Welfare assessment in pasture-based and confined dairy farms in São Paulo State, Brazil: a descriptive multi-case study identifying health and hygiene constraints

1 Introduction

Animal welfare is a multidimensional property of an animal’s physical and affective state. Its assessment therefore requires information that extends beyond productivity or the absence of overt disease. On-farm protocols commonly combine animal-based outcomes with resource- and management-based measures to evaluate feeding, housing, health, and behavior. The Welfare Quality^® framework formalized this approach into four principles, namely Good Feeding, Good Housing, Good Health, and Appropriate Behavior, and 12 criteria, with individual measures aggregated into herd-level scores and an overall classification (Welfare Quality, 2009; Blokhuis et al., 2010). Its emphasis on animal-based outcomes is a major strength because the same physical resource may have different consequences depending on the animals, climate, and management conditions.

However, aggregation also creates interpretive challenges. Individual measures differ in validity, feasibility, and influence on the final score, and a small number of adverse outcomes may contribute disproportionately to the overall herd classification (de Vries et al., 2013; Tuyttens et al., 2021). Consequently, an overall welfare category should be interpreted together with the underlying criteria, individual measures, and prevalence estimates. Furthermore, transferring a protocol across production environments is not necessarily neutral. Measures designed for cubicle or tie-stall housing may be less applicable to pasture-based systems, where shade availability, walking conditions, access to water, exposure to parasites, and thermal comfort may be particularly relevant. Although contextual adaptation can improve the practical relevance of an assessment, modifications to the measured indicators, observation locations, or aggregation rules may reduce comparability between protocol versions.

Brazilian dairy farming is highly heterogeneous, ranging from pasture-based herds with variable supplementation to continuously housed, high-producing herds. Tropical and subtropical conditions further add seasonal heat, solar radiation, rainfall, mud exposure, and parasite challenges to the welfare issues commonly observed in dairy production. To address these contextual differences, Garcia (2013) developed a pasture-adapted version of the 2009 dairy-cow protocol for Brazilian farms. This adaptation included modifications to water, behavioral, and somatic-cell-count assessments, as well as additional measures related to shade availability and tick infestation. Subsequent protocols for extensive cattle have also incorporated context-specific indicators, including distance to water, thermal comfort, environmental hazards, and behavioral responses during chute handling (Kaurivi et al., 2020; Valente and Stilwell, 2022). More recently, simplified internal audits have been evaluated as tools for routine monitoring before certification (Ferreira et al., 2025). Together, these studies demonstrate that protocol content should be aligned with the production environment and the purpose of the assessment. In addition, research applying a common assessment approach to 51 pasture-based, free-stall, and compost-bedded-pack dairy farms in Brazil confirmed substantial variation among farms and management systems (Kogima et al., 2022).

In this context, the present study characterized welfare indicators in six dairy farms, including three pasture-based farms and three continuously housed farms, distributed across three Brazilian milk-producing regions. One farm from each management context was represented in each region. The assessment combined the 2009 Welfare Quality^® framework with Garcia’s pasture adaptation to identify farm-level strengths, recurring critical domains, and practical priorities for welfare improvement. The objective was descriptive, namely to characterize dairy-cow welfare profiles and identify management areas requiring attention. Accordingly, this study was not designed to estimate current population-level prevalence, validate a new assessment protocol, or test a causal effect of production system. Instead, it provides a measure-level characterization of welfare conditions across contrasting Brazilian dairy-farming contexts and highlights the importance of interpreting farm-level assessments in relation to their environmental, management, and methodological settings. Because the underlying records were collected in 2013-2014, the present study represents a retrospective analysis of a historical dataset; its findings should therefore be interpreted as farm-level case profiles rather than as estimates of current welfare conditions or contemporary prevalence in Brazilian dairy farms.

2 Materials and methods

2.1 Study design and farms

A cross-sectional, descriptive multi-case field study was conducted on six commercial dairy farms in São Paulo State, southeastern Brazil. Farms were identified through professional contacts and selected purposively according to location and production system. Farm productivity was not used as an inclusion criterion. Three milk-producing regions were represented: Vale do Paraiba, Campinas, and São Jose do Rio Preto. Within each region, one pasture-based farm (P1-P3) and one continuously confined farm (C1-C3) were included. The design provided contrasting case contexts but was not intended to yield a representative sample of either system. Farm characteristics recorded during the visits are presented in Table 1. The welfare assessments originated from master’s research conducted at ESALQ/USP and completed in 2015 (Pinto, 2015). The archived records were revisited for the present retrospective multi-case analysis.

