Diurnal Behaviour, Health and Hygiene of Dairy Cows in Compost Barn Systems Under Different Climates in Argentina: A Bayesian Approach

Abstract

Compost barn systems are relevant alternatives to discussing production efficiency, welfare, and sustainability in dairy farming. However, studies evaluating these systems in different climates are still scarce, especially in subtropical climate zones. Here, we assess whether dairy cows’ behaviour, health and hygiene in compost barn systems are influenced by different climatic conditions and calving orders in Argentina’s central and extra-Pampean basins from the perspective of Bayesian inference. We evaluated dairy cows (n = 40) in a compost barn system simultaneously at two locations in Argentina: Rafaela and Salta. The following variables were evaluated: environmental factors, animal behaviour, respiratory rate, udder and hock hygiene, and locomotion degree of milking cows. There was a total of 10 primiparous cows and 10 multiparous cows at each location, randomly selected, which were in the first third of lactation (<90 DIM). Using Bayesian inference, we observed that Rafaela had a temperature-humidity index (THI) above 70, and Salta had a milder environment, with lower average temperature and higher relative humidity. Thus, climatic interference is evident in behaviour, triggering more behavioural and physiological mechanisms for heat abatement in primiparous females in Rafaela. At the same time, the mild conditions in Salta led to better thermal energy transfer by multiparous females compared to primiparous cows. This shows that the microclimate could interfere with the social hierarchy of cows when they are under heat stress. These findings highlight the importance of considering both calving orders and climate when designing management strategies for dairy systems.

1. Introduction

In recent years, dairy farming has undergone significant changes. Systems have evolved towards intensification, which poses challenges related to adapting processes to increase efficiency, improve animal welfare, and achieve more sustainable production [1]. In this context of intensification, confined housing systems for dairy cows have gained relevance to improve the quality of life of the animals and simplify the management associated with pastoral systems. Among confinement systems, the compost barn (CB), also referred to as a compost-bedded pack or “hot bed”,—has been widely adopted [2]. This system is based on barns where the feeding area is separated from the resting area. The cows remain in free circulation within this covered area and have access to a bedding area composed of organic material, usually sawdust. Daily, the bedding is stirred to incorporate the animals’ faeces and promote its oxygenation [3]. This way, the aerobic decomposition process is kept active to achieve a clean and dry surface [4]. There are many benefits to this system that have attracted the attention of the production and scientific sectors in terms of its validation for animal comfort.

Producers who choose this system are motivated by providing greater animal comfort, increasing their welfare standards, and improving longevity [5]. The greater adoption of these systems has occurred in regions where heat stress is constantly present in dairy systems, causing a drop in milk production and quality [6].

To cope with heat stress, dairy cattle rely on several physiological mechanisms such as increased respiration rate, panting and sweating, and consequent reductions in milk yield. Additionally, dairy cows modify their drinking and feed intake behaviour, increase standing time and shade seeking, and decrease activity and movement [7]. Also, behavioural responses and indices of welfare, reproduction, and herd health are negatively influenced [8].

Among the parameters considered to have the most significant impact and that condition the welfare and productivity of animals in CB confinement, the individual surface area assigned, the hygiene of the bedding and its temperature, as well as that of the barn interior, stand out [9]. To evaluate whether a CB system is well managed, in addition to the physical/chemical analyses of the bedding, Lobeck et al. [10] indicate that the study of behaviour and outcome-based measurements, such as locomotion and hygiene scores and milk production, are potential indicators to assess the welfare of dairy cows in different housing environments. Regarding heat stress, the main indicator in cattle is the ambient temperature-humidity index (THI), equal to or greater than 68 in dairy cows [11].

Argentina is located at the southern end of the American continent. It can be divided into large geographical regions that range from the subtropical climate in the north of the country to the cold and humid environment in the far south, passing through the semi-arid and temperate zones in the central area. It is worth mentioning that the central dairy basins of the country are in this latter region, where climatic conditions generate heat stress in dairy cattle for much of the year. For example, in Rafaela, a town in the country’s central region, an average of 21 days with a THI above 68 is recorded between December and March, totalling 158 days a year [12]. Additionally, there are other regions where milk production has regional economic importance, although its national impact is low, such as Salta (northwestern Argentina), where climatic conditions are more favourable for animal comfort. For example, the dairy basin of Salta, where the average annual temperature is 15.9 °C, has average maximum temperatures of 24.4 °C in the summer months and an average relative humidity of 45%. Although such thermal conditions are known in similar subtropical climate types worldwide, little has been explored about how animals in the same production system in different regions behave and react to different thermal stimuli. In Argentina, it is not known how the thermal comfort and behaviour of dairy cows in different calving orders are affected by the contrasting conditions between the Salta and Rafaela dairy basins. For this reason, a simultaneous comparative study is presented, aiming to verify the responses of cows kept in compost barn systems simultaneously to different thermal challenges. This article focuses primarily on behavioural and physiological responses, the thermal environment and its possible influence on productivity.

