A narrative review of factors affecting the welfare of dairy cows in larger Australasian pasture-based production systems

The extent to which cows in intensive grazing systems experience feelings of hunger is unclear. Intensively grazed dairy cows are typically provided fresh pasture twice per day following a morning and an afternoon milking. High intake rates are observed under this management with most of the available pasture being consumed soon after cows gain access to the new allocation (Chilibroste et al. 2007; Gregorini et al. 2017). One interpretation of these high intake rates is that cows are hungry. Hunger is an important subjective state, but the welfare implications of short-term fluctuations in hunger are unclear. Indeed, these fluctuations are essential in terms of motivating feeding behaviour (Tolkamp and D’Eath 2016). We posit that appetitive hunger associated with periods of fasting throughout the day is unlikely to present a meaningful risk to animal welfare.

Long-term fluctuations to a cow’s nutritional status may be more relevant to welfare. There is less control over individual-cow nutrition in pasture-based than in indoor dairy systems, where cows are fed a carefully balanced total mixed ration (TMR) diet. For example, in very large herds (>700 animals), cows spend up to 4 h out of the paddock at each milking (Beggs et al. 2015). Long milking times do not appear to have significant effects on lying times (Beggs et al. 2018a). However, that a cow’s position in the milking order is mostly consistent means that some animals will consistently be away from the paddock for several hours longer than others (Beggs et al. 2018b) and this can have implications for access to pasture (quantity and quality e.g. Dias et al. 2019). Research has shown that pastured cows that are milked first produce nearly 20% more milk per day than those that are milked last (Dias et al. 2019) and modelling reports milk yield losses of 0.61 kg and 0.20 kg for every 1 km increase in total walking distance and every 1 h increase in milking interval respectively (Islam et al. 2015).

In addition to difficulties in optimising individual-cow intake in pasture systems, the feed base is subject to seasonal shortages and surpluses, the extent of which depends on the environmental conditions of the time. As stated by Chilibroste et al. (2007), ‘it is not uncommon that available pasture allowances are insufficient to fully meet the nutrient requirements of the cows due to limited availability, low nutrient density and/or high nutrient requirements of the cows’ (p. 1079). Dairy cows also have preferences; they will avoid grazing areas near urine or faecal deposits, leaving patches (or clumps) of ungrazed pasture. The dairy industry in New Zealand, which focusses on a pasture-only diet (i.e. no supplementary concentrate), recommends that these clumps of grass have a ‘sharp shape’ after an efficient grazing with the tops and sides of clumps being well eaten into (Dairy NZ 2023). By contrast, Dairy Australia says that the right balance between pasture and supplementary concentrate is evidenced by a smooth curve up to the clump and that the sides or top of the clumps should not be eaten (Dairy Australia 2016). Assessing how hard cows graze into clumps of unpreferred pasture could potentially be used to assess cow hunger in alternative grazing systems, but requires validation.

The drive to maximise pasture production per hectare is associated with high stocking rates and a more intensive utilisation of homegrown feed (Stafford and Gregory 2008). A summary of the literature on the relationship between stocking rate and cow productivity can be seen in Table 1. If the amount of pasture grown does not change, pasture utilisation and milk production per hectare of land increases with stocking rate, but milk production and pasture utilisation per cow declines. These reductions are associated with reductions in lactation length, liveweight, and body condition in most studies (Table 1). The studies by Valentine et al. (2009), Baudracco et al. (2011) and Patton et al. (2016) were able to mitigate the effects of high stocking rates on cow production by importing more supplementary feed (Table 1). High stocking rates are generally supported by increased farming inputs, such as nitrogen fertilisation and water to achieve high pasture growth and concentrate feed to supplement cow energy intake. However there are physiological limits to how much forage a cow can consume in a day, resulting in periods where their metabolic demand exceed intake capacity such as in early lactation (Stockdale 2001).

Body condition score (BCS) is an accepted measure of energy reserves (Fisher 2020; Mee and Boyle 2020), and cows in pasture-based dairy systems are at a greater risk of low BCS than are more intensively fed cattle (Mee and Boyle 2020). The dairy industry has guidelines for target body condition at different stages of lactation (e.g. Dairy NZ 2012; Dairy Australia 2015). These are based on achieving optimal production and may not be a sufficiently sensitive reflection of cow welfare (Roche et al. 2013). As discussed by Fisher (2020), animals at a particular BCS could be maintaining, gaining, or losing condition with different implications for what they feel (e.g. hunger, malaise) and are likely to experience (e.g. metabolic diseases). Mellor (2017) suggested using both base-level and changes in body condition when inferring hunger during animal welfare assessments. For example, an animal with a mid-level but stable body condition is interpreted to have a low-level of hunger, whereas an animal with mid-level and declining body condition is interpreted to have a moderate level of hunger (Mellor 2017). Examining industry BCS targets through the lens of Mellor (2017) may provide an understanding of whether dairy cows are experiencing hunger. Dairy Australia (2015) recommends a BCS target of 4.5–5.5 at calving (on an 8-point scale), with a BCS decrease of 0.6 points or less from calving to mating. According to Mellor (2017), this target suggests that early lactation cows are, on average, moderately hungry. After early lactation, Dairy Australia (2015) recommends that cows maintain or gain body condition. Thus, during this period cows should have a low to no or very low levels of hunger (Mellor 2017).

It is not clear whether Mellor’s (2017) guide for the interpretation of body condition and its implications for hunger is applicable to animals in states of high physiological demand, such as early lactation. A slow rate of bodyweight loss in early lactation is natural (see review by Lederman 2004), and even cows fed a TMR ad libitum lose some condition in early lactation (e.g. Weber et al. 2013). Minimising the extent of loss in body condition post-calving may be more important to animal welfare than is preventing a loss from occurring. The question is whether the industry targets fall within a ‘normal’ rate of body condition loss for a dairy cow in a pasture-based system, or whether they are simply meeting a minimum physiological threshold that prevents disruption of reproduction. If the former, it is reasonable to conclude that cows achieving industry BCS targets are fully or close to fully fed in early lactation, but if the latter they may be experiencing hunger during this period. Cow reproduction is compromised with increasing severity and duration of the postpartum negative energy balance, but the effect of this relationship is biologically small (a 0.25-unit increase in nadir BCS is predicted to increase rate of pregnancy to first service by ~1%; Roche et al. 2007). It may be more helpful to consider the proportion of dairy producers successfully achieving industry targets. In New Zealand, Roche et al. (2007) found nearly 1/4 of cows to be thin at calving (BCS < 4 on a 10-point scale), but the average loss in BCS between calving and the nadir was 0.73 points, which is under the industries recommended threshold (≤1-point loss in BCS between calving and mating; Dairy NZ 2012).

As well as having enough food, it is important that the diet is appropriate and well balanced with the requirements of the cows. Grain-based concentrate is making up an increasing proportion of the cow’s diet in modern pasture-based systems. The increasing levels of concentrate being fed is partly attributed to intensification and increased climatic variability (Wales et al. 2013; Williams et al. 2020), although some dairy farms rely on supplementary feed as management tool to increase dry-matter intake and thus milk production (reviewed by Hills et al. 2015).

Concentrates that are pulse-fed to cattle during milking may increase the risk of diseases secondary to overfeeding of grain, with associated effects on feelings of pain, nausea, and malaise. The ingestion of a large quantity of concentrate with a high starch and low non-digestible fibre content in a short amount of time results in rapid ruminal fermentation and produces high levels of volatile fatty acids, causing the pH of the rumen to rapidly drop (Lean et al. 2008). Highly digestible pastures and winter crops, with high concentrations of rapidly fermentable carbohydrates and less effective fibre are also implicated in low ruminal pH (Bramley et al. 2005; O’Grady et al. 2008; Peyraud and Delagarde 2013). Subclinical ruminal acidosis (generally defined in the scientific literature as a pH between 5.5 and 6; abbreviated to ‘SARA’) is associated with diarrhoea, appetite suppression and behaviours indicative of stomach pain (Rushen et al. 2007; Plaizier et al. 2008; Beever and Bach 2016). Acute ruminal acidosis (pH below 5.5) can result in severe metabolic acidosis, inflammation and abscessation throughout the body, recumbency, coma and death (reviewed by Kleen et al. 2003 and discussed by Bramley et al. 2005). The point-prevalence of SARA in cows surveyed on 100 Australian dairy systems in 2008 was ~10% (average pH 5.7; Bramley et al. 2008). A similar point-prevalence survey of 12 Irish farms recorded 11% of cows with a pH of ≤5.5 (i.e. acidotic) and 42% with a pH between 5.6 and 5.8 (i.e. SARA; O’Grady et al. 2008). The dairy industry in Australia feeds more grain now than in 2008 (Dairy Australia 2019). More recent reports on the prevalence of acidosis in pasture-based systems are needed.