Table 1. Characteristics of the six dairy farms at the time of assessment.

P1–P3, pasture-based farms; C1–C3, confined farms. Lactating-cow counts represent herd size at the time of assessment and not the number of cows individually assessed.

Data were collected during one technical visit per farm between December 2013 and February 2014, the warm and rainy season in the study regions. Each assessment began after the first milking of the day and required approximately 7 h, including animal observations, resource inspection, and an interview with the person responsible for herd management. In all three pasture-based herds, lactating cows had year-round pasture access and left the paddocks only for milking or treatment; concentrated feed was offered at troughs where applicable. The archived records did not consistently document, by farm, the configuration of milking restraint, handling chutes, or dedicated indoor feeding and veterinary areas. These facility characteristics could not be reconstructed reliably and are therefore acknowledged as a contextual limitation. The findings represent a single seasonal snapshot of each farm.

2.2 Welfare assessment protocols

Because data were collected between December 2013 and February 2014, confined herds were assessed using the dairy-cow module of the Welfare Quality^® Assessment Protocol for Cattle, 2009 edition (Welfare Quality, 2009), hereafter referred to as the Welfare Quality^® Assessment Protocol for dairy cows. This historical edition, rather than Version 3 released in 2023, governed data collection and score calculation.

The protocol organizes measures under four principles and 12 criteria: absence of prolonged hunger and thirst; comfort around resting, thermal comfort, and ease of movement; absence of injuries, disease, and pain induced by management procedures; and expression of social and other behaviors, a good human-animal relationship, and a positive emotional state. When individual-cow somatic cell count records were unavailable, the bulk-tank mean from the preceding three months was used. This deviation from the Welfare Quality^® Assessment Protocol for dairy cows (Welfare Quality, 2009) is reported explicitly.

Pasture-based herds were assessed using Garcia’s (2013) adaptation of the Welfare Quality^® Assessment Protocol for dairy cows (Welfare Quality, 2009). In Garcia’s pasture adaptation, cubicle-dependent measures considered inapplicable to pasture-based systems were excluded from field assessment rather than assigned replacement or imputed values; criterion and principle scores were subsequently calculated according to the adapted procedures described by Garcia (2013). The setting or duration of selected behavioral tests was modified, bulk-tank somatic cell count was used, and measures of tick infestation, shade provision and quality, and water-trough accessibility were incorporated.

These changes improved contextual relevance but mean that the two implemented procedures did not contain identical measures or aggregation rules. Therefore, their scores were interpreted within the respective protocol and were not treated as directly exchangeable estimates. The principal modifications and their implications are summarized in Table 2.

Table 2. Principal differences between the original welfare quality protocol and the pasture-adapted version used in the study.

Adaptations follow Garcia (2013).

2.3 Animal-, resource-, and management-based measures

Animal-based measures included body condition, locomotion, cleanliness of the lower hind legs, udder, and upper hind legs/flank, integument alterations, nasal, ocular, and vulvar discharge, coughing, labored respiration, diarrhea, and qualitative behavioral assessment. Locomotion and integument outcomes were recorded using the categories specified by the corresponding protocol. The number of animals selected for individual observation followed the Welfare Quality sampling table and animals were chosen at random within the available lactating herd. Because this is a historical dataset, the archived records did not preserve the exact number of cows assessed for every measure on each farm. These denominators could not be reliably reconstructed from the reported percentages because several values were rounded and some measures involved different subsets of animals. The lactating-cow numbers presented in Table 1 therefore represent herd size and not the number of cows individually assessed.

All on-farm assessments were conducted by two trained observers. Before data collection, both were familiarized and calibrated using the 2009 Welfare Quality^® Assessment Protocol for Cattle and Garcia’s pasture adaptation, including a joint review of definitions, scoring categories, sampling, and field-recording forms. Both attended all visits and followed the same procedures. No formal inter- or intra-observer reliability test was documented; therefore, reliability coefficients and observer-related error could not be estimated retrospectively.

