially mixed ration (PMR) and the concentrate offered at the automated milking system (AMS) as reported from formulated diets submitted by study farm nutritionists (mean ± SE)

Table 5 .
Van Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS Regional differences in nutritional feeding strategies of the total diet as reported from formulated diets submitted by study farm nutritionists (mean ± SE) a-c Lines with superscripts occur when there is a significant regional effect and different superscripts in the same line indicate significant differences (P < 0.05) between the 4 regions within PMR or AMS concentrate (pairwise comparisons with Tukey adjustment).1 Highest standard error across regions. 2 cows/AMS), while being lower than Deming et al. (2013; 55 cows/AMS), Siewert et al. (2018; 56 cows/AMS), and the suggested capacity of 60 cows/AMS unit

Table 6 .
Van Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS Multivariable linear model of the factors associated with milk, milk fat, and milk protein yields (kg/d) of cows in study herds using automated milking systems (AMS) 1 Estimated regression coefficient.2

Table 7 .
Van Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS Multivariable linear model of the factors associated with the milk fat content (%) of cows in study herds using automated milking systems (AMS; n = 100 herds) 2 Van Soest et al.: NUTRITION AND MILK PRODUCTION IN AUTOMATED MILKING HERDS

Table 8 .
Multivariable linear model of the factors associated with the frequency of successful milkings and refused milkings (times/d) of cows in study herds using automated milking systems (AMS)

Table 9 .
Univariable linear models of the milking behavior factors associated with the milk, milk fat, and milk protein yield (kg/d) as well as milk fat and protein content (%) of cows in study herds using automated milking systems (AMS) 2 Estimated regression coefficient.