The maximum level of concentrate fed in the dairy increases from 5.2 kg/cow in small herds (<300 cows) to 10.1 kg/cow in large herds (>751 cows; Beggs et al. 2019a). The risk of ruminal acidosis increases when a grazed pasture diet is supplemented with high levels of concentrate fed in the dairy (Auldist et al. 2013; Golder et al. 2014; Wright et al. 2014), although the high levels of concentrate being fed per cow in these studies exceeds the maximum 10 kg/day reported by Beggs et al. (2019a). Indeed, Beggs et al. (2019a) found no effect of herd size on the prevalence of acidosis in Australian dairy farms assessed via the consistency of faecal pats and dairy farm records. Controlling ruminal fermentation is essential to reducing the risk of acidosis (Lean et al. 2008). One way of doing this is to gradually increase the level of grain fed during the transition period to prepare the rumen for the high levels of grain fed in early lactation. Another way is to increase crude-fibre intake during early spring and autumn periods (e.g. a long stem fodder, see review by Rivero and Anrique 2015). Farms that feed high levels of concentrate may also use feed additives that help attenuate SARA (e.g. yeast products, phytogenic compounds, buffering substances), although these may not fully compensate for appropriate feeding management (reviewed by Humer et al. 2018).

There is an emerging idea that the lack of diversity in the grazing cows’ diet has negative implications for animal welfare. For example, Villalba and Manteca (2019) suggested that diet building is a natural cognitive enrichment that fosters a positive affective state by relieving boredom. Cows in pasture-based dairy systems predominantly graze a monoculture or a simple grass–legume mix that is at a uniform state of development. These plants are typically chosen because of their high nutritive value and long growing season under temperate conditions. Some advocate a more holistic approach to manage grazing on the basis of the provision of diverse diets to increase resilience, improve nutrition, and, potentially, reduce environmental impacts (Gregorini et al. 2017). Although mixed swards have been studied in relation to their persistence and effects on animal production (e.g., Cranston et al. 2015; Jerrentrup et al. 2020), understanding of the relationship between diverse pastures and animal welfare remains in its infancy and requires further exploration.

A 582 kg lactating dairy cow fed a TMR and producing 28 kg of milk/day will drink ~75 L/day of water, whereas cows at pasture drink about half this because of the high water content in grass (see review by Jensen and Vestergaard 2021). Little is known about the condition of water provided to grazing dairy cows. Australian and New Zealand cows are provided water in their paddocks (Beggs et al. 2019a), but there is a question as to whether increasing herd sizes result in cattle spending longer periods away from the paddock and, thus, potentially away from the water source. Research by Beggs et al. (2019a) suggests the opposite, whereas only 50% of farms of fewer than 300 cows provided water on the track or at the dairy, 90% of large (500–749 cows) and very large (750+ cows) farms did.

Dairy cows on large pasture-based farms are more likely to drink water from a trough than from a creek or other natural water supply. This is due to the logistics of grazing larger herds combined with a push to keep cattle out of natural waterways. In an intensive rotational grazing system, dominant cows use aggression to monopolise a water source when the trough is placed in a 4 m wide corridor, but not when it is positioned in the paddock (Coimbra et al. 2012). This effect was most evident when the temperature peaked in the middle of the day. More intensive grazing generally reduces opportunities for shade and shelter (Stafford and Gregory 2008), highlighting the importance of accessibility to water in these systems. Other research suggests that the preference of cattle to drink out of large troughs may be associated with reduced competition for access (Jensen and Vestergaard 2021). Like confinement dairy systems, the design of facilities and the outdoor space to ensure animals can simultaneously access resources may become increasingly important as grazing intensifies.

Whereas water accessibility is obviously essential to welfare, water quality also needs to be considered. Water troughs and other water sources can be contaminated with manure and bore water can be contaminated with dissolved minerals (e.g. nitrates, sodium, sulfates, iron (Fe); Jensen and Vestergaard 2021) or bacteria such as cyanobacteria (blue-green algae). Contaminated water may prevent feelings of drinking pleasure, induce thirst if water is rejected, or feelings of poor health and nausea if it leads to illness (e.g. bacterial contamination). For example, dairy cows show no preference between water with 0 or 4 mg of Fe/L, but water intake declines when Fe concentrations exceed 8 mg/L (Genther and Beede 2013). Schütz et al. (2019) found that cow water consumption decreased with an increasing manure contamination. Appropriate groundworks around the trough are needed to reduce the risk of mud and other fresh-water contaminants, particular in the high-traffic areas on large pasture-based dairies. For water contaminated with dissolved solids, water treatments may improve the water quality but otherwise an alternative water source will need to be located. There are few, if any, reports on the quality of drinking water provided to cows on Australasian farms. In one Australian survey, larger farms self-reported as being more likely to provide cows water suitable for human consumption than did smaller farms (50% vs 90% of smaller vs larger farms; Beggs et al. 2019a). In New Zealand, Sapkota et al. (2022) found that only 4 of 23 observed farms met the researcher’s threshold of ‘acceptable’ water quality (clear water, easily visible trough base, no apparent dirt or dead insects), although most of the contaminated water troughs were located on a feed pad rather than in a paddock.

There may also be opportunity to promote drinking pleasure by providing water to cow preference. Cows in temperate climates generally prefer a lukewarm water temperature but prefer a cool temperature when experiencing heat stress in warmer weather (Jensen and Vestergaard 2021). Cows may also have taste preferences, albeit these differ from human preferences. For example, in a preference test, Thomas et al. (2007) found that cows drank 20 times more unflavoured than orange-flavoured water, but Jaster et al. (1978) found increased water intake in Holstein cows provided with saline water (2500 mg/L NaCl) compared with tap water. According to Australia’s and New Zealand’s National Water Quality Management Strategy (2000), water at the level of salinity used in the study by Jaster et al. (1978) is safe for dairy cattle.

A predominantly grazed diet promotes positive nutritional experiences such as masticatory and food-taste pleasures, but there is less control over intake from pasture than from TMR systems. This could increase the risk of hunger in some cows, particularly in early lactation. Directly monitoring the nutrition of individual cows can be difficult in larger herds without the support of technologies such as individual milk monitors, and automatic body condition scoring and weighing systems (technologies are discussed later in this review). Larger pasture-based farms feed higher amounts of concentrate and this could increase the risk of SARA with associated feelings of nausea or malaise; however, these farms are also more likely to use nutritionist-formulated diets to mitigate these risks. Appropriate infrastructure and groundwork around the water trough will ensure the supply of high-quality water and associated drinking pleasure in pasture-based herds.

Heat stress is characterised by an increase in body temperature where significant physiological mechanisms are triggered to maintain or return to homeostasis. Although the effect of heat stress on dairy farming is well studied globally, this is mostly due to its effect on milk production, reproductive performance and the implications climate change is predicted to have on the productivity of the dairy industry (e.g. de Rensis and Scaramuzzi 2003; West 2003; Carabano et al. 2016; Fabris et al. 2019; Chang-Fung-Martel et al. 2021). Dry-matter intake is reduced in heat-stressed cows and energy redirected from lactation to meet increased metabolic (i.e. nutrient partitioning) and physical (i.e. sweating and panting) demands (West 2003). This puts heat-stressed cows at a greater risk of health disorders, including mastitis, metritis, respiratory ailments, retained fetal membranes and digestive issues (Thompson and Dahl 2012). When it occurs in late gestation and early post-calving, heat stress reduces cow bodyweight and condition (Rhoads et al. 2009). Other research shows reduced calf birth weights when cows experience heat stress during late pregnancy (Baumgard and Rhoads 2012; Ouellet et al. 2020). The implications of heat stress on cow health and productivity extend to their following lactations, to their daughters and to their grand-daughters (Ouellet et al. 2020).

Polsky and von Keyserlingk (2017) highlighted significant research gaps regarding the effect of heat stress on pain, aggression, malaise, hunger, thirst, and frustration in dairy cows. These states have implications for animal welfare. They may also be affected by the management of dairy cattle in hot climatic conditions. For example, cattle seek shelter during adverse weather conditions (Polsky and von Keyserlingk 2017), but this can result in increased aggressive behaviours over competition for shade in hot conditions (Vizzotto et al. 2015). Deniz et al. (2021) found dominant cows more likely to lie in the shaded area of a silvopasture than are their lower-ranking herd mates. Sprinklers are effective at reducing body temperature and lessen the effect of heat stress on milk production (Mayer et al. 1999; West 2003; Legrand et al. 2011), but have also been associated with avoidance behaviours in cattle (Schütz et al. 2011; Chen et al. 2016).