Behavioral assessment included agonistic interactions, avoidance distance to an approaching human, and qualitative behavioral assessment. In confined farms, agonistic behavior was observed for 120 min; in pasture-based farms, the adapted procedure used 60 min of continuous observation in the grazing area. The 60-min duration was prescribed by the pasture-adapted procedure rather than selected ad hoc in the present analysis. Because a shorter observation window provides fewer opportunities to detect infrequent agonistic interactions, the resulting rates and scores should be interpreted within each protocol version and should not be compared directly between management contexts. Avoidance distance was assessed near the feeding area in confined farms and in the pasture in the adapted protocol. Qualitative behavioral assessment used the 20 descriptors and visual analog scales specified by Welfare Quality. Access to pasture was the measure used for the criterion expression of other behaviors.

Resource-based measures characterized water-point number, dimensions, cleanliness, function, flow, and access; lying areas and cubicles where applicable; freedom of movement; and shade in pasture-based farms. Under the Welfare Quality^® Assessment Protocol for dairy cows (Welfare Quality, 2009), absence of prolonged thirst was assessed using water-point number, dimensions, cleanliness, function and flow, the availability of at least two water points, and water space per cow. The pasture adaptation retained a trough-based requirement of 6 cm of accessible length or circumference per cow and 20 L/min flow, excluded natural surface-water sources and bowl-drinker equivalents, and counted only trough sections accessible at a height of 0.60–0.80 m (Garcia, 2013). The Version 3 water-scoring algorithm was not applied. The percentage of cows lying partly or completely outside the designated lying area was a scored resting-comfort measure. The cow-to-stall ratio, when noted, was recorded only as contextual information and did not contribute to the Welfare Quality^® score. A structured interview, developed from the farm-management and record-based components of the Welfare Quality^® Assessment Protocol for dairy cows (Welfare Quality, 2009) and expanded with items relevant to Garcia’s pasture adaptation (Garcia, 2013), provided information on herd production, mortality, dystocia, management procedures, bedding, staff training, and available health records. Hot-iron disbudding/dehorning and tail-docking practices, including use of anesthesia and analgesia, were recorded.

2.4 Score calculation and farm classification

Measure-level observations were converted into criterion scores using the algorithms specified by the Welfare Quality^® Assessment Protocol for dairy cows (Welfare Quality, 2009) or Garcia’s pasture adaptation (Garcia, 2013). Criterion scores were aggregated into the four principles on a 0–100 scale using the corresponding multi-criteria model, in which higher scores indicate a more favorable welfare state. Because bulk-tank somatic cell count replaced individual-cow records when necessary, these are protocol-derived research scores under the implemented historical procedures and should not be interpreted as official certification outcomes. The four principle scores were used to assign each farm to one of four official categories: Excellent, Enhanced, Acceptable, or Not classified. A farm was classified as Excellent when all four scores were >55 and at least two were >80; Enhanced when all were >20 and at least two were >55; and Acceptable when all were >10 and at least three were >20. Farms not meeting these conditions were Not classified (Welfare Quality, 2009; Garcia, 2013). The protocol’s five-point indifference threshold was retained; no classification in this dataset depended on equality at a boundary. These rules yielded four Enhanced farms (P1, P3, C1, and C2) and two Acceptable farms (P2 and C3). Farm C3 was particularly sensitive to these thresholds. With the other principle scores held constant, a one-point reduction in Appropriate Behavior from 11 to 10, or a two-point reduction in Good Health from 22 to 20, would change its final classification from Acceptable to Not classified. This represents deterministic threshold sensitivity rather than statistical uncertainty.

2.5 Data analysis

Farm was the principal descriptive unit. Scores and animal-level percentages were tabulated by farm and interpreted against protocol thresholds. To summarize the observed cases without implying population inference, principle scores are also presented as median and range within each management context. No hypothesis test, confidence interval, or model-based comparison was performed because the sample contained only three purposively selected farms per context and because the implemented protocols differed. Expressions such as higher or lower refer only to numerical patterns in these six cases and do not imply statistical significance or a production-system effect.

2.6 Ethical considerations

Assessments consisted of non-invasive observation of animals during routine farm activities, resource inspection, and interviews; no animal handling, experimental treatment, or change to routine husbandry was imposed. Ethical review and approval were therefore not required under the applicable local and institutional requirements. Farm owners or managers provided informed consent for participation and publication of anonymized farm-level data.

