Ethical considerations

Ethical approval for this project was granted by an Animal Ethics Committee at the University of Melbourne, Australia (approval ID: 29590). The research was conducted in accordance with the Australian Code for the Responsible Care and Use of Animals for Scientific Purposes.

Study animals, housing and experimental design

A total of 200 Landrace × Large White × Duroc or Landrace × Large White sows and their litters were studied for 12 days post-farrowing, over six consecutive time replicates. There were five primiparous sows, 124 sows of parity 2–4, and 71 of parity 5–10. All sows were artificially inseminated apart from eight in one replicate that had been mated naturally. Approximately one week prior to expected parturition, sows were moved from an outdoor gestation unit to group farrowing and lactation paddocks, with each paddock containing eight sows and eight farrowing huts. Sows were free to move around the paddock and choose a farrowing hut to perform nest-building behaviours. Either four (Replicate 1, 2, 4, 5 and 6) or five (Replicate 3) paddocks were studied in each replicate (n = 32 or n = 40 sows and their litters per replicate). There was a farrowing spread of 6–9 days between litters within the same replicate, and 1–2 weeks between farrowings for consecutive time replicates. Each paddock contained one water source, one feeder, and a wallow.

Each hut was designed to house one sow and her litter. The huts measured 2.4 × 1.8 × 1.3 m (length × width × height) and contained one entryway (1.3 × 0.7 m; length × width), one small window (either 0.6 × 0.6 m or 0.6 × 0.3 m) that was covered throughout the experiment, and a small outdoor area contained by a low metal fence (1.2 × 1.1 m; Figure 1). For the first 2 weeks post-farrowing, the fence was 0.4 m in height, a level that allowed the sow to step over the fence to access the rest of the paddock, but limited piglets from leaving the hut and small outdoor area early in life (Figure 1). After two weeks, a portion of the front of the fence was removed, lowering the height and allowing piglets full access to the paddock and other litters. The huts were constructed from metal and did not contain any supplementary heat sources. While they were insulated, some were relatively old, and it was not possible to determine whether insulation was fully present or partially degraded. The present research was conducted within a larger project examining the effect of an intervention, a door flap covering the entry way of pigs’ shelter, in reducing heat loss and piglet mortality (Lucas et al. Reference Lucas, Wotherspoon and Sofra2025). For this larger research project, a randomised complete block design with paddock as the blocking factor was used to assign treatments (with or without door flap) to huts. Thus, within the present research, half of the huts in each paddock contained a transparent rubber door flap that covered three-quarters of the hut entryway. All huts were positioned with the entryway facing north to north-east.

Figure 1. Diagram of the outdoor farrowing and lactation hut pigs were housed in for the experiment, with the outdoor area pictured. Half of the huts in each paddock also contained a transparent door covering over the entryway (not pictured).

Pigs were managed by stockpeople according to typical commercial practice, which included at least one daily visual inspection of all animals. An abundant amount of straw (roughly 2–3 bales per hut) was provided inside the hut prior to farrowing and was replenished as needed, based upon the discretion of stockpeople. Feed was provided ad libitum to sows, and additional feed was dropped into the small outdoor area once per day for the first few days post-farrowing. Within 24 h of farrowing (day 0), stockpeople undertook routine postnatal litter management, which involved cross-fostering to equalise the weight of piglets within litters and the number of piglets across litters. Stockpeople assigned each litter to a weight category of ‘Small’, ‘Medium’ or ‘Big’, based on visual inspection of piglet sizes during fostering. Fostering was conducted within the two treatments (huts with or without door flap) being examined in the larger research project. During this process, some female piglets that were selected as breeding stock were ear-notched for identification and had half of their tail docked, while the remaining piglets did not undergo any husbandry procedures. In addition to routine management by stockpeople, the researchers recorded in-person piglet behaviour, straw temperature, straw moisture content, and ammonia concentrations for each hut once a day from 0 to 4 days of age, for the purposes of the larger research project examining the impact of door flaps as reported in Lucas et al. (Reference Lucas, Wotherspoon and Sofra2025). Piglets were weaned at a mean (± SD) age of 26 (± 2.7) days.