Larger pasture-based dairy farms with more intensive grazing generally reduce opportunities for shade and shelter (Stafford and Gregory 2008). In grazing systems, trees, shrubs, and abiotic structures are the primary source of shading. Sapkota et al. (2022) found that none of the 23 convenience-sampled New Zealand farms provided adequate shade or shelter (i.e. enough for the entire herd in every paddock). A recent survey suggested that the Australian public may not accept permanent housing of cows or gene-editing as strategies to mitigate heat stress (Hendricks et al. 2022). Investment in how to incorporate shading structures in intensive grazing systems, the uptake of newly developed monitoring devices for heat stress or the introduction of tolerant breeds into the genetic pool of existing animals, in combination with other strategies (e.g. once-a-day milking, Schütz et al. 2023), may help reduce heat stress in a warming climate, but are not ‘one-size-fits-all’ solutions.

Cattle are generally well equipped to cope with cold conditions, but can enter thermal distress at temperatures below 0°C for windless conditions and below 11°C and 16°C in wet and windy conditions for lactating and dry cows respectively (see Laven and Holmes 2008). Schütz et al. (2010) used fans and sprinklers to experimentally manipulate the environmental conditions of dry cows. Exposure of 22 h to wet conditions markedly reduced lying time, feed intake and skin temperatures. There was little effect of wind per se, but wind in addition to the simulated rain reduced the average ambient temperature from 10°C to 8°C and magnified the observed responses (Schütz et al. 2010). The findings of Schütz et al. (2010) were replicated by Hendriks et al. (2020) under applied conditions, where decreased air temperature (range from −0.8 to 16.5°C) and increased rainfall (from 0 to 52.6 mm) were associated with reduced lying behaviour, but exposure to wet and cold together exacerbated this response. Tucker et al. (2007) also used sprinklers and fans to simulate winter conditions in an outdoor pen, finding that cows spent less time lying and had higher plasma cortisol, faecal cortisol metabolite and thyroxine than did those in indoor housing, although a high body condition helped insulate against the conditions. Thus, protection from rain and ensuring adequate feeding to maintain good body condition are important, even in mild temperatures (Tucker et al. 2007; Schütz et al. 2010). The latter is particularly important in seasonal dairy production systems where the transition period coincides with colder months. Indeed, prepartum Holstein cows provided with a shelter during winter spend more time lying, are cleaner and have lower non-esterified fatty acid concentrations in the blood, suggesting reduced adipose tissue mobilisation, than for similar cows without a shelter (Cartes et al. 2021).

Mud is an inevitable outcome of slow pasture growth, rain and intensive land use during winter-grazing periods, for example, with forage crops (Fisher 2020). Excessive or prolonged exposure to mud has negative implications for animal welfare through its effects on discomfort, chilling, skin irritation and pain, weakness, exhaustion, frustration and depression (Fisher 2020). A decrease in lying time in wet and muddy conditions may reflect a lack of cow comfort or a strategy to minimise heat loss (discussed by Hendriks et al. 2020). Neave et al. (2022) found that lying time on the day of and day after a rainfall event reduces by 24 and 29 min per 1 mm increase in rain respectively. On a day of high rainfall (12 mm), cows lay for only 2.5 h on average, and up to 38% of cows in some groups did not lie for 24 h (Neave et al. 2022).

Reduced lying times have welfare implications in terms of increased stress and immunosuppression (Fisher 2020). Pastoral dairy cows spend, on average, 10–12 h per day lying down (Fisher 2020). The previously mentioned experiment by Schütz et al. (2010) reported that cows spent an average of 2 of 22 hours lying down in wet conditions. Chen et al. (2017) experimentally manipulated soil moisture and found that muddy conditions were associated with reduced time lying (74% and 54% reduction for heifers and cows respectively), immunosuppression, dirtiness and an increase in lying with the legs tucked up to limit body contact with the muddy surface. This research was conducted in indoor pens, demonstrating the implications of mud even in the absence of wind and rain (Chen et al. 2017). Wet and muddy conditions can also increase the risk of lameness associated with softening of the keratin of the feet of cows (O’Driscoll et al. 2008a; Stafford and Gregory 2008) and increased susceptibility to mastitis, dermatitis, hypothermia and metabolic diseases (Fisher 2020). Dirty lower legs are common in Australian pasture-based systems but are thought to be related to walking on muddy tracks rather than lying in muddy areas (Beggs et al. 2019a). By contrast, New Zealand typically uses higher stocking rates than Australia (see Fariña and Chilibroste 2019) and has experienced public concern regarding images of cows deep in mud (discussed by Fisher 2020). Sapkota et al. (2022) recorded a median of 17.3% of cows with ‘very dirty’ hindquarters and/or flanks during their assessment of 23 New Zealand dairy farms.

Shelter is rarely provided in pasture-based systems and is becoming increasingly difficult to incorporate as grazing intensifies (Stafford and Gregory 2008). Windbreaks such as woody shelterbelts and riparian plantings may lessen the impact of environmental extremes on animal welfare and productivity (reviewed by Baker et al. 2018 and England et al. 2020). A shelterbreak reduces windspeeds by 40–70% within a horizontal distance of 10 times the break height (Haozhe et al. 2019) and changes the microclimate of the paddock so that warmer temperatures are felt during the cool of the night and cooler temperatures in the heat of the day (Baker et al. 2021). The most profound effects are felt close to the shelterbelt (Haozhe et al. 2019; Baker et al. 2021), highlighting the importance of access to the area near the windbreak. Despite these benefits, the adoption of windbreak shelter is low in Australia (Baker et al. 2018) and increasingly shelterbelts are being removed to facilitate pivot irrigation (Fisher et al. 2019). Farmer interviews in Australia (Fleming et al. 2019) and New Zealand (Fisher et al. 2019) have shown that many see shelter as a part of good farming, but barriers relating to costs, time, a lack of information and concerns over the impacts on production are preventing greater adoption of shelter such as trees on farms. Fisher et al. (2019) further identified different understanding on what constitutes good shelter and a lack of enforcement as contributing to the lack of shelter provision.

Removing cows from pasture and housing them on a drained loafing area, or ‘stand-off pad’, during periods of high rainfall is a strategy used by some farmers to minimise soil damage and nitrate loss. There is limited research on the features of the stand-off pad or their implications for animal welfare. Recent New Zealand research on uncovered stand-off pads found that sand flooring does not meet the minimum requirements for animal comfort and welfare, and a stone surface discouraged cows from transitioning between postures and from walking (Al-Marashdeh et al. 2019). Woodchip and a geotextile carpet were equal best performers in terms of animal comfort and welfare. A body of Irish research shows that providing cover on a winter pad reduces dirtiness of the pad (O’Driscoll et al. 2008b), improves hoof hardness (O’Driscoll et al. 2008a), reduces sole lesions (O’Driscoll et al. 2009) and reduces body heat energy lost to the environment (O’Driscoll et al. 2010) compared with those on an uncovered winter pad. A simple roofing structure over the winter-pad should be encouraged. Other features of the stand-off pad such as the number and location of feeding or water sources (which may have consequences for competition and social stress) and the frequency of bedding refreshment also require investigation.

Providing increased grain supplement as a mixed ration on a concrete feed pad after milking and then returning cows to pasture is the fastest-growing system for feeding dairy cows in Australia. These ‘partial mixed ration’ (PMR) systems made up 12% of Australian dairy farms in 2015 and 18% in 2019 (Dairy Australia 2019). Benefits of PMR systems include reduced feed and water wastage (Barber et al. 2019), as well as the potential to increase intake and thus milk production per cow (Wales et al. 2013), although none of Auldist et al. (2013), Wright et al. (2014) or Golder et al. (2014) found any difference in milk yield when supplements were fed in a PMR or as a concentrate in the dairy. Cows in a PMR system spend 2–3 h standing on a concrete feed pad when they would otherwise be on pasture. There is no evidence that this increases the risk of claw lesions or lameness (Coombe et al. 2013; Golder et al. 2014). Feeding supplements on a feed pad as a mixed ration does not affect the risk of ruminal acidosis compared with feeding the same level of supplement as a concentrate in the dairy (Auldist et al. 2013; Golder et al. 2014; Wright et al. 2014; Compton et al. 2015), unless the amounts being fed exceed ~14 kg DM/day (Golder et al. 2014). Daily grazing time is reduced in a PMR, likely because cows are spending less time on pasture, but lying- and rumination-time budgets are not affected (Hetti Arachchige et al. 2013).

The research described in the previous paragraph provided no indication of a negative effect of PMR feeding systems on cow welfare; however, there is a gap in the scientific literature regarding the levels of competition on the feed pad and how this relates to welfare outcomes. In indoor housing systems, factors such as total space allowance, stocking density and feeding spaces are important determinants of the levels of competition, agonistic behaviour and social stress, especially when access to feed is restricted (see review by Jensen 2018). The effect of feed-pad design on competition in a PMR system has been explored only by Hetti Arachchige et al. (2014). This research found that increasing trough space on a feed pad from 0.6 to 1.0 m per cow or putting up barriers to separate feeding positions increased the time cows spent feeding and decreased the number of feedings bouts, aggressive behaviours, displacements and heart rate. The reduction in heart rate and the number of feeding bouts was most pronounced for the smaller, subordinate cows. Other research has found that variation among cows in time spent eating doubled as the amount of supplement offered in a PMR increased from 6 kg DM/cow to 12 or 14 kg DM/cow per day (Wright et al. 2016), suggesting inequitable intake of concentrates on the feed pad. Comparable findings have been reported in cows housed indoors and provided with restricted temporal access to a TMR (Collings et al. 2011) or under high levels of competition for access to the TMR (Crossley et al. 2017).