The objective of this study was to assess how contrasting climatic conditions in Argentina’s central and extra-Pampean dairy basins influence the behaviour and health of primiparous and multiparous cows in compost barn systems, with the dual aim of generating preliminary scientific evidence on the interaction between parity, climate, and welfare, and providing practical guidance for producers to adapt management strategies—such as heat mitigation in warmer, drier regions and bedding management in cooler, humid regions—to improve cow health and welfare under increasing climate variability.

2. Materials and Methods

2.1. Experimental Site

The study was conducted simultaneously at two dairy farms located in the central (Rafaela, Santa Fe Province, Latitude 31°11′ S and Longitude 61°30′ W, 100 m above sea level) and northwestern (Rosario de Lerma, Salta Province, Latitude 24°58′ S and Longitude 65°34′ W, 2846 m above sea level) regions of Argentina, during February and March 2023.

2.2. Structure and Management

The selected commercial establishments (Figure 1) were characterized by having a similar biotype (Holstein), individual milk production (≥30 L/cow/day; 2.5 milking/cow/day), access to a compost barn housing shed (11–12 m²/cow, 25–30 cm bedding depth), diets formulated with similar energy density (2.7 Mcal ME/kg DM) and protein content (17% CP), and an assigned feed offer ad libitum, 3.7% of live weight (

Table 1. Ingredients of diets in Salta and Rafaela dairy farm in February and March 2023.

	Rafaela		Salta
Ingredients (kg Dry Matter/Cow/Day)	February	March	February	March
Pelleted concentrate feed (*)	6.10	6.30	5.35	4.70
Corn silage	5.80	4.80	5.00	7.30
Alfalfa pasture (mechanical harvest)	4.50	4.00	0.00	0.00
Alfalfa hay	3.40	3.90	3.50	3.10
High-protein soybean meal	2.40	2.50	3.20	2.90
Corn grain	2.10	3.00	5.00	4.40
Cotton seed	1.90	1.90	2.50	2.30
Soybean hull	0.00	0.43	0.00	0.00
Wheat bran	0.00	0.00	2.30	2.10
Total	26.20	26.83	26.85	26.80

(*) 16% crude protein, 3 Mcal ME/kg DM, includes vitamins and minerals.

). The manure and sawdust bedding management consisted of daily tilling (twice/day) with a rototiller and chisel to achieve dry and fluffy bedding. Both establishments lacked a ventilation and/or cooling system inside the shed.

2.3. Animal Management

Between mid-February and March 2023, over a period of 12 days (3 consecutive days per week), the following were evaluated: animal behaviour, respiratory rate, udder and hock hygiene, and locomotion degree of milking cows (MC): 10 primiparous cows and 10 multiparous cows on every dairy farm, randomly selected, which were in the first third of lactation (<90 DIM;

Table 2. Characteristics of the cows selected in the study in dairy farm Rafaela and Salta (mean ± standard deviation).

	Rafaela		Salta
Items	Primiparous Cows	Multiparous Cows	Primiparous Cows	Multiparous Cows
Milk production (L/cow/day)	32.0 ± 8.4	39.3 ± 7.9	31.1 ± 0.6	35.1 ± 0.6
Age (months)	23.3 ± 1.5	57.4 ± 1.6	23.8 ± 1.4	46.8 ± 1.9
Body weight (kg)	530 ± 45	675 ± 59	505 ± 52	623 ± 75
Number of lactations	1.0	2.8 ± 0.6	1.0	2.9 ± 0.7

). Eligible cows, including both multiparous and primiparous animals with no recorded health or reproductive events, were randomly selected using Microsoft Excel to form the subset of dairy cows included in the trial. The selected cows were managed under the same conditions as the rest of the herd and housed in the compost barn. Additionally, climatic conditions (ambient temperature and humidity) and the temperature of the environment and bedding (surface and at 20 cm depth) under the MC housing shed were evaluated.

2.4. Environmental Assessments

The indoor environmental temperature was recorded daily, every hour, using two automatic sensors (I Button DS1921G- F5, iButtonLink Technology, Whitewater, WI, USA), each placed inside a 0.16 m diameter black copper Vernon sphere [13]. These were located 2 m above the ground in the housing shed at both dairy establishments.

The external ambient temperature, relative humidity and wind speed at each establishment were obtained from the records of an automated weather station (model DZ-WT1081) located nearby. The temperature-humidity index (THI) was calculated according to Thom’s equation ([11]; Equation (1)):

$$THI=\left(1.8\times Ta+32\right)-\left(0.55-0.55\times HR\right)\times \left(1.8\times Ta-26\right)$$

(1)

where

• $Ta$: dry-bulb temperature (°C)
• $HR$: relative humidity (%).

2.5. Bedding Temperature

The surface temperature and at 20 cm depth were recorded daily before the animal behaviour evaluation (09:00 and 14:00 h), using a stainless-steel compost thermometer (model 19.2008, TFA Dostmann, Wertheim, Germany; accuracy ± 0.1 °C) at nine randomly selected points in the MC housing shed.

2.6. Animal Behaviour

Daytime behaviour was recorded between 09:00 and 14:00 h by three trained observers at each establishment. Observations were made every 15 min using the focal sampling method with 0/1 sampling, following the methodology described by [14]. The inter-observer agreement among the three evaluators was 97% in Salta and 95% in Rafaela. A partial ethogram (

Table 3. Ethogram of behaviour and respective definitions for lactating dairy cows.