Measurements

Behavioural observations

Video observations were recorded, downloaded and then analysed retrospectively by one trained researcher for a subset of litters (n = 59; control huts = 26, door flap huts = 33). Prior to farrowing, Swift 3C scouting trail cameras (Outdoor Cameras Australia, Toowoomba, Australia) were placed in ten huts per replicate. One camera failed, and three were placed in door flaps instead of control huts, leading to a slightly greater representation of litters from door-flap huts. Each camera was set to record for 5 s every 10 min. Instantaneous scan sampling was used to record various piglet and sow behaviours, including piglet proximity to the sow when both were resting.

As one of the aims of the present research was to examine if piglets spend more time resting in close proximity to the sow in cold conditions, behavioural observations of sows and litters were recorded when both were simultaneously lying stationary, without active or attempted nursing by the piglets. The litter was considered to be resting when at least 80% of piglets were lying stationary anywhere within the hut or the outdoor area, typically with their eyes closed, and not in oral contact with the sow’s udder. When both the sow and piglets were resting, piglet proximity to the sow was recorded by counting the number of piglets in contact with the sow, within one piglet length of the sow, more than one piglet length from the sow, and out of view of the camera (Figure 2). Whether or not the majority of piglets (>50%) in the litter were shivering, and whether or not the litter was lying a single huddle (see Figure 2[a]) was also recorded. These behaviours were recorded every hour from 1200 to 2300h from day 0 to 4 (a total of 60 attempted observations per hut). The first recorded clip after 1200h was observed, and so on every hour until 23000h. This window of observation was chosen as it was after routine stockperson observation and intervention in the morning and allowed for observations over a larger range of temperatures and sunlight. When observations could not be taken (e.g. sows/piglets were active), the sampling interval was skipped.

Figure 2. Shows example sows and litters with piglets at various proximities. The litter in diagram (a) has all piglets resting in a huddle greater than one piglet length away, and this litter is eligible for observation as the sow and all piglets are resting. The litter in diagram (b) has one group of piglets awake at udder and in contact, one group resting and < 1 piglet length away and one group resting and > 1 length away. The piglets are observed in separate groups and thus are not huddling. This litter is ineligible for observation, as some piglets are attempting to nurse, and thus less than 80% of the litter are resting.

Hut temperature

Prior to farrowing, a temperature logger was placed inside each hut (either a Thermochron TC Temperature Logger DS1921G or Hygrochron Humidity/Temperature Logger DS1923, both Thermochron, Castle Hill, Australia). The temperature loggers were positioned on the top of the midpoint of the short wall opposite the hut entryway (Figure 3), and recorded hut temperature every hour. The loggers measured temperature with a precision of 0.5°C. Data were downloaded retrospectively by the researchers when the temperature loggers were removed at 12 days of age to allow re-use in other huts.

Figure 3. The inside of the outdoor farrowing and lactation hut viewed from the entryway, with a sow and litter inside. The red temperature logger can be seen at the midpoint of the back wall of the hut.

Statistical analysis

All data were analysed in RStudio, Version 2024.04.2+764 (R Core Team 2024) by fitting either a Linear Mixed Effects Model or a Generalised Linear Mixed Model using the packages lme4 (Bates et al. Reference Bates, Maechler, Bolker and Walker2015), lmerTest (Kuznetsova et al. Reference Kuznetsova, Brockhoff and Christensen2017) and glmmTMB (Brooks et al. Reference Brooks, Kristensen, Benthem, Magnusson, Berg, Nielsen, Skaug, Maechler and Bolker2017). The experimental unit was always the litter of pigs. For all Linear Mixed Effects Models, model fit was examined by checking for outliers, normality and homoscedasticity by visually inspecting the residual plots. For all Generalised Linear Mixed Models, a Pearson Chi-squared test was performed to test for overdispersion. Odds ratios for these models are reported as back-transformed values calculated from exponentiating the model estimates. Pairwise comparisons were conducted using the pairs function of the emmeans package (Lenth et al. Reference Lenth, Banfai, Bolker, Buerkne, Giné-Vázquez, Herve, Jung, Love, Miguez, Piaskowski, Riebl and Singmann2025).