3 Results

3.1 Farm context and overall classification

The six farms differed substantially in herd size, breed composition, and milk production (Table 1). Pasture-based herds had 51–170 lactating cows and produced 12.7-23.7 L/cow per day. Confined herds had 54–375 lactating cows and produced 21.0-36.0 L/cow per day. Four farms (P1, P3, C1, and C2) were classified as Enhanced, whereas P2 and C3 were Acceptable (Table 3). No farm was classified as Excellent or Not classified. Within pasture-based farms, median (range) principle scores were 65 (34-100) for Good Feeding, 63 (44-71) for Good Housing, 33 (30-35) for Good Health, and 64 (38-67) for Appropriate Behavior. Corresponding values for confined farms were 58 (36-83), 71 (65-71), 22 (22-25), and 23 (11-31), respectively. These summaries describe only the assessed cases. The Acceptable classification of P2 coincided with lower Good Housing and Appropriate Behavior scores, whereas C3 had the lowest Good Feeding and Appropriate Behavior scores among confined farms. Exact farm- and measure-specific denominators were not preserved in the archived records. Percentages are therefore reported as originally recorded and should not be back-calculated to infer exact animal counts.

Table 3. Welfare-principle scores, within-context descriptive summaries, and final farm classifications.

Scores range from 0 to 100, with higher values indicating more favorable welfare. Summary rows report median (range) across the three farms in each management context and are descriptive only. NA, not applicable.

3.2 Good feeding

Good Feeding scores were 34, 100, and 65 for P1-P3 and 58, 83, and 36 for C1-C3, respectively (Table 3). On P1, 8.0% of assessed cows were very thin and 3.0% were classified as fat; water was not available in a sufficiently efficient, abundant, and clean manner. P3 had clean, functional drinkers but only one water source per group. On confined farms, the proportion of very thin cows was 8.1% on C1, 5.9% on C2, and 3.1% on C3. C3 received 32 points for absence of prolonged thirst because drinkers were dirty. Thus, water cleanliness, redundancy, or accessibility contributed to lower feeding scores on farms in both management contexts.

3.3 Good housing and cleanliness

Good Housing scores were 71, 44, and 63 on P1-P3 and 65, 71, and 71 on C1-C3 (Table 3). All six farms had at least one serious cleanliness problem according to protocol thresholds (Table 4). On P1 and P3, more than 50% of cows had dirty lower hind legs and more than 19% had dirty udders and upper hind legs/flanks. P2 had moderate problems for lower-leg and udder cleanliness but a serious problem for upper hind legs/flanks. All three confined farms exceeded the serious-problem threshold for all three body regions.

Table 4. Percentage of assessed cows classified as dirty by body region.

Protocol thresholds: lower hind legs, normal ≤20%, moderate >20-50%, serious >50%; udder and upper hind legs/flank, normal ≤10%, moderate >10-19%, serious >19%. Exact measure-specific denominators were unavailable in the archived historical records; percentages should therefore not be used to reconstruct animal counts.

Within the pasture adaptation, shade provision yielded criterion scores of 100 on P1, 20 on P2, and 66 on P3. P2 lacked sufficient and evenly distributed natural or artificial shade, while P3 did not provide the number or area of shade structures specified by the adapted protocol. In confined farms, the 2009 protocol had no direct animal- or environment-based thermal-comfort measure; this criterion inherited the better score of the other two housing criteria. Lying-down time and collisions with equipment were within protocol limits on all confined farms. On C1, 3.0% of cows were lying outside the cubicles. The assessor also recorded that the number of cows exceeded the number of available stalls; this cow-to-stall ratio was contextual information and was not included as a scored Welfare Quality^® measure. No cows were tethered on any farm.

3.4 Good health

Good Health was the lowest and least variable principle within each management context: 30, 33, and 35 on P1-P3 and 22, 25, and 22 on C1-C3 (Table 3). Locomotion and integument outcomes are shown in Table 5. Moderate locomotion impairment affected 28.0%, 25.0%, and 14.0% of assessed cows on P1-P3, and severe impairment affected 8.0%, 3.0%, and 0.0%. Corresponding values on C1-C3 were 30.0%, 34.0%, and 41.0% for moderate impairment and 24.0%, 22.0%, and 18.0% for severe impairment. Severe integument alterations affected 50.0%, 57.0%, and 25.0% of assessed cows on P1-P3 and 65.0%, 41.0%, and 65.0% on C1-C3.