Electrified fencing is a common infrastructure on pasture-based farms. It is used to contain and control the movement of cattle. Most permanent perimeter fences are electrified and mobile electrified poly-tape is used to front or back-fence cattle in intensive grazing systems. Electric fences represent a psychological rather than a physical barrier (Stookey 2010). Through the process of avoidance learning, cattle learn to associate the visual stimulus of the fence to an aversive electric shock, leading to active avoidance of the fence.

Grumett and Butterworth (2022) considered electric fencing a welfare issue because it delivers an electric shock and thus causes pain. Electric shock causes some discomfort, but the welfare consequences of this are not clear, and the costs of this discomfort need to be considered along with the management and human-safety benefits brought by electric fencing. Cattle quickly learn to avoid the electric fence, which reduces the number of shocks received. Verdon et al. (2020) observed the interactions of naïve heifers with electric fencing during avoidance learning. On average, the heifers received an electric shock eight times over 6 days following introduction to the electric fence (range 5–10). Nearly 50% of shocks were received on the first day of training and most were delivered as the animal was exploring the fence. By contrast, heifers received less than one shock per day over the remaining five observed days. The learning was maintained over time; only one interaction with the electric fence was observed when researchers repeated the observations 8 weeks later (Verdon et al. 2020). Other research has similarly found that dairy bull calves receive an average of 7.5 shocks from an electric fence over a 7-day training period, with most shocks being received during the first hour of Training day 1 (Martiskainen et al. 2008).

Ensuring fast and effective associative learning combined with the appropriate use of electric fencing are key to providing cows with control over the receipt of electric shock, thereby reducing the number of shocks received. For the former, Grandin (1999) recommended that the first contact cows have with an electric fence results in a significant shock to create a longer-lasting conditioned avoidance of the electric fence. Markus et al. (1998) found that steers learn to avoid an electric fence more quickly on a high-shock intensity (5.8 kV) than a medium (5.6 kV) or low (5.0 kV) intensities. There is no evidence that fear prevents the approach of cows to electric fencing following aversion learning. For example, cows frequently graze over and under the electric fence without breaking through (McDonald et al. 1981). Nonetheless, electric fencing should never be used in ways that compromise the animal’s ability to avoid contacting the fence, such as containing animals in very small areas, at very high stocking densities, with tight angles or in ways that obscures the fences visibility. It is possible that the greater available space provided in larger herds facilitates more effective avoidance of the electric fence, but this requires investigation.

New virtual-fencing technologies may replace mobile poly-wire electric fencing as well as some perimeter electric fencing. The founding principles of virtual fencing are similar to those of electric fencing, except that the visual cue of the fence is replaced by an audio cue (e.g. Lomax et al. 2019; Langworthy et al. 2021). Cattle rapidly learn the association between audio and electrical stimuli; in dairy cows, this is achieved within three to five interactions (Langworthy et al. 2021). Ensuring that dairy heifers have prior experience with electric fencing (Verdon et al. 2020) and are trained close to calving age compared with a younger age (Verdon and Rawnsley 2020) improves the rate of learning. Using a simple grazing regime featuring a single virtual front fence over a short study period (10 days with the virtual fence), Langworthy et al. (2021) found that the technology successfully managed grazing dairy cows off a fresh break of pasture, even when their allocation became depleted. The technology did not affect cow behaviour or welfare in the days immediately following its implementation, but there were some indications of increased stress from Day 4 with the virtual fence (Verdon et al. 2021). More recent research using a different VF technology has reported no difference in the welfare of dairy cows under VF compared with a conventional management, assessed through feed intake, bodyweight, milk yield, milk cortisol concentrations and activity (Fuchs et al. 2024). The virtual-fencing prototype used by Verdon et al. (2021) was designed for extensively grazed cattle rather than intensive pasture systems. New virtual-fencing technologies designed specifically for use in intensive dairy systems have recently become available (e.g. Verdon et al. 2024). These technologies enable virtual herding (i.e. bringing cows into the dairy) as well as virtual-fencing capabilities, and include features to enhance environmental predictability and controllability (e.g. individualised cue delivery). The welfare implications of this technology are being scientifically assessed.

Managing cows outdoors at pasture provides an enhanced physical environment with space for spontaneous locomotion, comfortable resting, fresh air, and natural lighting. These features of the physical environment promote visual, respiratory, and physical comfort. However, managing cows at pasture exposes them to thermal discomfort during environmental extremes and often requires their containment using electricity. We posit that the welfare benefits associated with housing and feeding of cows at pasture outweigh negatives associated with the receipt of electric shock, which appears to be infrequent once training is complete. Abundant shelter is rarely provided in pasture-based systems and is increasingly difficult to incorporate in larger herds. This presents welfare challenges that will be exacerbated by a changing climate in which an increased number of hot days and erratic weather events are expected. Overcoming this challenge will require innovative thinking, for example, sprinkler systems, mobile infrastructure and increased fenceline tree planting.

Lameness is defined as ‘impaired movement or deviation from normal gait’ (Cockram and Hughes 2011). Pain is its primary cause, and the resulting reduced mobility can prevent lame animals from accessing resources such as food, shade and water (see reviews by von Keyserlingk et al. 2009; Cockram and Hughes 2011; Fisher and Webster 2013; Mee and Boyle 2020). There is evidence that lame cows experience physiological stress, indicated by increased milk, hair and faecal cortisol concentrations (Gellrich et al. 2015; Janßen et al. 2016; Jurkovich et al. 2020), markers of immunosuppression (O’Driscoll et al. 2015) and heart-rate parameters (Jurkovich et al. 2020). Lame cows lose weight, have impaired reproduction, produce less milk and are at a greater risk of culling (Dewes 1978; Green et al. 2002; Booth et al. 2004; Green et al. 2010; Alawneh et al. 2011; Alawneh et al. 2012; Kamphuis et al. 2013; Somers et al. 2015; Cook 2020). These effects can have a chronic impact on animal welfare; it takes up to 35 days for a hoof lesion to fully recover (Tranter and Morris 1991), at least 28 days for locomotion and nociceptive thresholds to improve (Laven et al. 2008), and 4 weeks for lame cows to regain lost bodyweight (Alawneh et al. 2012).

The reported prevalence of lameness in pasture-based dairy herds is presented in Table 2. Trained researchers conducted lameness assessments in 13 of the available studies, with an average herd prevalence of 21.4% (Table 2). Lameness continues to be listed as one of the most common health conditions on Australian dairy farms (Beggs et al. 2019a, 2019b). Although pasture provides some level of protection from lameness (e.g. Haskell et al. 2006; Hernandez-Mendo et al. 2007), and can aid in lameness recovery (McLellan et al. 2022), low levels of lameness on pasture-based farms should not been seen as a guarantee.

A poorly maintained walking track increases the risk of hoof trauma (reviewed by Hund et al. 2019), and for decades has been listed as a primary risk factor for lameness on pasture-based dairies (Dewes 1978; Chesterton et al. 1989; Doherty et al. 2014; Moreira et al. 2019). A recent New Zealand study found that less than half of 23 convenience-sampled farms met track-condition targets (Sapkota et al. 2022). In Ireland, Browne et al. (2022) classified at least one track as rough or very rough on ~60% of surveyed dairy farms, despite 38% farmers having added new tracks and 34% having renovated an old cow track in the past 5 years. Impatient stock handling exacerbates the risk of lameness on a poorly maintained track (Chesterton et al. 1989; Doherty et al. 2014; Ranjbar et al. 2016; Moreira et al. 2019). Factors associated with movement of animals to the milking parlor accounts for 40% of the variation in lameness prevalence (Chesterton et al. 1989), and every 1 km/h increase in the average speed of movement increases the risk of lameness by 5% (Bran et al. 2018b). The condition of the pre-milking holding yard, in terms of cleanliness, damaged concrete and available space per cow, and sharp turns entering and exiting the dairy is also a lameness risk factor (Dewes 1978; Chesterton et al. 1989; Barker et al. 2010; Moreira et al. 2019). Every square-metre increase in available space per cow in the holding yard reduces the odds of lameness by 33% (Ranjbar et al. 2016).