Behaviour	Definition
Standing rest	Standing, not engaging in any activity.
Lying rest	Lying down, not engaging in any activity.
Standing rumination	Standing, regurgitating or chewing, including ruminal movement.
Lying rumination	Lying down, regurgitating or chewing food.
Eating	Animals at the feeder, with their heads inside the feeding alley, visibly ingesting food or interacting with food or peers in the feeder area.
Drinking	Animals at the water trough, with their heads inclined towards the supplied water, visibly ingesting water or interacting with the water slide or peers in the drinking area.
Walking	Moving within the compost barn area, moving within the social space.

) adapted from Pilatti et al. [15] was used [16].

2.7. Respiratory Rate

The respiratory rate (RR) was evaluated daily (at 9 and 14 h) by direct observation of all MC, counting the abdominal muscle movements of the right flank for 30 s, and then expressed as respiratory movements per minute (RMM; [16]).

2.8. Hygiene and Locomotion

On each day of animal behaviour evaluation, before the second milking, a trained observer conducted a general assessment of the hygiene level of the MC, considering the cleanliness of the udder, upper and lower legs, and flank, following the scale proposed by Cook and Reinemann [17] (score 1 = Free of dirt; 2 = Slightly dirty; 3 = Moderately covered with dirt; 4 = Covered with caked on dirt). Simultaneously, the hoof condition score (score 1 = normal; 2 = mild and 3 = severe) was evaluated according to Martínez and Suarez [9].

2.9. Milk Production

In both dairy farms, the cows were milked with automatic milking systems of the DeLaval company (model V 300 in Rafaela and model V 301 in Salta). The individual daily milk production (MP) data of the cows under study were obtained through the Delpro system (Delpro FarmManager 5.9 DeLaval Argentina, Victoria, San Fernando, Buenos Aires company).

2.10. Statistical Analysis

This study employs Bayesian analysis to scrutinise the complex interplay of environmental factors, animal behaviour, and milk production in dairy cattle across two distinct regions in Argentina. Additionally, we conduct a comparative assessment of two pivotal social groups within the dairy herd—primiparous and multiparous cows. The aim is to discern nuanced patterns and variations while maintaining scientific rigour and objectivity. Our primary focus is to utilise Bayesian methodology as a robust statistical tool for unravelling intricate relationships between environmental variables, social dynamics, and key performance indicators in dairy cattle. By adopting a Bayesian framework, we aim to transcend the limitations of traditional statistical approaches, providing a more nuanced understanding of the multifaceted interactions within the dairy farming ecosystem.

According to McElreath [18], Bayesian statistics emphasises the importance of assigning probabilities to uncertain events or parameters. Unlike traditional frequentist statistics, where parameters are considered fixed and unknown, Bayesian statistics treats them as random variables with probability distributions. This paradigm shift allows us to incorporate prior knowledge or beliefs into our analyses, making Bayesian methods particularly well-suited for handling complex and uncertain problems.

In our study, inference was carried out through posterior predictive contrasts. After fitting the Bayesian models for each response (Poisson, ordinal, or Gaussian depending on the variable), we generated posterior predictive distributions for each group of interest (e.g., primiparous vs. multiparous cows, or locations Rafaela vs. Salta). For every posterior draw, we computed the group differences, yielding a posterior distribution of contrasts. These distributions provide the basis for all inferential statements reported in the Results. Thus, rather than relying on p-values or asymptotic tests, we present the full uncertainty distribution of the contrasts, which allows direct probability statements about group differences.

This approach follows the logic described in [18] and builds upon applied implementations using hierarchical Bayesian models with the brms package [19] (see [20]). Posterior predictive contrasts are increasingly recommended as an intuitive and transparent means of Bayesian inference, as they provide evidence directly interpretable as the probability that one group differs from another. For the implementation of this approach, we utilised the R (version 4.3.1; [21]) language along with several packages, each playing a distinct role in data manipulation, Bayesian modelling, and visualisation:

• rstan [22]: For interfacing with the STAN probabilistic programming language.
• brms [23]: For fitting Bayesian models using a high-level syntax.
• tidybayes [19]: For converting Bayesian model objects into tidy data frames.
• emmeans [24]: For obtaining estimated marginal means from models.

2.10.1. Descriptive Analysis

We utilised ggplot2 to create line plots that depict the temporal trends of each variable in both Rafaela and Salta. The geom_smooth function was employed to fit a smooth curve, allowing for a clearer representation of trends over time. Additionally, shaded intervals around the smooth curve provide a visual representation of the uncertainty associated with the fitted model.

Given our dataset containing both hourly and daily measurements of environmental variables, we have chosen a dual plotting approach. Specifically, we offer hourly averages, derived from aggregating daily data, to capture finer-scale variations within each day. This strategy ensures a comprehensive exploration of temporal patterns, balancing the granularity of hourly data with the interpretability of daily trends.

Differences and similarities between Rafaela and Salta are visually discernible, providing valuable insights into the regional variations in these variables.