Table 5. Integument alterations and locomotion impairment among assessed cows.

Exact measure-specific denominators were unavailable in the archived historical records; percentages should therefore not be used to reconstruct animal counts.

Observed integument problems on pasture included lesions or alopecia compatible with prior tick infestation, photosensitization, dermatobiosis, injection-site abscesses, and papillomas. In confined herds, abrasions and swellings were observed around the hock, knee, flank, and neck; the field notes identified bedding condition and neck-rail or headlock position as plausible local contributors. These were descriptive observations rather than formally tested risk factors.

Disease-related threshold exceedances differed by farm. Bulk-tank somatic cell counts exceeded the protocol alarm threshold of 600,000 cells/mL on all pasture-based farms. P1 and P3 also exceeded alarm thresholds for dystocia and mortality, while P2 exceeded the diarrhea threshold and P3 reached the warning range for diarrhea. On confined farms, 8.0% of cows on C1 had ocular discharge; on C2, 17.0% had nasal discharge, 9.4% had diarrhea, and 6.3% showed labored respiration; C3 had elevated bulk-tank somatic cell count and 6.5% dystocia. Hot-iron disbudding was practiced without anesthesia or analgesia in all six herds; tail docking was not reported.

3.5 Appropriate behavior

Appropriate Behavior scores were 64, 38, and 67 on P1-P3 and 23, 31, and 11 on C1-C3 (Table 3). These farm-level values are presented descriptively and should not be interpreted as a between-system comparison because the implemented behavioral procedures differed. For expression of social behavior, P2 scored 32 following 0.52 head butts and 1.32 displacements per cow-hour near concentrated feed, whereas P1 and P3 scored 97 and 98. Confined farms scored 80, 97, and 76 for social behavior, with head butts ranging from 0.08 to 0.59 and displacements from 0.02 to 0.22 per cow-hour. The different observation durations and locations preclude direct comparison of these rates between protocol versions.

All pasture-based farms received the maximum score for expression of other behaviors because cows had year-round pasture access except during milking or treatment. All confined farms received zero for this criterion because pasture was not provided. Avoidance-distance criterion scores were high on C1-C3 (89, 92, and 86); 20.6%, 48.4%, and 23.7% of assessed cows, respectively, allowed touch. For positive emotional state, C1 and C2 scored 30 and 51, whereas C3 scored 1 and was described as showing more fearful, frustrated, bored, restless, apathetic, and stressed expressions. P3 had the most favorable qualitative behavioral profile among the pasture cases.

4 Discussion

This descriptive multi-case study provides historical, farm-level welfare profiles from six heterogeneous dairy farms assessed during an early period of application of the Welfare Quality^® framework in Brazil. Its contribution is empirical rather than methodological. The study applied existing procedures to document measure-level constraints across three pasture-based and three continuously housed farms without claiming protocol development or validation. Four farms met the historical protocol requirements for an Enhanced classification and two were classified as Acceptable. Nevertheless, every farm presented important welfare constraints. Good Health was consistently the lowest-scoring principle, serious cleanliness problems were widespread, and individual farms also showed limitations related to water provision, shade, behavior, locomotion, and pain management. The central practical finding is therefore not that one production system was uniformly better than the other, but that favorable global classifications coexisted with domain-specific problems requiring targeted management actions.

This coexistence illustrates why aggregated categories should not replace measure-level interpretation. The Welfare Quality^® model uses threshold-based classification rules that are not fully compensatory at key decision points, but farms may still receive an Enhanced classification while individual principles remain close to the lower threshold used for farm classification. Previous evaluations of the multi-criteria model have shown that some measures have greater influence on integrated scores than others and that discrimination between farms may be limited (de Vries et al., 2013). Expert work aimed at simplifying the protocol prioritized lameness, leanness, mortality, hairless patches, lesions or swellings, and somatic cell count as important adverse outcomes (Tuyttens et al., 2021). Ferreira et al. (2025) likewise found that simplified internal audits centered on body condition, water provision, lameness, integument alterations, and disbudding pain management can identify problems and encourage corrective action before certification, although simplified audits are not equivalent to complete protocol validation. These priorities closely matched the constraints observed in the present farms. For advisory purposes, a dashboard combining the four principles with sentinel animal-based measures may therefore be more useful than the final classification alone.