Cows in large herds walk further to and from the dairy (Beggs et al. 2015), which presumably could increase the risk of lameness. However, comparison of the single-herd studies reported in Table 2 show no obvious relationship between herd size and the incidence of lameness, nor did the research by Beggs et al. (2015, 2019a) and Sapkota et al. (2022) report such a relationship. One explanation for this is that the increased walking time in large herds provides more opportunities to identify lame cows. Large herds are also more likely to use technologies that may help in the identification of lameness (discussed further under ‘technology’), although commercially available products are less able to identify early onset lameness (e.g. de Mol et al. 2013; Beer et al. 2016). Early identification and treatment of hoof lesions is important to reducing the severity and duration of a lameness event and thus its impact on animal welfare. For example, therapeutic hoof trimming of mildly lame cows increases the chance of full recovery and decreases the time cows spend lame (Groenevelt et al. 2014), but the same effect is not observed in moderately or severely lame cows (Thomas et al. 2016; García-Muñoz et al. 2017; Miguel-Pacheco et al. 2017). Whereas one New Zealand study found that 40% of clinically lame cows were treated more than 3 weeks after the lameness was identified (Alawneh et al. 2012), a survey of Australian dairy farmers found that it took an average of 55 h (range 2–30 days) from the detection to treatment of lameness (Ranjbar et al. 2020).

There is a question as to why lameness is such a stubborn welfare problem for the dairy industry when the causes, effects, preventions, and treatments have been well researched. Wynands et al. (2022) identified a lack of agreement on a definition of lameness, normalisation to its signs, and the interconnectedness of lameness with other health and management issues as contributing factors to the perception of American dairy farmers that lameness is inevitable. According to Grandin (2018), a failure to measure chronic painful conditions such as lame livestock is a factor contributing to its persistence. Health data for dairy cows are sparsely recorded in Australia (Pryce et al. 2013). Survey data show that 72.5% of Australian dairy farmers fail to keep lameness records (Ranjbar et al. 2020), and Table 2 shows that farmers consistently underestimate the level of lameness in their herd. Other research has indicated that farmers level of knowledge of lameness affects the perceived lameness risk on their farm (Olmos et al. 2018). Regularly recording health and welfare outcomes and engagement in peer benchmarking activities make issues such as lameness visible, combat desensitisation and change blindness, and highlight what constitutes normal or abnormal performance (discussed by Verdon 2022). For example, a UK study recorded a reduction in the prevalence of lameness on farms that started monitoring lame animals, but reductions were greater when additional advice and support was provided (Main et al. 2012). An Australian survey found that dairy farmers perceived better equipment and facilities, improved knowledge and training, and a favourable cost–benefit ratio as factors that would enable them to improve their management practices and reduce lameness (Dutton-Regester et al. 2019). The authors suggested that strengthening farmer beliefs regarding improved animal welfare and increased milk production and challenging their belief regarding cost may increase the uptake of lameness-reduction strategies. These sentiments were echoed by a scientific review that concluded that although research has shifted its focus from the treatment to the prevention of lameness, behaviour changes among dairy farmers, consultants and veterinarians are needed before this shift is realised on farm (Sadiq et al. 2019).

Mastitis is an inflammation of the mammary gland that is primarily caused by a bacterial infection that enters through the teat canal causing an intra-mammary infection (IMI) (Whistance 2010). Behavioural (hypoalgesia, anorexia, reduced self-grooming, rumination, competitive behaviour and time lying, increased activity and restless behaviour during milking) and physiological (rectal temperature, udder inflammation) indicators of pain and illness are observed from Day −5 to Day +2, relative to the detection (or induction in experimental settings) and treatment of acute mastitis in dairy cows (Rasmussen et al. 2011; Fogsgaard et al. 2012; Medrano-Galarza et al. 2012; Sepúlveda-Varas et al. 2016). The symptoms of pain and illness become more obvious with increasing severity of the infection (Tomazi et al. 2018). Producers and veterinarians generally accept that severe mastitis is painful, but the potential impact of mild to moderate mastitis on animal welfare may be overlooked (Petersson-Wolfe et al. 2018). Indeed, there is behavioural evidence of pain in cows with mild cases of mastitis (Fitzpatrick et al. 1998).

There is often less mastitis in pasture-based than in more intensive indoor dairy systems, and the bacteria causing it differ (see reviews by Arnott et al. 2017; Mee and Boyle 2020). Nonetheless, Australian research has suggested that 30% of herds do not meet the industry target of fewer than 5% of cows with clinical mastitis (IMI with symptoms that are visible to the naked eye, e.g. udder swelling, redness) in the first 30 days of lactation (Beggs et al. 2019a). An ~10% lactational incidence of clinical mastitis in pasture-based dairy systems is reported in the published literature (McDougall 1999; Morton et al. 2014; Bates and Dohoo 2016). Subclinical mastitis is generally defined as an individual cow milk somatic-cell count (SCC) of >200,000 cells/mL and/or isolation of bacteria from a milk sample in the absence of signs of clinical mastitis (e.g. Compton et al. 2007; Cardozo et al. 2015; Villa-Arcila et al. 2017). The literature has reported a mean prevalence of 29–48.5% of cows with subclinical mastitis (Plozza et al. 2011; Suárez et al. 2017), suggesting that a proportion of IMI may be undetected. However, Australia’s national bulk milk-cell count (BMCC) statistics suggest that IMI is reducing over the years. In 2010, 34.2% of farms had an average BMCC of between 250,000 and 400,000 cells/mL, compared with 12.9% in 2019 (Morton 2021).

In-line milk sensors can aid the early detection of mastitis and thus expedite its treatment (reviewed by Shalloo et al. 2021). The published literature has reported the adoption of in-line sensors at 8% of New Zealand dairy farms (Yang et al. 2021) and 15% of Australian farms (Gargiulo et al. 2018). Dairy Australia’s ‘Animal Husbandry Survey’ suggested that this number has increased to 24% of Australian farmers in 2022, with a further 10% of farmers intending to install them (L. Sundermann, Dairy Australia, pers. comm.). Large herds are more likely to use this technology (Beggs et al. 2015). A 10-year-old survey found that 22% of dairy farms milking more than 700 cows and 17% of farms milking more than 500 cows used in-line sensors (Beggs et al. 2015). Another survey conducted in 2015 by Gargiulo et al. (2018) found that farmers of large herds were more likely to expect increased adoption of in-line sensors in the future. This was confirmed by Dairy Australia’s 2022 Animal Husbandry Survey, which reported that 46% of farms with >700 cows now have in-line sensors (Dairy Australia 2022). The Australian organisation DataGene (DataGene 2017), which has invested in national herd improvement, also has a target that 60% of the Australian herd will be subject to in-line measurement by 2024. Large farms are also more likely to use automatic post-milking disinfection (Gargiulo et al. 2018; Yang et al. 2021) and have written protocols for the identification and treatment of mastitis (Beggs et al. 2015). In addition to hygienic milking practices, rotational grazing means that cows are not resting on contaminated flooring, which may reduce the risk of environmental mastitis. Thus, although the current available evidence suggests that mastitis is not affected by herd size (Plozza et al. 2011; Beggs et al. 2015; Bates and Dohoo 2016), these systems include many of the management features required to achieve a low rate of mastitis.

Cows that are diagnosed with clinical mastitis typically require antibiotic treatment, which reduces the duration and severity of a mastitis infection (e.g. Shim et al. 2004). There is increasing social pressure on the livestock industries to reduce their reliance on antibiotics. In their review of the management of mastitis, Ruegg (2017) concluded that future research needs to define the appropriate use of antibiotics to control mastitis and prevent new infections from occurring in healthy cows, as a high priority. Selective antibiotic treatment can effectively reduce mastitis when there are low levels in the herd, but the reverse is true for moderate levels of mastitis (McDougall 2002; Plozza et al. 2011). Research from pasture-based dairy systems has shown that use of a teat sealant in conjunction with antibiotic therapy at dry-off reduces subclinical and clinical mastitis in early lactation compared with an antibiotic alone (Laven and Lawrence 2008; Runciman et al. 2010; Laven et al. 2014; Bates et al. 2016; Golder et al. 2016). In their systematic review and meta-analysis, Rabiee and Lean (2013) found that teat-sealant therapy alone, and in the presence of an antibiotic, reduced the risk of clinical mastitis in early lactation by 29% and 48% respectively. The same meta-analysis found that teat sealants reduced the risk of IMI by 73% compared with untreated cows (Rabiee and Lean 2013). Indeed, Bates and Saldias (2018) found that a teat sealant administered to cows with no history of mastitis and a low SCC halved the incidence of clinical mastitis in early lactation and reduced subclinical mastitis by 33%, compared with untreated cows. Other research suggests that the use of a teat sealant in heifers reduces IMI, clinical and subclinical mastitis and increases milk production in the first 30 days of lactation (Robertson et al. 2017). Thus, the blanket treatment of cows and late-gestation heifers with a teat sealant in conjunction with targeted use of antibiotics (based of clinical history and an average individual SCC, e.g. of fewer than 100,000 cells/mL for the year, with no test more than 200,000 cells/mL) may be an effective strategy to reduce the risk of IMI while minimising antibiotic reliance.