2.10.2. Modelling Animal Behaviour

Typically, behavioural data is modelled as a dichotomous variable, indicating the presence or absence of a specific behaviour at a given moment in time. However, due to the short monitoring period in our data, we opted to use a Poisson model. This model accounts for the number of times a particular behaviour occurs within the observed time frame (Equation (2)):

$$\begin{array}{c}{\mathrm{y}}_{\mathrm{i}\mathrm{j}\mathrm{k}\mathrm{l}}\sim \mathrm{Poisson}\left({\mathsf{\lambda }}_{\mathrm{i}\mathrm{j}\mathrm{k}\mathrm{l}}\right)\\ \mathrm{l}\mathrm{o}\mathrm{g}\left({\mathsf{\lambda }}_{\mathrm{i}\mathrm{j}\mathrm{k}\mathrm{l}}\right)={\mathsf{\alpha }}_{\mathrm{i}\mathrm{j}\mathrm{l}}+{\mathrm{C}}_{\mathrm{j}}+{\mathrm{D}}_{\mathrm{k}}+{\mathrm{L}}_{\mathrm{l}}+{\left(\mathrm{C}\mathrm{D}\mathrm{L}\right)}_{\mathrm{j}\mathrm{k}\mathrm{l}}\\ {\mathsf{\alpha }}_{\mathrm{i}\mathrm{j}\mathrm{l}}\sim \mathrm{Normal}\left({\mathsf{\mu }}_{\mathsf{\alpha }},{\mathsf{\sigma }}_{\mathsf{\alpha }}^2\right)\\ {\mathsf{\mu }}_{\mathsf{\alpha }}\sim \mathrm{Normal}\left(\mathrm{1,0.3}\right)\\ {\mathsf{\sigma }}_{\mathsf{\alpha }}^2\sim \text{Half-Cauchy}\left(\mathrm{0,25}\right)\\ {\mathrm{C}}_{\mathrm{j}}\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\\ {\mathrm{D}}_{\mathrm{k}}\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\\ {\mathrm{L}}_{\mathrm{l}}\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\\ {\left(\mathrm{C}\mathrm{D}\mathrm{L}\right)}_{\mathrm{j}\mathrm{k}\mathrm{l}}\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\end{array}$$

(2)

where

• $y_{ijkl}$ represents the Poisson-distributed outcome for observation $i$ within the combination of levels of four factors $j$, $k$ and $l$;
• ${\lambda }_{ijkl}$ represents the Poisson rate parameter for observation $i$ within the combination of levels of three factors;
• The log of the Poisson rate parameter ($\log \left({\lambda }_{ijkl}\right)$) is modelled as the sum of the random intercept (${\alpha }_{ijl}$) and the fixed effects $\left(C_j,D_k,L_{\mathrm{l}}\right)$, as well as the three-way interaction term $\left({\left(CDL\right)}_{jkl}\right)$;
• ${\alpha }_{ijl}$ represents a random effect for log counts within the individual $i$, which is nested within the category $j$, which is further nested within locality $l$;
• ${\mu }_{\alpha }$ represents the mean of the random effects for the intercepts;
• ${\sigma }_{\alpha }^2$ represents the variance of the random effects for the intercepts;
• $C_j,D_k,L_l$ are fixed effects for factors $j$, $k$, and $l$, where $C$ is the social category (primiparous and multiparous), $D$ is the day of the essay, and $L$ is the location (Salta and Rafaela);
• ${\left(CDL\right)}_{jkl}$ represents a three-way interaction term.

Our priors serve as regularisation priors. In this particular instance, we anticipate that the parameter estimates will conform to a Gaussian distribution. The variance of random effects is anticipated to adhere to a weakly informative half-Cauchy distribution, as recommended by Gelman et al. [25].

2.10.3. Modelling Respiratory/Heat Rate

Since this variable involves counting, the model aligns with the one established for behaviour analysis. The only modification is that ${\mu }_{\alpha }\sim \mathrm{Normal}\left(\mathrm{3.5,0.5}\right)$ instead of ${\mu }_{\alpha }\sim \mathrm{Normal}\left(\mathrm{1,0.3}\right)$.

2.10.4. Modelling Hygiene and Lameness Score

The following variables (hygiene and hoof conditions scores) present an exception in our analysis. Since they represent scores and are qualitative (ordinal) variables, it is not feasible to calculate the contrast of the estimates between groups. Instead, we can generate plots depicting the posterior predictive distribution of each score for each group in comparison to the others. This approach allows for a similar comparison to what was done previously, albeit in a slightly indirect manner.

For Hygiene and Lameness scores, our estimate consists of ordinal categories. Consequently, our model has been adjusted from the previous ones to align with this, defining a likelihood function that is now binomial logit and linear predictors that incorporate threshold intercepts (Equation (3)). The priors remain very similar to the previous models, with adaptations made solely to the values of the regularisation parameters:

$$\begin{array}{c}{y}_{ijkl}\sim \mathrm{Ordinal}\left({\theta }_{ijkl}\right)\\ \mathrm{logit}\left({\theta }_{ijkl}\right)={\alpha }_{ijl}+C_j+D_k+L_l+{\left(CDL\right)}_{jkl}\\ {\alpha }_{ijl}\sim \mathrm{Normal}\left({\mu }_{\alpha },{\sigma }_{\alpha }^2\right)\\ {\mathsf{\mu }}_{\mathsf{\alpha }}\sim \mathrm{Normal}\left(\mathrm{0,1.5}\right)\\ {\mathsf{\sigma }}_{\mathsf{\alpha }}^2\sim \text{Half-Cauchy}\left(\mathrm{0,25}\right)\\ C_j\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\\ D_k\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\\ L_l\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\\ {\left(CDL\right)}_{jkl}\sim \mathrm{Normal}\left(\mathrm{0,0.5}\right)\end{array}$$