Locomotion impairment varied substantially across the six assessed farms. Moderate and severe impairment ranged from 14.0-28.0% and 0.0-8.0%, respectively, among P1-P3, and from 30.0-41.0% and 18.0-24.0% among C1-C3. These are descriptive case-specific observations and cannot be attributed to housing system because herd size, milk yield, breed, flooring, bedding, hoof-care practices, stocking density, and region were not controlled. Nevertheless, the observed pattern is biologically consistent with experimental and observational evidence indicating that pasture access may support locomotion recovery and reduce exposure to hard or wet walking surfaces, whereas well-designed and properly managed housing can also mitigate these risks (Haskell et al., 2006; Hernandez-Mendo et al., 2007; Olmos et al., 2009). Recent large-scale studies in partly housed, pasture-based Irish herds have further demonstrated that lameness varies considerably among farms and is associated with both cow-level and herd-level management factors (Crossley et al., 2021; Browne et al., 2022). In Brazilian small-scale grazing herds, cow and farm characteristics were also associated with lameness, reinforcing that pasture access is not intrinsically protective when walking tracks, hoof care, nutrition, or early detection are inadequate (Bran et al., 2018). Thus, locomotion should be interpreted as a management outcome rather than as an automatic consequence of either pasture-based or confined production.

Integument alterations should likewise not be regarded as a problem exclusive to confined housing. In a recent Brazilian study of 335 grazing dairy cows from 12 small-scale herds under tropical conditions, 81.5% of cows had at least one integumentary injury. Hairless patches on the hock, lesions on the flank, side, and udder, and swellings in the back, shoulder, and neck regions were prominent findings (Santos et al., 2026). This evidence reinforces that pasture access does not eliminate injury risk because cows remain exposed to handling facilities, fences, walking tracks, vegetation, parasites, and other environmental or management hazards.

In the pasture-based farms assessed in the present study, lesions compatible with photosensitization, dermatobiosis, injection-site reactions, papillomas, and previous tick infestation were recorded descriptively. In the continuously housed farms, abrasions and swellings occurred around the hock, knee, flank, and neck, and field notes identified lying-surface conditions and contact with neck rails or headlocks as possible local contributors. Because the underlying causes were not clinically confirmed or formally tested, these observations should guide further inspection rather than be interpreted as definitive diagnoses. Practical responses will differ according to the production context. Pasture-based farms require attention to walking tracks, vegetation, ectoparasite control, injection technique, and handling areas, whereas continuously housed farms require review of stall dimensions, bedding, flooring, and equipment contact points.

Cleanliness was a cross-cutting constraint, although the likely pathways differed between management contexts. Every farm exceeded the serious threshold for at least one body region, and all continuously housed farms exceeded that threshold for the lower hind legs, udder, and upper hind legs or flank. In housed systems, bedding quantity and renewal, stall maintenance, alley scraping, stocking density, manure removal, holding-area time, and pre-milking hygiene may contribute directly to soiling. In pasture-based systems, cleanliness may instead be influenced by rainfall, soil moisture, drainage, congregation around shade, gateways and water points, walking tracks, holding and milking areas, and local stocking density. Therefore, poor cleanliness in grazing systems should not automatically be attributed to inadequate bedding or poor management.

The extensive-cattle framework proposed by Kaurivi et al. (2020), which evaluates body cleanliness together with water access, paddock and yard hazards, and handling resources, illustrates the importance of interpreting cleanliness within the wider outdoor environment. Because the present visits occurred during the rainy summer, mud and moisture probably contributed to this historical snapshot. However, seasonality does not eliminate the management relevance of dirty udders and legs. Cow hygiene is associated with somatic cell count and mastitis risk, and the location of dirt may help identify priority areas for corrective action (Sant'Anna and Paranhos da Costa, 2011).

Good Feeding scores varied widely within both management contexts. The protocol identified visibly thin cows and water-related constraints, but it did not provide a complete nutritional diagnosis. In pasture-based farms, interpretation should consider pasture availability and quality, seasonal walking demand, the number and distribution of water points, mud or obstructions around troughs, and competition where feed, water, and shade are concentrated. Farm P2 illustrated why the distribution of shade, rather than total shade area alone, may be important. Concentration of shade around feed or water points can increase competition and restrict access for subordinate cows.