Larger farms lead the adoption of technology on pasture-based dairies (Beggs et al. 2015; Edwards et al. 2015; Gargiulo et al. 2018; Beggs et al. 2019a). As discussed by Weary and von Keyserlingk (2023), researchers must consider the opportunities for technology to improve as well as to harm animal welfare. For example, there are concerns that incorporation of technology would increase cows’ fear of humans by reducing the number and length of stockperson interactions with cows and by altering the proportion of positive (e.g. feeding) to negative (e.g. hoof trimming) interactions (Hostiou et al. 2017), although Beggs et al. (2019a) found no evidence that herd size affected avoidance of humans in the paddock. Alternatively, technology could reduce the risk of negative interactions with stockpeople particularly during mundane or repetitive tasks (e.g. moving cows to the dairy). The relationship between humans and cows, and how this may affect cow welfare in large herds, is further explored later in this review (see ‘Interactions with humans’).

A perceived benefit of technology in large herds is an ability to counteract the reduced capacity of stockpeople to monitor the health of individual cows (Beggs et al. 2019a). Whereas technologies that save labour and time have the greatest uptake (Yang et al. 2021), some of these may help prevent health challenges. For example, automatic post-milking teat sprayers and cup removers are two of the most implemented technologies and have the greatest uptake on larger farms (Gargiulo et al. 2018; Yang et al. 2021). These technologies promote hygienic milking practices that can reduce the risk of mastitis (e.g. Williamson and Lacy-Hulbert 2013). Technologies used to capture data (e.g. automatic weighing, automatic heat detection) can also be used to monitor cow health and help in the early identification of health challenges (e.g. Beauchemin 2018). The uptake of these types of technologies is generally low compared with uptake of time-saving technologies, and reportedly unaffected by herd size (Yang et al. 2021). For example, in 2018 only 2.6% of Australian farmers and 9% of New Zealand farmers used automatic walk-over weighing technology, 4% and 7% used automatic oestrus detection and 15% had in-line mastitis-detection capabilities (Gargiulo et al. 2018; Yang et al. 2021). Twenty-six per cent of Australian farmers expected increased adoption of automatic weighing in the future compared with 64% and 76% expecting increased uptake of automatic mastitis- and oestrus-detection tools, respectively (Gargiulo et al. 2018). Recent Australian industry data suggest half of farms with >300 cows used sensors to manage herd health, although the type of sensors used was not reported (Dairy Australia 2022). Production variables such as weight, body condition and reproduction can provide insight into a cow’s nutritional, disease, and chronic stress status (discussed by Verdon 2023). The slow uptake of technologies such as walk-over scales may be related to confusion over how to use and apply the data to achieve benefits on farm (Gargiulo et al. 2018).

Wearable accelerometer technologies that continuously monitor the cow behaviour are a good example of a product marketed as a production tool (e.g. oestrus detection), but that may also be useful to monitor changes in animal health (Weary et al. 2009). For example, a reduction in rumination time can indicate ruminal acidosis (Beauchemin 2018), increased time lying and reduced time ruminating postpartum can indicate metritis (Held-Montaldo et al. 2021), increased restlessness and reduced rumination are related to the onset of parturition (Saint-Dizier and Chastant-Maillard 2015; Borchers et al. 2017; Kovács et al. 2017), hindleg activity in the crush can indicate stress or discomfort (Gomez et al. 2017; Stewart et al. 2017), and decreased rumination and feeding time are symptomatic of mastitis (Jaeger et al. 2019) and lameness (Beer et al. 2016; Barker et al. 2018). Wearable technologies may also be used to improve the precision of nutrition-related decisions to better manage variability of nutritional status among grazing cows (González et al. 2018). Researchers have used behaviour-monitoring technologies to understand how cows respond to different grazing regimes (e.g. Verdon et al. 2018) and predict an insufficient pasture allocation (Shafiullah et al. 2019; Werner et al. 2019). Technology that measures the dry-matter intake of grazing cattle is still under development, but recent studies have suggested that this may be determined through algorithms based on time spent grazing (Greenwood et al. 2018) or chewing sounds (Galli et al. 2018).

However, there are some challenges to relying on accelerometer technologies to detect health disorders. First, although their ability to detect moderate to severe challenges has been demonstrated (e.g. Beer et al. 2016), their ability to detect minor health disorders requires further investigation. Second, only 30% of retailed accelerometers have been externally validated, and technologies used for health evaluations are most likely to be classified as ‘low performance’ (Stygar et al. 2021). Third, much of the accelerometer technologies have been developed for cows housed indoors, and although some have been validated for pasture-based systems (Werner et al. 2017; Rombach et al. 2018), others are less successful transferring across the systems (e.g. Ambriz-Vilchis et al. 2015; Rombach et al. 2018). Validation of these technologies for the detection of health challenges in pasture-based settings is needed.

Machine vision is developing as an alternative to wearable technologies to monitor livestock behaviour (Nasirahmadi et al. 2017). The lying behaviour of cows in a free-stall barn has been detected by machine vision with 92% sensitivity (Porto et al. 2013) and aggressive behaviour with 85% sensitivity (Guzhva et al. 2016). Machine vision may be useful in detecting gait changes associated with lameness (Abdul Jabbar et al. 2017; Halachmi et al. 2019) and remotely monitoring weight or body condition in pasture-based systems (Nasirahmadi et al. 2017). Respiration rate (Stewart et al. 2017), mastitis (Colak et al. 2008) and hoof lesions (Alsaaod and Büscher 2012) may all be detected with infrared thermography and machine vision, but this technology is rarely seen in pasture-based dairies.

Culling describes the removal of cows from the farm for immediate slaughter if they present with persisting or severe health issues (e.g. illness, injury, infertility; termed involuntary culling), or are no longer productive or desirable as breeding animals (e.g. poor milk production, temperament; termed voluntary culling). As described by Cook (2020), ‘An individual dairy cows productive life in the herd is determined by her health and fertility and the economic need for dairy producers to optimise the milk production of each individual in the herd, for a duration determined by replacement availability’ (p. 3846). There is a paucity of peer-reviewed scientific literature regarding culling decisions and outcomes for cows in pasture-based dairy systems. Consequently, the following paragraphs provide a more international perspective of culling in the dairy cow.

Average culling rates reported in the literature range from 21.3% to 33.7% of the herd each year (Gardner et al. 1990; Maher et al. 2008; Bell et al. 2010; Chiumia et al. 2013; Shock et al. 2018). Nearly 70% of culling decisions are classified as involuntary (Bell et al. 2010), with mastitis and lameness the most frequently listed reasons (e.g. Bell et al. 2010; Brickell and Wathes 2011; Langford and Stott 2012; Cha et al. 2013; Chiumia et al. 2013; Adamczyk et al. 2017; Dahl-Pedersen et al. 2018). Cow factors such as high parity, BCS at service, SCC and milk yield acceleration, are significant predisposing risks to culling (Hadley et al. 2006; Bell et al. 2010; Cha et al. 2013; Chiumia et al. 2013; Gussmann et al. 2019), whereas external factors, such as milk price, management factors, such as herd expansion and transitioning to robotic milking, and farmer attitudes also influence culling decisions (Hadley et al. 2006; Bugueiro et al. 2019; Stojkov et al. 2020b).

A systematic review found that the incidence risk of culling for individual dairy cattle on pasture-based farms ranged from 0.14 to 0.28 (Compton et al. 2017). There is no evidence that this risk changed over a study period from 1986 to 2007, a time during which the average New Zealand herd size increased from 145 to 351 (Dairy NZ 2009; Wilkinson et al. 2020). More recent data are required to assess whether culling rates are affected by modern improvements in reproductive efficiency and the availability of sexed semen (De Vries 2020). The previously mentioned review by Compton et al. (2017), as well as an Australian study of nearly 2.5 million culling records from 1995 to 2016 (Workie et al. 2021), found that the risk of voluntary culling has decreased over the years. More specifically, Workie et al. (2021) found that culling for fertility increased and culling for production decreased between 1995 and 2016. Earlier research similarly pointed to a decrease in voluntary culling and an increase in involuntary culling in expanding dairy herds (Weigel et al. 2003).

Transportation of cull animals to slaughter is an integral step in livestock production. Considerable attention must be given to the fitness of animals for the intended journey (Cockram 2019). Explicitly put by Grandin (2001), ‘it is impossible to ensure good animal welfare during transport if the animal is unfit’ (p. E201). Animals sent for slaughter with pre-existing conditions have an increased risk of dying in transit, becoming non-ambulatory, or being euthanised on arrival compared with healthy animals (Cockram 2019). Stojkov et al. (2020b) found that more cull cows died during transport (up to 11.9%) and were euthanised on arrival to the processors (up to 20.5%) when the number of animals classified as having poor fitness on the farm prior to loading increased from 16% to 26%. Danish research found that 8 h of road transport increased the incidence of lameness by 10%, milk leakage by 16% and wounds by 12% in cull dairy cows (Dahl-Pedersen et al. 2018). Moreover, 20% of cows either became lame or more lame during transport, including 2.2% cows that were severely lame on arrival (Dahl-Pedersen et al. 2018).