(3)

where

• $y_{ijkl}$ represents the ordinal-distributed outcome for observation $i$ within the combination of levels of four factors $j$, $k$ and $l$;
• ${\theta }_{ijkl}$ represents the cumulative log-odds of the ordinal response categories for observation $i$ within the combination of $j$ categories, $k$ localities, and $l$ days;
• The logit link function $\left(\mathrm{logit}\left({\theta }_{ijkl}\right)\right)$ is modelled as the sum of the random intercept $\left({\alpha }_{ijl}\right)$ and the coefficients for the main effects ($C_j,D_k,L_l)$ and the triple interaction $\left({\left(CDL\right)}_{jkl}\right)$ between the three factors;
• ${\alpha }_{ijl}$ represent a random effect for the cumulative log-odds within the individual $i$, which is nested within the category $j$, which is further nested within locality $l$;
• ${\mu }_{\alpha }$ represents the mean of the random effects for the cumulative log-odds;
• ${\sigma }_{\alpha }^2$ represents the variance of the random effects for the cumulative log-odds;
• $C_j,D_k,L_l$ are fixed effects for factors $j$, $k$, and $l$, where $C$ is the social category (primiparous and multiparous), $D$ is the day of the essay, and $L$ is the location (Salta and Rafaela);
• ${\left(CDL\right)}_{jkl}$ represents a three-way interaction term.

2.10.5. Modelling the Temperature of the Bed

The continuous variables (bedding temperature—Equation (4) and milk production—Equation (5)), or estimates, were ultimately modelled assuming a Gaussian distribution:

$$\begin{array}{c}{y}_{ijkl}\sim \mathrm{Normal}\left({\mu }_{ijkl},{\sigma }^2\right)\\ {\mu }_{ijkl}=\mathsf{\alpha }+M_j+D_k+L_l+{\left(MDL\right)}_{jkl}\\ \mathsf{\alpha }\sim \mathrm{Normal}\left(\mathrm{25,10}\right)\\ M_j\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ D_k\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ L_l\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ {\left(MDL\right)}_{jkl}\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ {\sigma }^2\sim \text{Half-Cauchy}\left(\mathrm{0,25}\right)\end{array}$$

(4)

where

• $y_{ijkl}$ represents the Gaussian-distributed outcome for observation $i$ within the combination of levels of three factors $j$, $k$ and $l$;
• ${\mu }_{ijkl}$ represents the parameter mean of Gaussian distributions and is modelled as the sum of the intercept ($\mathsf{\alpha })$ and the predictors ($M_j,D_k,L_l$), as well as the three-way interaction term (${\left(MDL\right)}_{jkl}$);
• $M_j,D_k,L_l$ are the effects for factors $j$, $k$, and $l$, where $M$ is parts of the day (Morning and afternoon), $D$ is the day of the essay and $L$ is the location (Salta and Rafaela);
• ${\left(MDL\right)}_{jkl}$ represents a three-way interaction term;
• ${\sigma }^2$ represents the variance of the outcome variable.

2.10.6. Modelling Milk Production

$$\begin{array}{c}{y}_{ijkl}\sim \mathrm{Normal}\left({\mu }_{ijkl},{\sigma }^2\right)\\ {\mu }_{ijkl}={\alpha }_{ijk}+C_j+D_k+L_l+{\left(CDL\right)}_{jkl}\\ {\alpha }_{ijk}\sim \mathrm{Normal}\left({\mu }_{\alpha },{\sigma }_{\alpha }^2\right)\\ {\mu }_{\alpha }\sim \mathrm{Normal}\left(\mathrm{35,10}\right)\\ {\sigma }_{\alpha }^2\sim \mathrm{Half-Cauchy}\left(\mathrm{0,25}\right)\\ C_j\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ D_k\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ L_l\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ {\left(CDL\right)}_{jkl}\sim \mathrm{Normal}\left(\mathrm{0,5}\right)\\ {\sigma }^2\sim \mathrm{Half}-\mathrm{Cauchy}\left(\mathrm{0,25}\right)\end{array}$$

(5)

where

• $y_{ijkl}$ represents the gaussian-distributed outcome for observation $i$ within the combination of levels of three factors $j$, $k$ and $l$;
• ${\mu }_{ijkl}$ represents the parameter mean of Gaussian distributions and is modelled as the sum of the random intercept (${\alpha }_{ijl}$) and the fixed effects ($C_j,D_k,L_l$), as well as the three-way interaction term (${\left(CDL\right)}_{jkl}$);
• ${\alpha }_{ijl}$ represents a random effect for the intercept within the individual $i$, which is nested within the category $j$, which is further nested within locality $l$;
• ${\mu }_{\alpha }$ represents the mean of the random effects for the intercepts;
• ${\sigma }_{\alpha }^2$ represents the variance of the random effects for the intercepts;
• $C_j,D_k,L_l$ are fixed effects for factors $j$, $k$, and $l$, where $C$ is the social category (primiparous and multiparous), $D$ is the day of the essay and $L$ is the location (Salta and Rafaela);
• ${\left(CDL\right)}_{jkl}$ represents a three-way interaction term;
• ${\sigma }^2$ represents the variance of the outcome variable.