In continuously housed farms, relevant challenges include ration access, competition for feeding space, drinker cleanliness and flow, the availability of alternative water points, and obstruction of cow movement. Thin cows in either production context require individual investigation because lactation stage, disease, social competition, diet formulation, and walking demand may all contribute. Water points should be clean, functional, accessible, adequately sized, and distributed in a way that allows subordinate cows to obtain water without excessive competition.

Appropriate Behavior requires particularly cautious cross-context interpretation. The farm-level scores for this principle differed across the six cases, but the values should not be interpreted as a direct system comparison because all continuously housed farms received zero for the expression of other behaviors under the implemented scoring procedure when pasture access was not provided. The pasture adaptation also changed the duration and location of agonistic-behavior and avoidance-distance observations. Consequently, the resulting scores combine actual behavioral observations with structural properties of the assessment procedures.

Welfare outcomes may also vary by season and housing period within the same farms (de Graaf et al., 2017; Crossley et al., 2021), and grazing duration has shown heterogeneous relationships with Welfare Quality^® outcomes (Burow et al., 2013; Wagner et al., 2021). The low emotional-state score observed on farm C3 remains a meaningful within-farm signal, but it should prompt repeated assessment by trained observers rather than be interpreted as evidence of a causal effect of housing system.

Avoidance distance primarily reflects the human-animal relationship and must be interpreted in light of previous opportunities for human contact. Cows maintained predominantly on pasture may encounter people mainly during milking, treatment, vaccination, or movement, whereas housed cows may experience more frequent routine human presence. Breed composition, available escape space, handling history, test location, and the frequency and quality of previous interactions may therefore influence behavioral responses. A longer avoidance distance in a grazing herd does not, by itself, demonstrate poorer overall welfare, as it may reflect lower habituation or the specific testing context. Methodological work conducted under pasture conditions also indicates that avoidance-distance testing requires further refinement and may be more informative at the group level than at the individual level (Aubé et al., 2023). The values obtained under the two implemented procedures should therefore not be compared directly.

The behavioral results also reflect the broader evolution of welfare science. Traditional protocols are effective for detecting harms but may be less able to demonstrate whether animals experience positive affective states. Recent on-farm studies have evaluated indicators such as behavioral synchrony, relaxed postures, rumination while resting, voluntary human contact, and facial or ear expressions (Mattiello et al., 2024). The inclusion of validated positive-welfare indicators and repeated observations could help distinguish the absence of severe harm from opportunities for agency, comfort, exploration, and positive social experiences. Any extension of the protocol should, however, retain practical feasibility and include observer-reliability assessment, since a complete assessment required approximately 7 h per farm in the present study.

More recent pasture-oriented protocols have also incorporated measures that were absent from Garcia’s (2013) adaptation. Valente and Stilwell (2022) included distance to water points, thermal comfort, ear-tag positioning, responses during chute restraint, and entry and exit speed when assessing Portuguese autochthonous suckler herds. Kaurivi et al. (2020) incorporated water distance and availability, shade, environmental hazards, stockpersonship, and behavior during handling in extensive cow-calf systems. These measures may be informative for cattle managed outdoors for most of the year. However, they were not collected in the present historical dataset and could not be reconstructed retrospectively. Their absence represents a limitation of the Garcia adaptation and a priority for future feasibility, reliability, weighting, and validation studies in tropical pasture-based dairy herds.

Pain management was a consistent and preventable concern. All six farms reported hot-iron disbudding without anesthesia or analgesia. Disbudding is painful, and evidence indicates that local anesthesia reduces acute nociception, whereas its combination with an appropriate non-steroidal anti-inflammatory drug provides more sustained relief than local anesthesia alone (Stilwell et al., 2009; Herskin and Nielsen, 2018). Brazilian studies have identified limited awareness of pain-relief options and trade-offs between calf pain and perceived production or handling benefits (Cardoso et al., 2016). Extension professionals may facilitate or impede the adoption of pain-control measures depending on their knowledge, attitudes, and confidence, highlighting the importance of practical veterinary guidance and targeted education (Hötzel and Sneddon, 2013). This welfare concern was communicated to the farm owners or managers at the time of the assessments.