Where relevant, models included the following fixed effects: day of piglet age (day 0, 1, 2, 3 or 4, with day 0 representing the first day of life); the litter’s weight category (small, medium or big, based on categorisation by the stockperson on day 0 as described previously); the time of day to the closest hour (1200, 1300 and so on until 2300h) and sow parity (parity 1, parity 2–4 or parity 5–10). To account for the hierarchical structure of the data, random effects for hut, paddock and replicate were included, with huts nested within paddocks which were nested within replicates, as well as day of age nested within the others for models examining piglet behaviour. Further detail pertaining to specific variables and each of the model structures is presented below.

Effect of hut temperature on piglet mortality

$$ {\displaystyle \begin{array}{l} Piglet\ mortality\ variable\sim Temperature\ variable+ Weight\\ {}\hskip1em +\hskip2px Sow\; parity+\left(1| Replicate/ Paddock/ Hut\right)\end{array}} $$

Generalised Linear Mixed Models using a binomial distribution with a logit-link function were fitted to examine the effect of hut temperature on piglet mortality. For analyses of piglet survival, the hut temperature variables included the average minimum and average hut temperature during three time-periods that corresponded to the piglet mortality variables: day 0, days 1 to 4, and days 5 to 12. Minimum and average temperature variables for these periods were calculated from averaging the daily minimum temperature over these days, and averaging the hourly hut temperature each day, respectively. Piglet mortality variables after litter management on day 0 included the number of piglet deaths due to any cause from days 1 to 4 and 5 to 12. For analyses of piglet survival before litter management on day 0, the piglet mortality variable was the total number of piglet deaths on day 0 for all sows who farrowed in a hut.

While specific causes of death were recorded by stockpeople according to routine practice on-farm, overwhelmingly most deaths had the recorded cause of overlay. For instance, on day 0, 100% of all piglet deaths inside a hut were recorded as overlay, with no deaths listed as being caused by starvation and only two who farrowed outside by hypothermia. It is likely that overlay was a secondary cause of death in some of these cases (i.e. the piglet died of starvation and was subsequently overlain). As such, it was decided to analyse piglet mortality by looking at all causes of death together as there was a degree of uncertainty regarding the accuracy of mortality causes.

Sow parity was included as a fixed effect in these models as previous studies have shown that sow parity can be a risk factor for increased mortality (Lossec et al. Reference Lossec, Herpin and Le Dividich1998; Galiot et al. Reference Galiot, Lachance, J-P and Guay2018). Days 1 to 4 were chosen specifically to look at the number of piglets that died early in life, as most piglet mortality occurs before this time (Svendsen et al. Reference Svendsen, Bengtsson and Svendsen1986; Marchant et al. Reference Marchant, Broom and Corning2001; Tucker et al. Reference Tucker, Craig, Morrison, Smits and Kirkwood2021), while days 5 to 12 were the remaining days in which piglets were still confined to the hut and hut temperature measurements were collected. In all piglet mortality models, the number of piglet deaths were compared to the number of piglets that had been alive in the litter at the start of the time-period; either the number born alive, or number alive on day 1 or day 5. In these cases, the number of piglets that had died on day 0 was compared to the number born alive. For piglet mortality on day 0, the litter’s weight classification was not included as a fixed effect as piglets had not yet been fostered and thus were of different weights.