Some cull dairy cows are transported for slaughter with identifiable pathological conditions. Seventy-five per cent of Danish cull cows deviate from normal in at least one clinical measure when assessed on farm prior to transport (Dahl-Pedersen et al. 2018). Most would have been considered ‘slightly ill or injured’ and, consequently, fit to transport under the guidance of the European Regulation, despite 16% of cows being thin (BCS ≤ 2.75) and 31% being lame (locomotion score (LS) ≥ 3). Similar on-farm assessments in Canada found that up to 9% of cows are transported when thin (BCS ≤ 2) and up to 26% were classified as having poor fitness to transport (Stojkov et al. 2020b). Assessments of cull dairy cows at livestock markets in northern America reported 10–40% of cows being thin (BCS ≤ 2), 60–73% having locomotion problems (LS ≥ 2), 7% being severely lame (LS ≥ 4), and 13% having engorged or inflamed udders (Moorman et al. 2018; Stojkov et al. 2020a). Stojkov et al. (2020a) deemed that 30% of the cows sold at the observed livestock markets had poor fitness for transport. Similar issues have been identified in dairy cows arriving at the processing facility (37% BCS ≤ 2, 8% LS ≥ 4; Nicholson et al. 2013).

Marketing cull cows when they are still fit will improve their fitness for transport, whereas delaying the decision to cull allows a cow to deteriorate (Grandin 2001). A recent survey of Canadian dairy farmers by Roche et al. (2020) found that the majority of lame cows are transported within 1 week of a culling decision having been made, but 1/4 of respondents reported an interval of 1–3 weeks between the cull decision and transport, 11% reported an interval between 3 and 6 weeks and 11% reported an interval >6 weeks. The authors also found several gaps between producer perception and reality, including a strong confidence that the cull cow would arrive at slaughter in the same condition as they left (Roche et al. 2020). Dairy Australia (2022) found that 67% and 58% of lactating and dry cull cows in Australia were consigned direct to the abattoir, with most of the remaining cows sold via a sale yard, although there was regional variation around this. Anecdotally, long wait times for direct consignment and the hope that someone else will buy the cows for a ‘second chance’ affected farmer decision-making around cull-cow management (Dairy Australia, pers. comm.). Providing farmers with information relating to the condition of their cows on arrival to the processor had no clear effect on their subsequent culling decisions (Stojkov et al. 2020b). A potential conflict between the financial loss associated with not transporting an unfit animal and the financial return from transporting that animal to slaughter may be affecting farmers’ decision to transport an unfit cow (Langford and Stott 2012; Cockram 2019). Other factors, such as high milk prices, can lead to delayed culling decisions and an increase in the number of compromised animals being sent to slaughter (Stojkov et al. 2020b). Developing a pathway for cull cows to enter the beef production chain as a niche product (e.g. ‘vintage beef’) may increase their economic value and make conditioning healthy ex-milking cows at pasture a viable alternative to slaughter as a low-value product. This pathway requires further research and development.

Ideally, the farmer will not present cows for culling that are not fit for transport and, in cases where they do, the livestock driver will not load any unfit animals. In Denmark, 72% of livestock drivers admitted to loading unfit dairy cows (Herskin et al. 2017). Although 94% of the drivers said they knew the rules regarding fitness for transport, only 52% correctly answered two questions on the legislation (Herskin et al. 2017). There is variation among stakeholders in how severely affected an animal must be to be considered unfit for transport (Cockram 2019). Dahl-Pedersen et al. (2018) found a low to moderate level of agreement among farmers, veterinarians and livestock drivers assessing fitness to transport in lame cows, with veterinarians being more likely to score a cow as lame and farmers less likely to score a cow as unfit to transport. Considering that current guidelines and regulations are not ensuring that only fit animals are transported to slaughter (Cockram 2019), the development of clear and specific guidelines to assess fitness for transport of cull dairy cows should be a priority.

Lameness and mastitis are the most common health challenges observed on dairy farms and the most frequently listed reasons for involuntary culling. Both conditions are painful and early detection and treatment, as well as prevention strategies, are important for good animal welfare. On pasture-based farms of all sizes, a well-maintained walking track and patient stockperson handling reduces the risk of lameness owing to bruising and other claw injuries. The welfare outcome of a poorly maintained track could be more significant on large farms where cows walk long distances to and from the dairy. Hygienic milking practices and rotational grazing reduce the risk of contagious and environmental mastitis, and written protocols and technologies aid its early identification and treatment. Thus, good lameness and mastitis outcomes should be achievable on well-managed large herds with maintained walking tracks, automatic cup removal, post-milking teat disinfection, in-line mastitis detection, written protocols on the identification and treatment of mastitis, and rotational grazing.

Zero-grazing dairy farms are uncommon in countries such as Australia and New Zealand, although the percentage of permanently housed intensive dairy production systems in Australia doubled to 2% between 2017 and 2019 (Dairy Australia 2019). From 2015, at least 25 farmers or investors in the Australian state of Victoria considered developing new dairies or transitioning old dairies into fully housed systems, representing 2% of the farms in the region but planning to house 10% of the cows (Williams et al. 2020). One Australian dairy farm looking to expand and transition to a free-stall barn received 430 objections from the public (Williams et al. 2020). Access to pasture may not always be a given for cows residing in countries currently recognised for their pasture-based dairy industries.

Time at pasture is important to cattle and its provision promotes positive affective states (see reviews by Špinka 2006; Schütz et al. 2018; Mee and Boyle 2020). The motivation of cattle to access pasture is well demonstrated by experimental research on housed dairy cows (see research by Charlton et al. 2011a, 2011b, 2013; Falk et al. 2012; Nielsen and Wredle 2016; Shepley et al. 2016, 2017; Kismul et al. 2019). Pasture provides a complex environment, increased space, a comfortable lying surface, and the opportunity to graze, manipulate feed, explore and engage in social activities (reviewed by Charlton and Rutter 2017; Schütz et al. 2018). One or a combination of these factors may be influencing the preference of cattle for pasture. Von Keyserlingk et al. (2017) compared how hard cows were willing to work (by pushing on a weighted gate) for access to pasture or fresh TMR feed. Contrary to expectations, most cows pushed as hard or harder to access pasture as they did to access the TMR. The authors suggest that the motivation to access pasture is not driven by hunger, but by a motivation to be outside and engaging in behaviours such as grazing. Other research has found that alternative outdoor areas (e.g. exercise paddock or sand pack) are not as attractive to cows as is pasture (Kismul et al. 2018; Smid et al. 2018), and that dairy cows prefer to graze even when the same cut forage is provided indoors (Shepley et al. 2016). Taken together, these data show that the opportunity to graze is important to cows (Smid et al. 2020).

It is not known how cows perceive the daily routine of being moved from the paddock to the dairy for milking. On one hand, this routine is not conducive to the expression of agency. Alternatively, the predictability and structure of the routine could promote feelings of calm and control. Cows experience both positive (e.g. provision of concentrate, positive human interactions, removal of milk from a heavy udder) and negative (being hurried through the dairy, negative human interactions, high stocking density in the collecting yard, pain if teat cups are left on too long, teat damage from incorrect vacuum settings, risk of injury due to slipping) stimuli prior to, during and after being milked in a parlor (Phillips 2002). Prescott et al. (1998) assessed the motivation of cows to be milked over several Y-maze experiments. Of five opportunities to choose being milked or not, on average cows chose to be milked 2.7 times. The authors observed high individual variability and no difference in choice between high- or low-producing cows. A subsequent experiment by the same authors found that cows exclusively chose concentrate pellets over being milked. These results suggest that the motivation to be milked is generally weak but also variable (Prescott et al. 1998). More recent research by Kézér et al. (2015) and Kovács et al. (2019) examined cardiac measures of autonomic nervous-system activity in mid-lactation multiparous dairy cows at various stages of milking in either a herringbone or rotary system (e.g. calm standing baseline, moving to pre-milking pen, waiting in the holding pen, udder preparation, milking, waiting in the stall post-milking). Neither study found evidence that milking per se is stressful (i.e. from attachment to removal of the last teat cup).

The handling of cows during movement to and at the parlour may be a more important consideration for animal welfare (Endres and Schwartzkopf-Genswein 2018). Kovács et al. (2019) found that heart rate increased from baseline when cows are being driven from their home pen to a rotary milking parlour and while being held in the pre-milking holding pen. Although the increased heart rate during driving could be associated with activity, other recorded cardiac parameters support the notion that being in the holding pen is stressful (Kovács et al. 2019). For example, high stocking densities in the holding pen force close proximity between subordinate and more dominant cows. Some farms use a backing gate to help move cows forward towards the milking parlour as the number of animals waiting to be milked decreases. This can potentially cause injuries by cows slipping, especially if the backing gate has an electric wire (Endres and Schwartzkopf-Genswein 2018). A published survey of Australian dairy farmers found that between 13% and 33% used a backing gate to push cows into the dairy and this was mostly unaffected by herd size (Beggs et al. 2015). Whereas only 3% of small farms (<150 cows) used an electric wire on the backing gate, over 25% of large farms used an electric wire (>500 cows). Conversely, small farms were more likely to use polyethylene pipe as goads when moving cows (63% vs 33%; Beggs et al. 2015).