The present cases do not estimate current national prevalence, but they identify a clear welfare priority for veterinary oversight, multimodal pain-control protocols, trained restraint and drug administration, and auditable treatment records. Because the same practice was reported across all six farms, this finding also illustrates the value of including management-based indicators alongside animal-based outcomes.

An additional limitation is that the assessments and scores were anchored to the 2009 edition of the Welfare Quality^® dairy-cow protocol and to Garcia’s (2013) pasture adaptation. Version 3, released after data collection, changed the calculation of absence of prolonged thirst by using a continuous availability function weighted by water-point cleanliness and also revised the aggregation of injury and disease measures (Welfare Quality Network, 2023). The present scores were not recalculated under Version 3 because the historical pasture adaptation has not been revalidated against those algorithms and some required measure-level data were unavailable. Accordingly, these results should not be interpreted as current Version 3 classifications or compared directly with assessments generated using the revised protocol. The Version 3 revisions directly modify the scoring of absence of prolonged thirst and the aggregation of injury and disease measures, and therefore may affect the Good Feeding and Good Health principle scores.

This study has several additional limitations. The dataset is historical, originating from field assessments conducted in 2013–2014, and therefore cannot be interpreted as representing current welfare conditions or contemporary prevalence. The long interval between data collection and the present analysis increased reliance on archived records and limited the retrospective reconstruction of some sampling and facility-related details. Farms were purposively selected, only three cases represented each management context, and each farm was assessed during a single visit in the warm and rainy season. Exact farm- and measure-specific animal denominators were not preserved in the archived records, and no formal inter- or intra-observer reliability coefficients were documented. Some facility characteristics were also incompletely recorded. In addition, pasture-based and confined farms were assessed using partially different historical procedures, including differences in behavioral observation settings and durations and the exclusion of measures considered inapplicable to pasture systems. Scores were calculated using the historical 2009 Welfare Quality^® protocol and Garcia’s pasture-adapted procedure rather than Version 3, and the Acceptable classification of C3 was sensitive to small changes around protocol thresholds. Collectively, these limitations preclude population-level inference, contemporary prevalence estimates, formal between-system comparisons, and causal interpretation. Nevertheless, the study provides a detailed historical, measure-level account of welfare conditions in the six assessed farms and supports context-sensitive, farm-specific interpretation.

5 Conclusion

Across these six dairy farms, overall Welfare Quality^® classifications did not fully capture recurring constraints in health, cleanliness, and pain management. Four farms were classified as Enhanced and two as Acceptable, yet Good Health scores remained relatively low across all cases. The Acceptable classification of C3 should be interpreted cautiously because it was threshold-sensitive: with the other principle scores unchanged, a one-point reduction in Appropriate Behavior from 11 to 10 would have resulted in a Not classified outcome. Locomotion impairment and severe integument alterations were prominent, cleanliness thresholds were frequently exceeded, and farm-specific improvements were needed in water access, shade provision, disease monitoring, and behavioral conditions. Appropriate Behavior scores varied across the six farms, but these values cannot be interpreted as a direct production-system comparison because the implemented protocols differed, particularly in the scoring of pasture access and the context of behavioral tests. The findings support farm-specific action plans focused on routine animal-based monitoring, lameness prevention, hygienic resting and walking areas, clean and accessible water, multimodal pain control, and worker education. Because farms were purposively selected, assessed during a single visit, evaluated using partially different historical procedures, and incompletely characterized with respect to pasture-farm handling infrastructure, the results should be interpreted as case profiles rather than current prevalence estimates, official Version 3 classifications, or evidence of a causal effect of production system.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The author(s) declared that generative AI was used during the preparation of this manuscript. ChatGPT (OpenAI) was used solely for language editing and text refinement. No data analysis, scientific interpretation, or figure generation was performed using this tool. All authors reviewed and edited the final manuscript and take full responsibility for its content.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Pol Llonch, Autonomous University of Barcelona, Spain

Reviewed by: Luz Y. Peña-Avelino, Universidad Autónoma de Tamaulipas, Mexico

Ferreira, M.F., University of Trás-os-Montes e Alto Douro, Quinta de Prados, Portugal

Snežana Paskaš, University of Novi Sad, Serbia