Effect of temperature on piglet-sow proximity

$$ {\displaystyle \begin{array}{l}Piglets\ in\ contact\sim Temperature\ at\hskip0.2em the\ time\ of\ observations\\ {}\hskip0.5em +Day\hskip0.2em of\hskip0.2em age + Weight+Time\ of\hskip0.2em day\\ {}\hskip0.5em +(1|Replicate/Paddock/Hut/Day\hskip0.2em of\hskip0.2em age)\end{array}} $$

Generalised Linear Mixed Effects Models with a beta-binomial distribution were fitted to examine how hut temperature and day of age affected the proportion of the litter observed in physical contact with the sow from days 0 to 4 of age. Time of day was included in this model to control for systematic variation that may influence behaviour from day to night, such as influences of diurnal behaviour and presence of humans. An interaction between hut temperature at the time of observation and day of age was fitted, but there were no significant (P < 0.05) interactions, thus it was dropped, and odds ratios were reported from the additive model. While piglet behaviours were quantified based on the number of piglets resting in contact with the sow, within one piglet length of the sow, and more than one piglet length from the sow, the direction of effects for these different behaviour variables was similar. Thus, for conciseness, only statistical modelling for piglets resting in contact with the sow is reported. Temperature was quantified as the temperature of the hut at the closest hourly time-point to behavioural observations of piglet-sow proximity.

Effect of piglet-sow proximity on piglet mortality

$$ {\displaystyle \begin{array}{l} Piglet\ mortality\ variable\sim Piglets\ in\ contact+ Weight\\ {}\hskip1em +\hskip2px \left(1| Replicate/ Paddock/ Hut/ Day\; of\; age\right)\end{array}} $$

Generalised Linear Mixed Models using a binomial distribution with a logit-link function were fitted to examine the effect of piglet-sow proximity on piglet mortality. One piglet mortality variable was the total number of piglet deaths from days 1 to 4. For this model, average values for piglets in contact with the sow were calculated from the proportion of piglets in physical contact with the sow when both the sow and piglets were resting from 1200 to 2300h across days 1–4. Mortality and proximity over day 0 was also examined, where the piglet mortality variable was the number of piglet deaths on day 0 post-litter management, and the proximity variable was the resting behaviour of piglets from 1200 to 2300h.

Effect of temperature on shivering and huddling behaviours

$$ {\displaystyle \begin{array}{l}Piglet\ behaviour\sim Temperature\ at\hskip0.2em the\ time\ of\ observations\\ {}\hskip1em +Day\hskip0.2em of\hskip0.2em age + Weight + Time\ of\hskip0.2em day\\ {}\hskip1em +(1|Replicate/Paddock/Hut/Day\hskip0.2em of\hskip0.2em age)\end{array}} $$

Generalised Linear Mixed Models using a binomial distribution with a logit-link function were fitted to examine the effects of hut temperature on piglet thermoregulatory behaviours each hour. Piglet thermoregulatory behaviours included binary variables for whether most piglets in the litter were shivering (yes or no), and whether the litter was huddling as one group (yes or no) when resting. Temperature was quantified as the temperature of the hut at the closest hourly time-point to behavioural observations.

Number of huts and observations included in each model

A total of 200 sows and their litters were included in the study, but not every model contained a complete dataset from each of these 200 litters. Similarly, not all 59 huts that had cameras in them contained a full set of observations. A full dataset of 200 litters or 3,540 behaviour/temperature observations were not always present due to several reasons: four temperature loggers malfunctioned, weight category was not assigned to two litters, and piglet behavioural observations were unable to be completed if the sow or piglets were feeding or active. Further, when analysing piglet survival, there were 14 sows that farrowed outside of a hut or in another occupied hut, which were excluded from piglet mortality on day 0 and behavioural observations, as any piglet deaths that occurred at this time were not affected by hut temperature. The eleven sows that farrowed outside were moved into a farrowing hut as soon as possible, and as it was only a small number of huts, they were included in analyses of piglet mortality and temperature but excluded from any behaviour or mortality models that included observations on day 0.