3. Results

3.1. Environmental Description

Throughout the experimental days, Rafaela’s THI was higher than Salta’s. This resulted in an underlying increase in internal temperatures inside the barns, with a highlighted difference between the two locations (

Table 4. Mean ± standard deviation of the thermal variables regarding the compost barn during the experimental period, between 09:00 and 14:00 h, in Rafaela and Salta, Argentina.

	Rafaela				Salta
Day	ST *	T20	IT	THI	ST	T20	IT	THI
1	26.7 ± 2.3	29.7 ± 2.4	24.2 ± 6.2	78 ± 2.1	20.8 ± 2.1	46.2 ± 7.8	23.7 ± 2.8	74 ± 5.5
2	26.4 ± 3.3	28.6 ± 2.3	28.1 ± 4.5	77 ± 4.0	23.5 ± 3.7	51.8 ± 6.2	25.1 ± 1.0	75 ± 3.2
3	23.2 ± 2.3	27.3 ± 2.4	21.3 ± 2.0	69 ± 3.0	19.3 ± 2.5	50.0 ± 4.4	13.0 ± 1.5	56 ± 1.7
4	20.8 ± 0.9	22.7 ± 2.3	19.8 ± 0.4	67 ± 0.5	20.8 ± 1.6	39.4 ± 5.5	21.3 ± 1.9	68 ± 4.7
5	24.8 ± 3.4	25.1 ± 3.3	22.5 ± 1.1	71 ± 1.5	20.1 ± 1.7	38.9 ± 4.9	24.0 ± 1.9	73 ± 3.8
6	24.6 ± 2.3	25.4 ± 2.4	27.6 ± 2.5	75 ± 3.2	22.0 ± 2.2	41.0 ± 5.2	21.2 ± 1.9	70 ± 3.2
7	28.3 ± 3.4	28.7 ± 2.8	32.3 ± 2.7	80 ± 3.8	20.7 ± 1.5	38.7 ± 3.9	22.4 ± 1.6	72 ± 4.2
8	27.8 ± 4.3	27.7 ± 3.7	30.1 ± 2.6	78 ± 2.8	23.2 ± 1.5	38.1 ± 6.3	23.9 ± 2.0	75 ± 4.2
9	26.3 ± 3.6	27.6 ± 2.6	29.1 ± 2.0	77 ± 2.6	23.5 ± 1.4	40.4 ± 6.4	23.9 ± 1.9	74 ± 3.9
10	28.7 ± 4.2	31.3 ± 4.7	32.7 ± 3.1	78 ± 3.2	20.5 ± 0.9	41.7 ± 5.6	23.3 ± 1.6	74 ± 3.0
11	30.4 ± 4.8	32.3 ± 4.8	33.6 ± 3.4	80 ± 3.6	22.9 ± 1.6	44.4 ± 4.5	22.4 ± 2.0	73 ± 4.4
12	33.1 ± 5.4	36.2 ± 5.6	34.4 ± 2.6	82 ± 3.0	22.4 ± 1.0	40.2 ± 4.4	23.1 ± 1.8	75 ± 3.5

* ST = surface temperature (°C); T20 = bedding temperature (°C) at 20 cm depth; IT = indoor temperature (°C); THI = external temperature-humidity index (THI).

).

However, the bedding temperature at 20 cm depth was higher in Salta than in Rafaela, even though the surface temperature of the beds was higher in this location. Based on the above, we stated that Rafaela is a hotter and drier region, while Salta stands out for having a milder but wetter temperature (Figure 2a,b). This difference can be explained by the fact that cows in Rafaela were managed under a cold compost bed system, whereas in Salta they were kept under a hot compost bed system. Internally, the wind speed recorded (Figure 2c) was higher in Rafaela’s compost barn, which may favour reducing the animals’ heat load during the hottest hours. In summary, at the external THI (Figure 2d) in Rafaela, we observe that during the hottest hours, the values were higher than 70, which indicates possible thermal stress for the animals inside the shed. On the other hand, Salta recorded lower values, keeping the THI variation below 70, which shows a milder thermal situation.

3.2. Bedding Temperature Results

The Bayesian models did not converge for bedding temperature during the data analysis. As a result, verifying the difference between calving orders and the two locations was impossible.

3.3. Behaviour and Respiratory Rate Results

In Rafaela, primiparous cows (Vaq) had a higher probability of standing behaviour during most of the days (Figure 3a). Apart from day 3 (probability of 0.4805), all the other days showed a probability of occurrence of standing rest above 0.5 for primiparous cows, with nine days having a probability above 0.7. The opposite was found in Salta, where multiparous cows (Vaca) were more likely to be idle while standing throughout the experimental period. The exception was days 4 (0.4760), 7 (0.21275), 8 (0.15650) and 12 (0.2305). Most days in Salta were characterised by a high probability of standing multiparous cows (above 0.7).