Little is known about the effect of large herd sizes on the organisation of social behaviour in grazing dairy cows. Observations of the few remaining feral groups of domestic cattle suggests that natural herd sizes consist of ~20 individuals (Bouissou et al. 2001). Increased competition and agonistic behaviour are observed in stable groups of 16 compared with groups of four and eight grazing cattle (Rind and Phillips 1999). Research in housed cattle controlling for stocking density reports increased agonistic behaviour in dynamic groups of 24 compared with groups of six cattle (Jensen and Proudfoot 2017), but there is no evidence of increased agonistic behaviour in larger herds of housed cattle (9 vs 130 cows, Burow et al. 2009; greater than 79 cows, de Vries et al. 2015). These latter findings are comparable to observations in other species (e.g. pigs, Verdon et al. 2015; hens, D’Eath and Keeling 2003). Dairy cows in large herds may split into subgroups on the basis of affiliative relationships and within which social hierarchies may develop. Such subgrouping has been observed in beef cattle grazing rangelands (discussed in Jensen 2018), feral free-grazing cattle (Lazo 1994) and domesticated cattle living under semi-natural conditions (Reinhardt and Reinhardt 1981). Whether subgrouping occurs in large herds of grazing dairy cattle requires investigation.

The stocking rates (cows/ha) listed in Table 1 describe the number of animals per hectare of available farming land and are not an instantaneous measure of the number of animals in 1 ha of land per se. The latter is a measure of how much area is allocated per cow, referred to as the ‘instantaneous stocking rate’ or ‘stocking density’. Strip-grazing involves providing cows access to a restricted amount of pasture at high stocking densities. Teixeira et al. (2017) reported increased agonistic behaviour and reduced grazing behaviour when the stocking density was increased from 200 to 500 cows/ha. Although 500 cows/ha (corresponding to an area allocation of 20 m²/cow) is less common, it is not unusual for cows to be allocated ≤50 m²/cow in a grazing bout. The resulting competitive grazing environment may have consequences for animal welfare in terms of social stress and hunger, particularly for the smaller, less experienced or more subordinate animals in the herd. Verdon et al. (2018) found that older, heavier cows produce more milk than do younger, smaller cows when pasture was provided in seven small strips per day (average 10 m²/cow per pasture allocation) but did not observe this relationship when the same amount of pasture was provided in two larger strips per day (average of 35 m²/cow per pasture allocation). It is surprising that the competitiveness of feeding in an intensive pasture-based dairy system is rarely mentioned, aside from a few scientific reviews (Stafford and Gregory 2008; Mee and Boyle 2020), considering that increasingly high instantaneous stocking rates are likely to feature in the intensification of pasture-based dairy production (Stafford and Gregory 2008).

Humans prominently feature in the life of the dairy cow. This can have a positive or negative effect on animal welfare, depending on the context of the specific interaction. According to Mellor et al. (2020), the human’s attitude, voice, aptitude, and handling skills are all characteristics that affect animal welfare. For example, humans interacting with animals using an uncertain, domineering, or impatient attitude, speaking in a loud or angry voice, being unsuitably trained or qualified, and using erratic or rough handling cause feelings of fear, anxiety, panic, and confusion in the animal. Conversely, having a confident, patient, and empathetic attitude when interacting with animals, using a calm, clear and encouraging voice, being experienced, skilled and suitably trained, and using gentle, reward-focussed handling can promote positive welfare through feelings of calm, control, confidence, and positive bonding (Mellor et al. 2020).

The causal relationships between human attitudes and human behaviour, and how this affects the animals’ fear of humans and, consequently, stress during human–animal interactions, have been demonstrated by well-known intervention studies on dairy farms (Hemsworth et al. 2002) as well other livestock species (Hemsworth et al. 1981a, 1981b, 1986, 1989, 1994; Gonyou et al. 1986). Importantly, it is the human’s attitude towards animals that determines their behaviour towards animals. For example, if a stockperson has a positive attitude towards cows (thinks they are intelligent, learn easily, like being stroked), they are more likely to show positive behaviours towards cows, and the cows are less afraid and stressed in the presence of humans (see review by Waiblinger et al. 2006) and are, consequently, more productive (Bertenshaw and Rowlinson 2009). Conversely, a negative attitude towards cows increases the occurrence of negative behaviours such as slaps, hits and shouting, resulting in increased fear of humans and stress in human presence (Hemsworth et al. 2000; Waiblinger et al. 2002), with proven negative effects on milk yield, protein and fat (Hemsworth et al. 2000) as well as conception rates (Hixon et al. 1981).

Whether cows hold positive or negative perceptions of a human in their environment will depend on factors such as their previous experiences with humans, the context of the interaction, and who the specific human is. Dairy cows show reluctance to move into a location where they have previously received negative handling (Pajor et al. 2000), and will avoid or approach a specific person in a specific location on the basis of their previous positive or negative experiences with that person in that location (Rushen et al. 1999). The presence of a cow’s normal caretaker in the paddock may promote excitement associated with the provision of fresh pasture, but the presence of that same person while the cow is confined in a crush where she has experienced painful husbandry procedures may be associated with uncertainty and fear. Cows are also able to discriminate among different handlers with whom they have had positive or negative handling experiences (Munksgaard et al. 1997). Thus, a cow being milked by a known quiet and calm stockperson may elicit a different response by the same cow when milked by a known loud and erratic stockperson.

The above research has demonstrated causation between human behaviour and animal welfare by manipulating the valence of human–animal interactions, i.e. some interactions were consistently negative, and others were consistently positive. Rather than the occurrence of a single negative human interaction, we hypothesise that generally it is the cumulative effect of negative interactions that produce a pessimistic expectation of future interactions with humans and results in a chronic state of fear in human presence. Some negative behavioural interactions with humans are unavoidable or even necessary to animal health and welfare. For example, confining cattle in the crush for treatment of lameness, or giving vaccinations. Others may currently be a necessary part of animal production, such as artificial insemination. Unavoidable negative human interactions such as these need to be minimised in duration, frequency, and severity. Incidental interactions are also involved in day-to-day management, for example, when moving cows in and out of the dairy or when approached by an animal in the paddock. In these instances, negative human interactions need to be minimised and replaced with neutral or positive interactions. Replacing repetitive mundane handling tasks with technology may help reduce the risk of negative interactions with humans during day-to-day handling. For example, virtual-fencing technology that herds cows to the dairy could reduce the risk of fear or injury associated with cattle being pushed by a stockperson on a bike or with a dog.

The knowledge, skills and attitudes of the stockperson and the workplace culture are key to improved human–animal interactions (Coleman and Hemsworth 2014). Recruitment and retention of a sufficient number of competent stockpeople may become challenging as herd sizes increase, particularly for some regional communities (Eastwood et al. 2020). Social research to better understand what makes working on a dairy farm attractive may help producers adapt to the changing needs and wants of employees (e.g. flexible working hours; Eastwood et al. 2020). Larger farms are more likely to provide formal training for their staff, have written protocols for disease management and incorporate technologies that can help monitor cow health (Stafford and Gregory 2008; Beggs et al. 2015; Gargiulo et al. 2018). These factors combined may reduce the risk of negative human–animal interactions, and, indeed Beggs et al. (2019a), found no evidence of increased human avoidance in cows from large herds. However, it would be beneficial to also understand the opportunities for positive interactions in large herds.

Pasture access allows the more frequent expression of natural and motivated behaviours. The more time cows spend at pasture, the more pronounced the positive effects are. Cows do not express agency in when they are removed from the paddock for milking in conventional dairy systems, but do express some agency in their behavioural interactions with the environment while in the paddock. Although cows in larger herds spend more time off pasture, proportionately they continue to spend more time at pasture than not each day. Further, when they are at pasture, they have access to a greater total area for exploration and movement than do cows in smaller herds. It is not clear whether larger herds challenge the function of a social hierarchy that requires recognition of individual cows, and there is a risk that high instantaneous stocking rates increase competition for resources such as space and food. Guidelines on minimum stocking rates exist in other intensive livestock industries and could reduce this risk of overstocking on the welfare of grazing dairy cows. The quality of the stockperson interactions with the cow has the potential to provoke negative states such as fear or positive states such as excitement and comfort. Larger herds are more likely to use technologies for routine tasks, which could reduce the opportunity for negative interactions with stockpeople. They are also more likely to formally train their staff which, by targeting stockpeople attitudes, may improve the quality of human–animal interactions.