Multiparous cows are more prone to idleness while resting than primiparous cows in Rafaela (Figure 2b). With the exception of days 2 (0.49750), 3 (0.10725), and 12 (0.33750), multiparous cows consistently exhibited a high likelihood of lying down, with probabilities exceeding 0.70 on the remaining days. Similar results were found in Salta, but with a probability above 0.8 on most days, except for days 3 (0.56400) and 7 (0.49075).

Primiparous cows showed a high probability of performing this behaviour in both localities (Figure 4a). However, lying rumination had a higher probability of occurrence for multiparous cows in Rafaela and Salta (Figure 4b).

For standing rumination, the difference between the locations refers to the level of probability. In Salta, the probability of this behaviour occurring is greater than 0.8. The same happened in Rafaela on only seven days, with the probability of occurrence on the other days around 0.6–0.7. However, the behaviour of lying down was more expressive for multiparous cows in both locations (Figure 5b). In both Rafaela and Salta, the animals had seven days with a probability of occurrence above 0.8.

Multiparous cows raised in Rafaela were more likely to engage in eating behaviour than primiparous cows, while in Salta, the opposite was observed, with primiparous cows more likely to engage in this behaviour (Figure 5a). However, the variability recorded in both locations was high, especially in Salta, which can be seen in the flattening of the predictive posterior distribution.

The probability of water drinking behaviour (Figure 5b) was higher (above 0.6) for multiparous cows than for primiparous cows in both locations. However, in Rafaela, the variability was lower than in Salta, which indicates greater accuracy in this finding. On days 2, 5, 6, 10, and 11, the probability of occurrence was higher (above 0.5) for primiparous cows.

Regarding walking behaviour, primiparous cows had a higher probability (above 0.55) than multiparous cows in Rafaela (Figure 6a). The opposite was found in Salta, where the probability above 0.55 was verified for multiparous cows in relation to primiparous cows. Regarding respiratory rate (Figure 6b), multiparous cows had a higher probability than primiparous cows in Rafaela. Only on three days was the relationship reversed. In Salta, primiparous cows were more likely to pant than multiparous cows during most of the experimental period (11 days), although the variability was greater, as verified by the flattening of the posterior predictive distributions.

3.4. Hygiene and Locomotion Results

The following variables (hygiene and hoof conditions scores) present an exception in our analysis. Since they represent scores and are qualitative (ordinal) variables, it is not feasible to calculate the contrast of the estimates between groups. Instead, we can generate plots depicting the posterior predictive distribution of each score for each group in comparison to the others. This approach allows for a similar comparison to what was done previously, albeit in a slightly indirect manner.

Following this interpretation for the hygiene score (Figure 7), we found that in both locations, there were no differences between groups of dairy cow categories. In both Rafaela and Salta, there was a higher probability of score 1 than score 3, showing that cows kept in a compost barn system have a high cleanliness score.

A similar pattern was observed for the hoof condition score (Figure 8). There was no difference between parity order. However, there was a difference in probabilities between Rafaela and Salta, with the latter location showing a marked probability (above 0.9) of score 1 occurring in relation to levels 2 and 3, i.e., normal condition and practically zero for score 3.

3.5. Milk Production Results

Multiparous cows were more likely to produce milk than primiparous cows in both locations (Figure 9). In Salta, this was true on all experimental days, while in Rafaela, on days 8, 14, and 15, primiparous cows were more likely to produce milk than multiparous cows.

4. Discussion

This study aimed to assess how the behaviour, locomotor health and hygiene of dairy cows of different calving orders were affected by different climatic conditions in central Argentina, from the perspective of Bayesian inference. The results showed that a hot, dry climate characterises the Rafaela region, while Salta has milder thermal conditions. The impact of these thermal environments on the cattle’s behaviour is discussed in detail in this paper.

In Rafaela, primiparous cows were more likely to be idle and ruminate while standing, with a greater probability of exhibiting walking behaviour. In Salta, multiparous cows were more likely to engage in lying down and walking. In both locations, there was a higher probability of standing rumination behaviour and a higher respiratory rate for primiparous cows. As for the scores, the compost barn system shows a higher probability of clean cows with normal locomotion. However, in Salta, the probability of normal locomotion was higher than in Rafaela. Finally, milk production showed a higher probability for multiparous cows than primiparous cows, although the probability was higher in Salta. Therefore, climatic interference is evident in behaviour, with primiparous females triggering more behavioural and physiological mechanisms for heat abatement in Rafaela. At the same time, the mild conditions in Salta led to better thermal energy transfer by multiparous females compared to primiparous cows.

4.1. Animal Behaviour and Respiratory Rate Discussion

Standing behaviour is an indicator of postural discomfort in farm animals [26,27,28]. The explanation for the high frequency of this behaviour is also correlated with the animal’s need to increase its surface area when under heat stress [29]. This makes it easier for the environment’s acclimatisation resources to be effective in reducing the internal heat load. In this study, in the town of Rafaela, which is characterised by a hot, dry climate (Table 2 and Figure 1), primiparous cows were more likely to stand and move around (walking behaviour).

According to Pilatti et al. [15], the hierarchy among cows is a key factor in determining the use of resources for heat relief. According to the same authors, the hotter it is, the more multiparous cows seek lying down behaviours, while primiparous cows are more likely to stand and walk. These statements also explain our results, i.e., the greater likelihood of walking behaviour, which shows an increase in the level of activity, which is a response to the need to search for heat abatement resources. Frigeri et al. [30] stated that in environments with THI above 70, cows spent reduces their lying down behaviour and increases walking and standing behaviours. In Rafaela, the THI was above 70 and our results followed these prior findings.

This stress in primiparous cows was also evident in the high probability of respiratory rate. Thermal exchange by panting is the main latent mechanism of heat and mass transfer in production animals. It is a reflex when animals have a surface temperature that is nearby the air temperature. In Rafaela, the average surface temperature was 26.8 °C, in addition to the mean air temperature (28.0 °C) and lower relative humidity (Figure 1b), thus favouring panting in the primiparous cows [6,31].

In Salta, where the average THI was below 70 (Figure 2d), the multiparous females found a favourable thermal environment for a greater likelihood of lying down. This finding is corroborated by the surface temperature of the bedding, which favours thermal exchange by conduction. As the animals remain in the sternal decubitus position during the day, the surface area transfers heat from the surface to the inside of the bedding. This significantly reduces the animal’s heat load in the hottest periods [32].

4.2. Hygiene and Locomotion Discussion

The compost barn system in Argentina provided conditions for the animals to have good hygiene and normal locomotion. When the bedding is well turned, it allows for good drainage and incorporation of manure for composting. The result is cows with high hygiene scores, as evidenced in this study (Figure 6). The predominant scores were 1 and 2, with a difference between 1 and 3 [15,33].

The low compaction of the bedding is the result of good turning and bedding material. This was observed by the hoof condition score. The cows had a normal locomotion score in both locations, but in Salta, the probability was even higher (Figure 7). Table 1 shows that the temperature at a depth of 20 cm was higher in Salta than in Rafaela. High temperature indicates greater composting activity. This better degrades the bedding components, leaving it drier and with a better structure for the animals to move around [34]. The result is a low incidence of locomotor problems, such as lameness, for example. In general, as observed by Danieli et al. [35], compost-bedded pack barn systems provide a soft surface into which the animal’s hooves can sink, allowing for better traction and reducing locomotor problems. In general, the prevalence of lameness in cows in the compost barn system varies between 4% and 9% [10,36]

4.3. Milk Production Discussion

This was an expected result. In general, multiparous cows produce more milk than primiparous cows. This explains the high probability found in this study. What is striking is the fact that Salta has a higher probability of production from multiparous cows than from primiparous cows (Figure 8). As described in Materials and Methods, the animals received the same feed in both locations. The climatic variation between the northwestern and central regions of Argentina may explain this finding. Salta has a milder climate in the hottest seasons and times of day, as the THI is found below 70. Multiparous cows in this location expend less metabolic energy to maintain body temperature, which results in higher productivity [37,38]. According to Guesine et al. [8], multiparous cows exposed to high temperature in a compost barn have their milk production decreased.

4.4. Limitations of the Study and Next Steps Forward

Our sample size is small in this study, which is one of its main limitations. We used 10 primiparous cows and 10 multiparous cows (a total of 20 cows per location), although the study was longitudinal and conducted over time. To overcome the difficulty of obtaining a larger number of animals, we used the Bayesian approach. One of the great advantages of this inference, in addition to those already listed in the Materials and Methods section, is the iterative method. That is, we performed n samples within a limited dataset, which ensures greater accuracy and precision when compared to frequentist models. [39].

Another limitation of this study is the lack of continuous monitoring by video cameras. Focal observation allows for closer interaction between the observer and the animal, but limits the study in terms of the frequency of these observations. With advances in computational analysis and the use of artificial intelligence, the variety of information improves the observation and verification of cows’ behavioural repertoire.

As for the next steps in the research, as evidenced by Frigeri et al. [30], the behavioural aspect needs further study. Assessing heat waves and their effects on production indicators, as well as the health and reproduction of cows of different calving orders, will certainly increase the level of scientific information aimed at elucidating the animals’ responses, not only physiological but also behavioural, in conjunction with other responses.

5. Conclusions

This study shows that climatic conditions in the central and extra-Pampean dairy basins of Argentina influence dairy cow behaviour and health differently according to parity. Primiparous cows were more affected by heat stress, whereas multiparous cows in cooler and more humid regions spent more time lying down. The compost barn system provided suitable hygiene and comfort, particularly in temperate climates. These findings highlight the importance of considering both parity and local climate when designing management strategies for Argentine dairy systems.

Our results indicate that contrasting climatic conditions between the central and extra-Pampean dairy basins influence cow behaviour and certain health indicators depending on parity. In warmer and drier regions, primiparous cows tended to stand and walk more frequently, while in cooler and more humid regions, multiparous cows spent more time lying down. Furthermore, the higher respiratory rate observed in primiparous cows suggests that microclimate may interact with social hierarchy, intensifying the effects of heat stress.